The Scrum Papers:: Nuts, Bolts, and Origins of An Agile Process

The Scrum Papers:
Nuts, Bolts, and Origins of an Agile Process
Jeff Sutherland, Ph.D.

Ken Schwaber
Co-Creators of Scrum

CTO & Worldwide Scrum Consulting Practice Manager
PatientKeeper, Inc.
One Newton Place
275 Washington Street – 2nd Floor
Newton, MA 02458
+1 617 947-7400 U.S. mobile
+1 508 644-8298 Voicemail everywhere
+45 60 67 87 90 Danish mobile (617-987-0403 from U.S.)
+46 709 827833 Swedish mobile
+47 90 62 84 63 Norwegian mobile
+37 2532 03761 Explorer mobile (Italy, Dubai, Russia, Netherlands, UK, ...)
[email protected]
Draft 10/14/2007 ©Jeff Sutherland 1993-2007 1

Dedication ....................................................................................................................................... 9
Introduction................................................................................................................................... 11
Chapter 1: Introductions to Scrum................................................................................................ 13
A Brief Introduction to Scrum ...................................................................................................... 14
The Product Owner has the following responsibilities ......................................................... 14
The ScrumMaster is a facilitative team leader working closely with the Product Owner. He
must:...................................................................................................................................... 14
The Team is cross functional ................................................................................................ 15
Getting started with Scrum ................................................................................................... 16
Sprint Planning Meeting ....................................................................................................... 16
Daily Scrum Meeting............................................................................................................ 17
Burndown Chart.................................................................................................................... 17
Sprint Review Meeting ......................................................................................................... 18
Continuing the Scrum ........................................................................................................... 19
Scrum Primer ................................................................................................................................ 20
Traditional Software Development........................................................................................... 20
Agile Development and Scrum................................................................................................. 21
Scrum Basics............................................................................................................................. 22
Scrum Roles .............................................................................................................................. 23
Starting Scrum .......................................................................................................................... 24
Sprint Planning Meeting ........................................................................................................... 25
Daily Standup Meeting ............................................................................................................. 27
Sprint Review............................................................................................................................ 28
Sprint Retrospective.................................................................................................................. 29
Starting the Next Sprint ............................................................................................................ 29
Release Planning....................................................................................................................... 29
Common Challenges................................................................................................................. 30
Results From Scrum.................................................................................................................. 31
Rolling out Agile at a large Enterprise ......................................................................................... 33
Abstract ..................................................................................................................................... 33
Introduction............................................................................................................................... 33
Background ............................................................................................................................... 33
Kick-off..................................................................................................................................... 38
Management Support ................................................................................................................ 39

Employee Support..................................................................................................................... 39
Feedback ................................................................................................................................... 39
Roll-out ..................................................................................................................................... 40
Challenges................................................................................................................................. 40
Training..................................................................................................................................... 41
Next time................................................................................................................................... 42
Fund the internal coaching team adequately......................................................................... 42
Encourage deeper engagement from coaching staff ............................................................. 42
Deeper management involvement......................................................................................... 43
Focus on the majority............................................................................................................ 43
Conclusion ................................................................................................................................ 43
Results from Scrum................................................................................................................... 43
Capturing Extreme Business Value: 1000% Annual Return on Investment in Scrum Trainers .. 45
Introduction............................................................................................................................... 45
IDX Systems (now GE Healthcare): Scaling Scrum for the First Time ................................... 45
Chapter 2: The First Scrum........................................................................................................... 49
Agile Development: Lessons Learned from the First Scrum........................................................ 50
Introduction............................................................................................................................... 50
Reviewing Software Development Processes........................................................................... 50
Why the Easel CEO Supported the First Scrum ....................................................................... 51
Scrum Basics............................................................................................................................. 52
Scrum Results ........................................................................................................................... 54
The First Scrum: How Scrum provides energy, focus, clarity, and transparency to project teams
developing complex systems ........................................................................................................ 55
Chapter 3: What is Scrum? The First Papers on the Scrum Development Process...................... 57
Scrum Development Process ........................................................................................................ 58
Abstract ................................................................................................................................. 58
1. Introduction....................................................................................................................... 58
2. Overview........................................................................................................................... 59
3. Current Development Situation ........................................................................................ 59
4. Scrum Methodology.......................................................................................................... 64
5. Advantages of the Scrum Methodology ........................................................................... 72
6. Scrum Project Estimating ................................................................................................. 73
7. System Development Methodologies : Defined or Empirical .......................................... 73

Chapter 4: Scrum on Large Projects: Distributed, Outsourced Scrum ......................................... 77
First Scrum Scaling at IDX Systems 1996-2000 ...................................................................... 77
Linear Scalability in Large, Distributed, and Outsourced Scrums ........................................... 77
Agile Can Scale: Inventing and Reinventing Scrum in Five Companies ..................................... 79
Introduction............................................................................................................................... 79
Easel Corporation: The First Scrum ......................................................................................... 81
“All-at-Once” Software Development...................................................................................... 81
Software Evolution and Punctuated Equilibrium ..................................................................... 82
Stimulating Software Evolution with SyncSteps.................................................................. 83
Achieving a Sustainable HyperProductive State .................................................................. 85
VMARK: The First Senior Management Scrum ...................................................................... 86
Individual: The First Internet Scrum......................................................................................... 86
IDX Systems: The First Scrum in the Large............................................................................. 87
PatientKeeper Scrum: The First Scrum Company.................................................................... 87
Conclusions............................................................................................................................... 88
Distributed Scrum: Agile Project Management with Outsourced Development Teams .............. 89
Abstract ..................................................................................................................................... 89
Introduction............................................................................................................................... 89
“All-at-Once” Development Models ........................................................................................ 90
Hyperproductivity in Scrum ..................................................................................................... 91
Distributed, Outsourced Scrum..................................................................................................... 92
Distributed Team Models ..................................................................................................... 92
SirsiDynix Case Study .......................................................................................................... 94
Hidden Costs of Outsourcing................................................................................................ 94
Intent of the Integrated Scrums Model ......................................................................................... 95
Context...................................................................................................................................... 95
Forces........................................................................................................................................ 96
Complexity Drivers............................................................................................................... 96
Top Issues in Distributed Development................................................................................ 97
Solution: Integrated Scrums...................................................................................................... 97
Team Formation.................................................................................................................... 97
Scrum Meetings .................................................................................................................... 98
Sprints ................................................................................................................................... 99

Product Specifications .......................................................................................................... 99
Story for Simple Renewals Use Case: ................................................................................ 100
Testing................................................................................................................................. 101
Configuration Management ................................................................................................ 102
Pair Programming, Refactoring, and other XP practices.................................................... 102
Measuring Progress............................................................................................................. 103
Resulting Context with Integrated Scrums ............................................................................. 103
Conclusions............................................................................................................................. 104
Scrum and CMMI Level 5: The Magic Potion for Code Warriors............................................. 106
Abstract ................................................................................................................................... 106
Introduction............................................................................................................................. 106
CMMI ..................................................................................................................................... 106
Scrum ...................................................................................................................................... 107
CMMI and Agile methods ...................................................................................................... 108
Scrum and CMMI: a magic potion ......................................................................................... 108
How Systematic adopted Scrum ............................................................................................. 110
Systematic experience from pilots .......................................................................................... 113
Guide for mixing CMMI and Agile ........................................................................................ 115
How CMMI can improve Agile.......................................................................................... 116
Critiques of CMM................................................................................................................... 119
Conclusions............................................................................................................................. 120
Chapter 5: Scrum Metrics ........................................................................................................... 121
Reporting Scrum Project Progress to Executive Management through Metrics ........................ 122
Introduction............................................................................................................................. 122
Transparency into Projects...................................................................................................... 122
Executive Dashboard .............................................................................................................. 123
Work Breakdown Structure ................................................................................................ 125
Earned Business Value ....................................................................................................... 126
The Underlying Data........................................................................................................... 127
Putting the Data into the Report.......................................................................................... 127
Creating the Parking Lot..................................................................................................... 128
Earned Business Value ....................................................................................................... 128
Conclusions............................................................................................................................. 129

Notes ................................................................................................................................... 129
Chapter 6: Scrum Tuning............................................................................................................ 130
Type B Scrum Continuous Flow: Advancing the State of the Art ............................................. 132
Introduction............................................................................................................................. 132
Background on Scrum............................................................................................................. 132
Improving Scrum Implementations ........................................................................................ 134
Enhancing Scrum Implementations ........................................................................................ 135
Management Challenges for a Continuous Flow Scrum ........................................................ 137
Summary ................................................................................................................................. 138
Postscript................................................................................................................................. 138
Type C All-At-Once Scrum: Creating a Scrum Company ......................................................... 140
Abstract ................................................................................................................................... 140
1. Scrum Evolution ................................................................................................................. 140
2. Scrum Evolution in Practice ............................................................................................... 142
2. The First Scrums – Team Scrum and Continuous Flow Scrum.......................................... 143
3. Continuous Flow Scrum ..................................................................................................... 145
3.1 Staging Functional Specifications for a Type B Sprint ................................................ 146
3.2 Product Owner as Part of the Scrum Team................................................................... 148
3.3 Type B Scrum Can Enable a Hyperproductive Team................................................... 149
4. Type C Scrum ..................................................................................................................... 150
4.1 Case Study Context....................................................................................................... 151
4.2 Case Study Market Requirements................................................................................. 151
4.3 Case Study Problem...................................................................................................... 152
4.4 Case Study Forces......................................................................................................... 152
4.5 Type C Scrum Solution................................................................................................. 153
4.6 Type C Scrum Rationale............................................................................................... 162
4.7 Type C Scrum Resulting Context ................................................................................. 163
5. Conclusions......................................................................................................................... 166
Chapter 7: Case Studies .............................................................................................................. 168
Ssh! We are adding a process… (at Google) .............................................................................. 169
1. Introduction............................................................................................................................. 169
2. First agile attempts.................................................................................................................. 170
2.1. The guinea pig projects.................................................................................................... 171
2.2. The first process steps...................................................................................................... 172

2.3. Issues to overcome........................................................................................................... 173
2.4. Working with the remote team ........................................................................................ 174
3. Adding agility – one practice at a time ............................................................................... 174
3.1. Daily standup meetings................................................................................................ 174
3.2. Small steps – Completely finishing a feature/task....................................................... 175
3.3. Spikes........................................................................................................................... 176
4. Release experience.............................................................................................................. 177
5. Feedback and next steps...................................................................................................... 177
6. The second version ............................................................................................................. 178
6.1. The world is better, but … ........................................................................................... 179
6.2. The project manager is dispensable ............................................................................. 180
7. Where are we going from here............................................................................................ 180
8. Summary ............................................................................................................................. 181
5. References............................................................................................................................... 183
Appendix I: ................................................................................................................................. 188
Scrum: A Pattern Language for Hyperproductive Software Development ................................ 188
How does Scrum Work? ......................................................................................................... 189
Issues or blocks that need to be resolved. (The ScrumMaster is a team leader role
responsible for resolving the blocks.) ................................................................................. 190
The Patterns ............................................................................................................................ 190
Sprint....................................................................................................................................... 190
Context................................................................................................................................ 190
Problem ............................................................................................................................... 191
Forces.................................................................................................................................. 191
Solution ............................................................................................................................... 191
Rationale ............................................................................................................................. 192
Known Uses ........................................................................................................................ 192
Resulting Context................................................................................................................ 192
Backlog ................................................................................................................................... 193
Context (From: Sprint)........................................................................................................ 193
Problem ............................................................................................................................... 193
Forces.................................................................................................................................. 193
Solution ............................................................................................................................... 193

Resulting Context................................................................................................................ 194
Scrum Meetings ...................................................................................................................... 194
Context (From: Backlog) .................................................................................................... 194
Problem ............................................................................................................................... 195
Forces.................................................................................................................................. 195
Solution ............................................................................................................................... 196
Rationale ............................................................................................................................. 198
Index ........................................................................................................................................... 201

Dedication
This book is dedicated to the Godfathers of Scrum, Takeuchi and Nonaka [1], who unknowingly
gave Scrum its name and caused a worldwide transformation of software development. In
addition to Takeuchi and Nonaka, others contributing to the creation of Scrum have been:
• Jim Coplien and the ATT Bell Labs Pasteur Project for the paper on the most productive
software development team ever documented – the Borland Quattro Pro Project [2]. The
first Scrum team implemented the Scrum daily meeting after reading this paper.
• Nobel Laureates Muhammad Yunus and the Grameen Bank for originating
microenterprise development and the Accion International President’s Advisory Board,
responsible for much of microenterprise development in the western hemisphere. The
strategy for bootstrapping the poor out of poverty has been a model for freeing tens of
thousands of software developers from developer abuse caused by poor management
practices.
• Alan Kay and his team at Xerox Parc [3] for inventing Smalltalk, the mouse, the
graphical user interface, the personal computer, the Ethernet, and the laser printer.
Listening to his insights on innovation inspired the first Scrum team to go from “good” to
“great” [4].
• The Creative Initiative Foundation for their work with Silicon Valley volunteers to help
make the world a better place, the underlying motivation driving the founders of Scrum.
• Professor Rodney Brooks for launching the startup now known as iRobot in space leased
from Jeff Sutherland. He taught us the subsumption architecture [5], how to create simple
rules to produce highly intelligent performance from a complex adaptive system.
• Christopher Langton of Los Alamos Labs and the Sante Fe Institute for coining the term
“artificial life” and showing that increasing the degrees of freedom of computer life
forms up to the edge of chaotic behavior accelerated their evolution [6]. Scrum feels a
little “chaotic” by intent, so as to accelerate software evolution.
• Capers Jones and his productivity experts at Software Productivity Research who
analyzed and reanalyzed the output of early Scrum teams, as well as many of the software
products built with Scrum during 1994-2000 [7]. These analyses allowed us to provide a
money back guarantee that users would double productivity during the first month using
tools created by the first Scrum.
• The first Scrum team – John Scumniotales (ScrumMaster), Don Roedner (Product
Owner), Jeff McKenna (Senior Consultant), Joe Kinsella, Laurel Ginder, and others.
They endured repeated failure, depressing analysis of these failures in front of their
technical peers from other companies, and transcendence of their missteps. They were the
first Scrum team to achieve the hyperproductive state for which Scrum was designed and
their product, Object Studio, was reported as the industry leader by computer trade
journals.
• PatientKeeper, Inc., the first company to fully implement an “All at Once” or Type C
Scrum involving the entire company in Scrum practice. This innovation in process design
has been documented by Mary and Tom Poppendieck in their book on Lean Software
Development [8]. “I find that the vast majority of organizations are still trying to do too
much stuff, and thus find themselves thrashing. The only organization I know of which

has really solved this is Patient Keeper [9].” PatientKeeper was the first company to
achieve a hyperproductive revenue state driven by Scrum in July, 2007
• Last, but not least, some Scrum practitioners experience the quality without a name
(QWAN) - a phrase used by Christopher Alexander in his book “The Timeless Way of
Building” [10]. Alexander describes a certain quality that we seek, but which cannot be
named. This may be the most important feature of Scrum and can only be spoken of as a
set of core values - openness, focus, commitment, courage, and respect. It could be
viewed as the “speed of trust” or one of the sources of “ba” often seen on Scrum teams.
Ba is the Japanese term for the creative flow of innovation described by Takeuchi and
Nonaka [11].
Thanks to the reviewers of the text who include among many others:
• Tom Poppendieck
• Henrick Kniberg
• Rowan Bunning
• Clifford Thompson

Introduction
Agile software development processes are influenced by best practices in Japanese industry,
particularly by lean development principles [12] implemented at companies like Toyota [13] and
Honda [14], and knowledge management strategies developed by Takeuchi and Nonaka [11],
now at the Hitotsubashi Business School in Japan, and Peter Senge [15] at MIT.
Scrum is an Agile methodology that delivers software to customer and end users faster, better,
and cooler [16, 17]. As the Chief Product Owner at Yahoo observed, coolness is a requirement at
Google, Yahoo, and most software game companies. It must be highly suited to a creative
approach to development of complex and innovative systems and it must scale. It is used on
some of the world's largest projects at British Telecom and Siemens because of its high
productivity with distributed and outsourced development teams. It is the only software
development process that has repeatedly demonstrated linearly scalable productivity when
adding resources to large projects [18, 19]. The most profitable software product ever created
(Google Adwords [20]) is powered by Scrum and the most productive large project with over a
million lines of code (SirsiDynix and Starsoft [18]) used a distributed, outsourced, Scrum
implementation. CMMI Level 5 companies cut costs in half with Scrum while simultaneously
improving quality, customer satisfaction, and the developer experience (Systematic Software
Engineering [19]). At the same time, Scrum remains the process of choice in small
entrepreneurial companies where it has it roots.
The first software development Scrum was created at Easel Corporation [21] in 1993 based on
extensive research on successful projects worldwide, a deep analysis of the computer science
literature, close collaboration with leading productivity experts, and decades of experience with
advanced software technologies. Jeff Sutherland was the Chief Engineer for the Object Studio
team that defined roles, hired the first Product Owner and ScrumMaster, developed the first
Product Backlog and Sprint Backlog and built the first portfolio of products created with Scrum.
In 1995, Jeff introduced the Scrum team to Ken Schwaber, CEO of Advanced Development
Methods. Ken agreed that Scrum was a better way to build software and they worked together to
formalize the Scrum development process at OOPSLA’95 [22]. In the same year, Sutherland
provided support for development of eXtreme Programming [23] by giving Kent Beck all
background information on the creation of Scrum [24] and the results of two years of product
development with the Scrum process from 1993-95. XP engineering practices evolved along
with Scrum and the two leading Agile development processes work well together. Scrum and XP
are the most widely used Agile processes worldwide and their creators are signatories of the
Agile Manifesto.
Agile development is now accepted globally as the best way to develop, maintain, and support
software systems. Several papers on the early implementation of Scrum are of general interest.
Later papers provide some of the nuts, bolts, and best practices of Scrum implementations. The
design and implementation of an All-at-Once Scrum (Type C Scrum) at PatientKeeper to enable
enterprise agility has been emulated by innovative companies worldwide. Case studies of CMMI
Level 5 Scrum implementations and hyperproductive distributed, outsourced teams are of
particular interest. In this compilation, papers are organized into a single volume to be readily
accessible. More may be available in a subsequent edition.

Scrum is the only Agile process that has made its way through the Pattern Languages of
Programming Design (PLoP) process. Both Scrum and eXtreme Programming were affected by
the software patterns movement and Mike Beedle, the third Scrum signatory of the Agile
Manifesto, led the effort to formally codify Scrum as an organizational pattern. His work is
published in Volume 4 of Pattern Languages of Program Design [25] and the Scrum Pattern is
included here as Appendix I.
Scrum is designed to add energy, focus, clarity, and transparency to project planning and
implementation. It will:
• Increase speed of development

• Align individual and corporate objectives
• Create a culture driven by performance
• Support shareholder value creation
• Achieve stable and consistent communication of performance at all levels
• Enhance individual development and quality of life
The continued expansion of the global rollout of Scrum in both the largest and smallest software
companies is a testimony to the fact that Scrum delivers on its promise. While it is often said that
Scrum is not a silver bullet, Scrum can be like a heat seeking missile when pointed in the right
direction. It’s inspect and adapt approach to continuous quality improvement can do serious
damage to outmoded business practices and many companies are now transforming entire
organizations to take advantage of Scrum productivity, to delight customers, and to make the
work environment better and more fun for development teams. It’s focus on building
communities of stakeholders, encouraging a better life for developers, and delivering extreme
business value to customers can release creativity and team spirit in practitioners and make the
world a better place to live and work.
For every member of the first Scrum team, life was changed by Scrum. Some were concerned
that they would be searching for the rest of their life to find another development team that
would provide a similar exhilarating experience. They feared they would never be able to find
one. Because of their blood, sweat, and tears, and the careful shepherding of Scrum by Ken
Schwaber and the Scrum Alliance, they can relax. Scrum teams are found everywhere on the
planet, from Silicon Valley to Katmandu.
Jeff Sutherland - Boston, MA USA, 2007

Chapter 1: Introductions to Scrum
The Scrum Alliance requested a Scrum Overview for publication on their web site in 2006 and
the result was a brief overview of Scrum for general use. While intended for the uninitiated, it
has become more generally useful as a short summary of Scrum that can help all Scrum
practitioners refine their overview of Scrum.
Jeff Sutherland was asked by Tobias Meher to help educate engineers at Yahoo on Scrum in
November, 2005. Yahoo Vice President Pete Deemer set up a Yahoo Senior Management
meeting the following month for a Scrum briefing at an evening dinner in Palo Alto. By the end
of the dinner, Yahoo management decided to roll out Scrum companywide. They felt it fit the
Yahoo style of development they used as a startup while giving them a structure that would
support a global organization.
Pete recruited Gabrielle Benefield from in 2006 to lead the Scrum rollout at Yahoo in the U.S.
and then moved to become Yahoo Chief Product Officer in India. Together, they have written a
more thorough introduction to Scrum and provided some interesting survey data. Their paper is
included as a second introduction to Scrum and includes survey data from Scrum teams at
Yahoo.
In 2007, Gabrielle published updated survey data from over 100 Scrum teams at Yahoo. Included
as a third paper in this section, these data show overwhelming support for Scrum as a “better,
faster, cooler” method for building software. It also shows an annual Return on Investment (ROI)
of 1000% for investment in Scrum trainers. This was exactly the estimated ROI on Scrum
training proposed to Pete Deemer in 2005 by Jeff Sutherland.

A Brief Introduction to Scrum
Scrum Alliance, 2006
Scrum is a "lean" approach to software development. The term Scrum comes from a 1986 study
[1] by Takeuchi and Nonaka that was published in the Harvard Business Review. In that study,
Takeuchi and Nonaka note that projects using small, cross-functional teams historically produce
the best results. They write that these high-performing teams were like the Scrum formation in
Rugby. When Jeff Sutherland developed the Scrum process at Easel Corporation in 1993, he
used their study as the basis for team formation and adopted their analogy as the name of the
process as a whole. Ken Schwaber formalized the process for the worldwide software industry in
the first published paper on Scrum at OOPSLA 1995 [22] .
Scrum is a simple framework used to organize teams and get work done more productively with
higher quality. It allows teams to choose the amount of work to be done and decide how best to
do it, thereby providing a more enjoyable and productive working environment. Scrum focuses
on prioritizing work based on business value, improving the usefulness of what is delivered, and
increasing revenue, particularly early revenue. Designed to adapt to changing requirements
during the development process at short, regular intervals, Scrum allows teams to prioritize
customer requirements and adapt the work product in real time to customer needs. By doing this,
Scrum provides what the customer wants at the time of delivery (improving customer
satisfaction) while eliminating waste (work that is not highly valued by the customer).
Scrum is a simple “inspect and adapt” framework that has three roles, three ceremonies, and
three artifacts [26] designed to deliver working software in Sprints, usually 30-day iterations.
• Roles: Product Owner, ScrumMaster, Team.

• Ceremonies: Sprint Planning, Sprint Review, and Daily Scrum Meeting
• Artifacts: Product Backlog, Sprint Backlog, and Burndown Chart
The Product Owner has the following responsibilities
• Define the features of the product;

• Decide on release date and content;
• Be responsible for the profitability of the product (ROI);
• Prioritize features according to market value;
• Adjust features and priority every 30 days, as needed; and
• Accept or reject work results.
The Product Owner is responsible for the first of the three Scrum ceremonies: Scrum Planning.
The ScrumMaster is a facilitative team leader working closely with the Product
Owner. He must:

• Ensures that the team is fully functional and productive;
• Enable close cooperation across all roles and functions;
• Remove barriers;
• Shield the team from external interferences; and
• Ensure that the process is followed, including issuing invitations to Daily Scrum, Sprint
Review and Sprint Planning meetings.
The Team is cross functional
• Seven plus/minus two members

• Selects the Sprint goal and specifies work results
• Has the right to do everything within the boundaries of the project guidelines to reach the
Sprint goal
• Organizes itself and its work
• Demos work results to the Product Owner
Sprint Cycle

Getting started with Scrum
Preparation for a Scrum Sprint begins with the Product Owner developing a plan for a product or
a project. The Product Owner could be a customer representative or a customer proxy. For
product companies, the customer is a market, and the Product Owner serves as a proxy for the
market. A Product Owner needs a vision for the product that frames its ultimate purpose, a
business plan that shows what revenue streams can be anticipated from the product in what time
frames, and a road map that plans out several releases with features ordered by contribution to
return on investment (ROI). S/he prepares a list of customer requirements prioritized by business
value. This list is the Product Backlog, a single list of features prioritized by business value
delivered to the customer.
In order to prioritize the Product Backlog, the Product Owner needs help from the Scrum Team
to estimate the cost of developing features. The Product Backlog needs to include all features
visible to the customer, as well as technical requirements needed to build the product. The
highest priority items in the Product Backlog need to be broken down small enough to be
estimable and testable. About 10 developer days of work is a good size for a Product Backlog
item that is ready for implementation by a Scrum team in the next Sprint. Features that will be
implemented further out in time can be less detailed.
Sprint Planning Meeting
The Scrum begins when enough of the Product Backlog is defined and prioritized to launch the
first Sprint. A Sprint Planning Meeting is used to develop a detailed plan for the iteration. It
begins with the Product Owner reviewing the vision, the roadmap, the release plan, and the
Product Backlog with the Scrum Team. The Team reviews the estimates for features on the
Product Backlog and confirms that they are as accurate as possible. The Team decides how much
work it can successfully take into the Sprint based on team size, available hours, and level of
team productivity. It is important that the team “pull” items from the top of the Product Backlog
so they can commit to deliver in a Sprint of no longer than 30 days. Pull systems deliver
significant productivity gains and are a key component of lean product development.
When the Scrum Team has selected and committed to deliver a set of top priority features from
the Product Backlog, the ScrumMaster leads the team in a planning session to break down
Product Backlog features into Sprint tasks. These are the specific development activities required
to implement a feature and form the Sprint Backlog. This phase of the Sprint Planning Meeting
is time-boxed to a maximum of four hours. Tasks are broken down into pieces that will require
less than 2 days or 16 hours of work. When the Sprint Backlog is complete, the total work
estimated is compared with original estimates from the Product Backlog. If there is a significant
difference, the team negotiates with the Product Owner to get the right amount of work to take
into the Sprint in order to ensure a high probability of delivering “potentially shippable” code at
the end of the Sprint.

Daily Scrum Meeting
The Sprint begins its cycle and every day, the ScrumMaster leads the team in the Daily Scrum
Meeting. This is a 15 minute meeting designed to clarify the state of the Scrum. Each team
member speaks to three questions – what did I do yesterday, what will I do today, and what
impediments are in my way? While anyone can attend this meeting, only team members who
have committed to deliver work to the Scrum are allowed to speak. The goal is to get a global
snapshot of the project, discover any new dependencies, address any personal needs of
committed individuals, and adjust the work plan in real time to the needs of the day.
The ScrumMaster has three primary responsibilities in addition to leading the Daily Scrum
meeting:
1. The ScrumMaster needs to know what tasks have been completed, what tasks have
started, any new tasks that have been discovered, and any estimates that may have
changed. This makes it possible to update the Burndown Chart which shows the
cumulative work remaining day by day. The ScrumMaster must also look carefully at the
number of open tasks in progress. Work in progress needs to be minimized to achieve
lean productivity gains.
2. The ScrumMaster needs to surface dependencies and blocks which are impediments to
the Scrum. They need to be prioritized and tracked. A remediation plan needs to be
implemented for impediments in priority order. Some can be resolved with the team,
some can be resolved across teams, and others will need management involvement as
they may be company issues that block all teams from achieving their production
capacity. For example, a telecom company recently implemented Scrum and found 18
items on their impediment list, only three of which were directly related to Scrum teams.
The other impediments were company issues that needed management attention.
3. Last but not least, the ScrumMaster may notice personal problems or conflicts within the
Scrum that need resolution. These need to be clarified by the ScrumMaster and be
resolved by dialogue within the team, or the ScrumMaster may need help from
management or the Human Resources group. Jim Coplien developed over 200 case
studies of notable projects while working at ATT Bell Labs. He reports that over 50% of
productivity losses were caused by personnel issues. The ScrumMaster must pay
attention to them to ensure the team is fully functional and productive.
Burndown Chart
At the Sprint Planning Meeting the Scrum Team identifies and estimates specific tasks that must
be completed for the Sprint to be successful. The total of all Sprint Backlog estimates of work
remaining to be completed is the cumulative backlog. When tasks are completed as the Sprint
proceeds, the ScrumMaster recalculates the remaining work to be done and the Sprint Backlog
decreases, or burns down over time. If the cumulative Sprint Backlog is zero at the end of the
Sprint, the Sprint is successful.
The Product Backlog items brought into the Sprint are fixed for the duration of the Sprint.
However, the Sprint Backlog may change for several reasons:

• The development team gains a better understanding of work to be done as time
progresses and may find that they need to add new tasks to the Sprint Backlog to
complete the Product Backlog items selected.
• Defects may be identified and logged as additional tasks. While these are viewed
primarily as unfinished work on committed tasks, it may be necessary to keep track of
them separately.
• The Product Owner may work with the team during the Sprint helping to refine the
team’s understanding of the Sprint goal. The ScrumMaster and Team may decide that
minor adjustments that do not lengthen the Sprint are appropriate to optimize customer
value.
The Burndown Chart is used as a tool to guide the development team to successful completion of
a Sprint on time with working code that is potentially shippable as a product.
Sprint Burndown Chart
2500
2000
Hours of Work Remaining
1500
1000
500
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Tim e
Sprint Review Meeting
At the end of a Sprint, the potentially shippable code is demonstrated to the Product Owner in the
first half of a Sprint Review Meeting that is timeboxed to a maximum of four hours. The Product
Owner leads this part of the meeting and all interested stakeholders are invited to attend. The
state of the business, the market, and the technology are reviewed. The Product Owner
determines which items on the Product Backlog have been completed in the Sprint, and discusses
with the Scrum team and stakeholders how best to reprioritize the Product Backlog for the next
Sprint. The goal for the next Sprint is defined.

The ScrumMaster then leads the second half of the Sprint Review meeting with the Scrum team
for a Sprint Retrospective. The way the team worked together in the Sprint is assessed. Positive
ways of working together are identified and encouraged as future practice. Things that could
work better are identified and strategies for improvement are defined and agreed upon.
Continuing the Scrum
The next step after completing the Scrum Review Meeting is to start the process over again.
Iterations proceed until sufficient features are done to complete a project or release a product. A
well functioning Scrum will deliver highest business value features first and avoid building
features that will never be used by the customer. Since industry data shows over half of the
software features developed are never used, development can be completed in half the time by
avoiding waste, or unnecessary work.
In most companies, development is slowed down by impediments identified during the daily
meetings or planning and review meetings. When these impediments are prioritized and
systematically removed, further increases in productivity and quality are the result. Well run
Scrums achieve the Toyota effect – four times industry average productivity and 12 times better
quality.
This is accomplished by removing management pressure from the team, allowing the team to
select its own work, and then self-organize through close communication and mutual agreement
within the team on how best to accomplish the work. In a successful Scrum, this autonomy can
significantly improve the quality of life for developers and enhance employee retention for
managers.
The simple rules of Scrum allow for continual inspection, adaptation, self-organization, and
emergence of innovation. This can produce an exciting product for the customer, develop high
team spirit and satisfying work, generate high productivity and customer satisfaction making
management successful, and achieve the market and financial goals of the company. As a result,
Scrum is being widely adopted worldwide in companies large and small, localized or distributed,
open source or proprietary, for virtually any type or size of project.
Certified Scrum Trainer Craig Larman notes that “Scrum is arguably the oldest and most widely
applied agile and iterative method, with an emphasis on iterative and adaptive PM practices. It
has been applied in thousands of organizations and domains since the early 1990s, on projects
large and small, from Yahoo to Medtronics to Primavera, with great results when leadership
commits to the deep required changes moving away from command-control and wishful-
thinking-predictive management, and with poor results when leadership can’t or won’t make
those changes. Scrum can be easily integrated with practices from other iterative methods, such
as practices from the Unified Process and Extreme Programming, including test-driven
development, agile modeling, use cases, user stories, and so forth. On the surface Scrum appears
to be simple, but its emphasis on continuing inspect-adapt improvement cycles and self-
organizing systems has subtle implications.”

Scrum Primer
Pete Deemer and Gabrielle Benefield
Yahoo, 2006
Pete Deemer is Chief Product Officer, Yahoo! India Research and Development. Gabrielle
Benefield is Senior Director of Agile Development at Yahoo! Inc. They lead Yahoo!’s large-scale
global adoption of Scrum.
A note to readers: There are many concise descriptions of Scrum available online, and this
primer aims to provide the next level of detail on the practices. It is not intended as the final step
in a Scrum education; teams that are considering adopting Scrum are advised to equip
themselves with Ken Schwaber’s Agile Project Management with Scrum or Agile Software
Development with Scrum, and take advantage of the many excellent Scrum training and
coaching options that are available; full details are at Scrumalliance.org. Our thanks go to Ken
Schwaber, Dr. Jeff Sutherland, and Mike Cohn for their generous input.
Traditional Software Development

The traditional way to build software, used by companies big and small, is commonly known as
“The Waterfall”. There are many variants, but it typically begins with a detailed planning phase,
where the end product is carefully thought through, designed, and documented in great detail.
The tasks necessary to execute the design are determined, and the work is planned using tools
like Gantt charts and programs like Microsoft Project. The team arrives at an estimate of how
long the project will take by adding up detailed estimates of the individual steps involved. Once
stakeholders have thoroughly reviewed the plan and provided their approvals, the team starts to
build. Team members complete their specialized portion of the work, and then hand it off to
others in production-line fashion. Once the work is complete, it is delivered to a Quality
Assurance organization, which completes testing prior to the product reaching the customer.
Throughout the process, strict controls are placed on deviations from the plan, to ensure that
what is produced is actually what was designed.
This approach has strengths and weaknesses. Its great strength is that it is supremely logical:
think before you build, write it all down, follow a plan, and keep everything as organized as
possible. It has just one great weakness: humans are involved, and humans don’t work very well
this way.
For example: this approach requires that the good ideas all come at the beginning of the
development cycle, where they can be incorporated into the plan. But as we all know, good ideas
appear spontaneously throughout the process – in the beginning, the middle, and sometimes even
the day before launch, and a process that doesn’t permit change will stifle this innovation. With
the Waterfall approach, a great idea late in the development cycle is not a gift, it’s a threat.
The Waterfall approach also places a great emphasis on writing things down as a primary
method for communicating critical information. The very reasonable assumption is that if I can
write down on paper as much as possible of what’s in my head, it will more reliably make it into

the head of everyone else on the team; plus, if it’s on paper, there is tangible proof that I’ve done
my job. The reality, though, is that most of the time, these highly detailed 100-page requirements
documents just don’t get read. And that’s probably just as well, because when they do get read,
the misunderstandings are often compounded. A written document is an incomplete abstraction
of a picture I have in my head; when you read that document, you create yet another abstraction,
which is now two steps away from what I’m really thinking of. It should come as no surprise that
serious misunderstandings occur.
Something else that happens when you have humans involved is the “hands-on aha” moment –
the first time that you actually use the working product, and you immediately think of 20 ways
you could have made it better. Unfortunately, these very valuable insights often come at the end
of the development cycle, when changes are most difficult and disruptive – in other words, when
doing the right thing is most expensive.
Humans also have a poor ability to predict the future. For example, the competition makes an
announcement that wasn’t expected. Unanticipated technical problems crop up that force a
change in direction. Furthermore, people tend to be particularly bad at planning things far into
the future – guessing today how you’ll be spending your week eight months from now is
something of a fallacy, and it’s been the downfall of many a Gantt chart.
In addition, the Waterfall also tends to foster an adversarial relationship between the team-
members that are handing work off from one to the next. “He’s asking me to build something
that’s not in the spec.” “She’s changing her mind about what she wants.” “I can’t be held
responsible for something I don’t control.” And this gets us to another observation about the
Waterfall – it’s not that much fun to work within. In fact, we’d go a step further and say that the
Waterfall is a cause of great misery for the people who build products, and the resulting products
fall well short of expressing the creativity, skill, and passion of their creators. People aren’t
robots, and a process that expects them to act like robots often results in unhappy people.
A rigid, change-resistant process will also tend to produce mediocre products. Customers may
get what they first ask for, but is it what they really want once they see the product begin to
emerge? By gathering all the requirements up front and having them set in stone with little
chance of change, the product is condemned to be only as good as the initial idea, instead of
being the best it could be once the team knows more about the possibilities.
Many users of the Waterfall experience these shortcomings again and again, but it seems like such a
logical approach, the natural reaction is to turn the blame inward: “If only we did it better, it would work”
– if we just planned more, documented more, resisted change more, everything would work smoothly.
Unfortunately, many teams find just the opposite: the harder they try, the worse it gets!
“I believe that the prevailing system of management is, at its core, dedicated to mediocrity, If forces
people to work harder and harder to compensate for failing to tap the spirit and collective intelligence
that characterizes working together at its best.” Peter Senge, MIT
Agile Development and Scrum

The Agile family of development methodologies was born out of a belief that an approach more
grounded in human reality would yield better results. Agile emphasizes building working
software that people can get hands on with quickly, versus spending a lot of time writing
specifications up front. Agile focuses on small, cross-functional teams empowered to make
decisions, versus big hierarchies and compartmentalization by function, and Agile focuses on
rapid iteration, with as much customer input along the way as possible. Often when folks learn
about Agile, there’s a glimmer of recognition – it sounds a lot like back in the start-up days,
when we “just did it”.
One of the fastest-growing Agile methods is Scrum. It was formalized over a decade ago by Ken
Schwaber and Dr. Jeff Sutherland, and it’s now being used by companies large and small,
including Yahoo!, Microsoft, Google, Lockheed Martin, Motorola, SAP, Cisco, GE Medical,
CapitalOne and the US Federal Reserve. Many teams using Scrum report significant
improvements, and in some cases complete transformations, in both productivity and morale. For
product developers – many of whom have been burned by the “management fad of the month
club” – this is significant. Scrum is simple, powerful, and rooted in common sense.
Figure 1. Scrum
Scrum Basics
Scrum is an iterative, incremental process. Scrum structures product development in cycles of
work called Sprints, iterations of work which are typically 1-4 weeks in length. The Sprints are
of fixed duration – they end on a specific date whether the work has been completed or not, and
are never extended. At the beginning of each Sprint, a cross-functional team selects items from a
prioritized list of requirements, and commits to complete them by the end of the Sprint. Each
work day, the team gathers briefly to report to each other on progress, and update simple visual
representations of work remaining. At the end of the Sprint, the team demonstrates what they
have built, and gets feedback which can then be acted upon in the next Sprint. Scrum emphasizes
producing working product that at the end of the Sprint is really “done”; in the case of software,
this means code that is fully tested and potentially shippable.

Scrum Roles
In Scrum, there are three primary roles: The Product Owner, Team Members, and The
ScrumMaster.
The Product Owner is responsible for taking all the inputs into what the product should be –
from the customer or end-user of the product, as well as from Team Members and stakeholders –
and translating them into a product vision. In some cases, the Product Owner and the customer
are one and the same; in other cases, the customer might actually be millions of different people
with a variety of needs. The Product Owner role maps to the Product Manager or Product
Marketing Manager position in many organizations.
Team Members build the product that the customer is going to consume: the software, the
website, or whatever it may be. The team in Scrum is typically five to ten people, although teams
as large as 15 and as small as 3 commonly report benefits. The team should include all the
expertise necessary to deliver the finished work – so, for example, the team for a software project
might include programmers, interface designers, testers, marketers, and researchers. Team
members build the product, but they also provide input and ideas to the Product Owner about
how to make the product as good as it can be. Projects with more than 15 people are organized as
multiple Scrum teams, each focused on a different aspect of the product development, with close
coordination of their efforts. While team members can split their time with other projects, it’s
much more productive to have team members fully dedicated to the Scrum. Team members can
also change from one Sprint to the next, but that also reduces the productivity of the team.
The ScrumMaster is tasked with doing whatever is necessary to help the team be successful.
The ScrumMaster is not the manager of the team; he or she serves the team, by helping remove
blocks to the team’s success, facilitating meetings, and supporting the practice of Scrum. Some
teams will have someone dedicated fully to the role of ScrumMaster, while others will have a
team member play this role (carrying a lighter load of regular work when they do so). Great
ScrumMasters have come from all backgrounds and disciplines: Project Management,
Engineering, Design, Testing. The ScrumMaster and the Product Owner probably shouldn’t be
the same individual; at times, the ScrumMaster may be called upon to push back on the Product
Owner (for example, if they try to introduce new requirements in the middle of a Sprint). And
unlike a Project Manager, the ScrumMaster doesn’t tell people what to do or assign tasks – they
facilitate the process, to enable the team to organize and manage itself.
In addition to these three roles, there are other important contributors to the success of the
project: Perhaps the most important of these are Managers. While their role evolves in Scrum,
they remain critically important – they support the team in its use of Scrum, and they contribute
their wisdom, expertise and assistance to the project. In Scrum, these individuals replace the time
they previously spent “playing nanny” (assigning tasks, getting status reports, and other forms of
micromanagement) with more time “playing guru” (mentoring, coaching, playing devil’s
advocate, helping remove obstacles, helping problem-solve, providing creative input, and
guiding the skills development of team members). In making this shift, managers may need to
evolve their management style; for example, using Socratic questioning to help the team discover
the solution to a problem, rather than simply deciding a solution and assigning it to the team.

Starting Scrum
The first step in Scrum is for the Product Owner to articulate the product vision. This takes the
form of a prioritized list of what’s required, ranked in order of value to the customer and
business, with the highest value items at the top of the list. This is called the Product Backlog,
and it exists (and evolves) over the lifetime of the product (figure 2). The Product Backlog will
include a variety of items, such as features (“enable all users to place book in shopping cart”),
development requirements (“rework the transaction processing module to make it scalable”),
exploratory work (“investigate solutions for speeding up credit card validation”), and known
bugs (“diagnose and fix the order processing script errors”).
Figure 2. The Product Backlog
The Product Backlog is regularly updated by the Product Owner to reflect changes in the needs
of the customer, announcements by the competition, new ideas or insights, technical hurdles that
appear, and so forth. At any point during the project, the Product Backlog is the single, definitive
view of “everything that needs to be done.” Only a single Product Backlog exists; this means the
Product Owner is required to make prioritization decisions across the entire spectrum of work to
be done.
Items in the Product Backlog will vary widely in size; the larger ones will often be broken into
smaller pieces during the Sprint Planning Meeting, and the smaller ones may be consolidated.
One of the myths about Scrum is that it prevents you from writing detailed specifications; in
reality, it’s up to the Product Owner and Team to decide just how much detail is required, and
this may vary from one Product Backlog item to the next. The general advice is to state what’s
important in the least amount of space necessary – in other words, one doesn’t have to describe
every possible detail of an item, one should just make clear what is necessary for it to be
considered completed. The further down the Product Backlog one goes, the larger and less
detailed the items will be; as they get closer to being worked on, additional detail gets filled in by
the Product Owner.

Sprint Planning Meeting
At the beginning of each Sprint, the Sprint Planning Meeting takes place. In the first part of the
Sprint Planning Meeting, the Product Owner and Scrum Team (with facilitation from the
ScrumMaster) review the Product Backlog, discuss the goals and context for the items on the
Backlog, and provide the Scrum Team with insight into the Product Owner’s thinking. In the
second part of the meeting, the Scrum Team selects the items from the Product Backlog to
commit to complete by the end of the Sprint, starting at the top of the Product Backlog (in others
words, starting with the items that are the highest priority for the Product Owner) and working
down the list in order. This is one of the key practices in Scrum: the team decides how much
work they will commit to complete, rather than having it assigned to them by the Product Owner.
This makes for a much more reliable commitment; first, because the team is making it, rather
than having it “made” for them by someone else; and second, because the team itself is
determining how much work will be required, rather than having someone else decide how much
“should” be required. While the Product Owner doesn’t have any control over how much the
team commits to, he or she knows that the items the team is committing to are drawn from the
top of the Product Backlog – in other words, the items that he or she has rated as most important.
The team does have the ability to pull in items from further down the list if it makes sense (for
example, pulling in a slightly lower priority item that can be quickly completed as part of higher
priority work).
Figure 3. Estimating Available Hours
The Sprint Planning meeting will often last a number of hours – the team is making a very
serious commitment to complete the work, and this commitment requires careful thought to be
successful. The team will begin by estimating how much time each member has for Sprint-
related work – in other words, their average workday minus the time they spend doing things like
critical bug-fixes and other maintenance, attending meetings, doing email, taking lunch breaks,
and so on. For most people this works out to 4-6 hours of time per day available for Sprint-
related work. (Figure 3.)
Once the time available is determined, the team starts with the first item on the Product Backlog
– in other words, the Product Owner’s highest priority item – and working together, breaks it
down into individual tasks, which are recorded in a document called the Sprint Backlog (figure
4). Once tasks are identified, team members will volunteer for them, thinking through
dependencies and sequencing, making time estimates for each task, and making sure the

workload of each individual is reasonable. There will be back and forth with the Product Owner
during this process, to clarify points, verify tradeoffs, break down bigger Backlog items into
smaller pieces, and generally ensure that the team fully understands what’s being asked of it. The
team will move sequentially down the Product Backlog in this way, until it’s used up all its
available hours. At the end of the meeting, the team will have produced a list of all the tasks, and
for each task who has signed up to complete it and how much time they estimate it will take
(typically in hours or fractions of a day). Many teams also make use of a visual task-tracking
tool, in the form of a wall-sized task board where tasks (written on Post-It Notes) migrate during
the Sprint across columns labeled “Not Yet Started”, “In Progress”, “To Verify”, and
“Completed”.
Figure 4. Sprint Backlog
One of the key pillars of Scrum is that once the Scrum Team makes its commitment, the Product
Owner cannot add new requests during the course of the Sprint. This means that even if halfway
through the Sprint the Product Owner decides that they want to add something new, he or she
cannot make changes until the start of the next Sprint. If an external circumstance appears that
significantly changes priorities, and means the team would be wasting its time if it continued
working, the Product Owner can terminate the Sprint; this means the team stops all the work they
are doing, and starts over with a Sprint Planning meeting, and so forth. The disruption of doing
this is great, though, which serves as a disincentive for the Product Owner to resort to it except in
extreme circumstances.
There is a powerful, positive influence that comes from the team being protected from changing
goals during the Sprint. First, the team gets to work knowing with absolute certainty that its
commitments will not change, which only reinforces the team’s focus on ensuring completion.
Second, it disciplines the Product Owner into really thinking through the items he or she
prioritizes on the Product Backlog. Knowing that the commitment is for the duration of the
Sprint makes the Product Owner much more diligent in deciding what to ask for at the
beginning.
In return for all this, though, the Product Owner gets two things. First, he or she has the
confidence of knowing the team has made a very strong commitment to complete the work
they’ve signed up for, and over time Scrum teams get to be very good at delivering this. Second,
the Product Owner gets to make whatever changes he or she likes to the Product Backlog before
the start of the next Sprint. At this point, additions, deletions, modifications, and re-
prioritizations are all completely acceptable. While the Product Owner is not able to make

changes during the Sprint, he or she is always only a Sprint’s duration or less away from making
any changes whatsoever. Gone is the stigma around change – change of direction, change of
requirements, or just plain changing your mind – and it may be for this reason that Product
Owners are among the most ardent Scrum enthusiasts.
Daily Standup Meeting

Once the Sprint has started, the Scrum Team engages in another of the key Scrum practices: The
Daily Stand-Up Meeting. This is a short (15 minute) meeting that happens every workday at an
appointed time, and everyone on the Scrum Team attends; in order to ensure it stays brief,
everyone stands (hence “Stand-Up Meeting”). It’s the team’s opportunity to report to itself on
progress and obstacles. One by one, each member of the team reports just three things to the
other members of the team: What they were able to get done since the last meeting, what they’re
aiming to get done by the next meeting, and any blocks or obstacles that are in their way. The
ScrumMaster makes note of the blocks, and then helps team members to resolve them after the
meeting. There’s no discussion during the Daily Stand-Up Meeting, just the reporting of the
three key pieces of information; if discussion is required, it takes place right after the meeting.
The Product Owner, Managers, and other stakeholders can attend the meeting, but they should
refrain from asking questions or opening discussion until after the meeting concludes – everyone
should be clear that the team is reporting to each other, not to the Product Owner, Managers or
ScrumMaster. While it’s non-standard, some teams find it useful to have the Product Owners
join and give a brief daily report of their own activities to the team.
After the meeting, the team members update the amount of time remaining to complete each of
the tasks that they’ve signed up for on the Sprint Backlog. This information is recorded on a
graph called the Sprint Burndown Chart (figure 5). It shows, each day, how much work
(measured in hours or days) remains until the team’s commitment is completed. Ideally, this
should be a downward sloping graph that is on a trajectory to hit zero on the last day of the
Sprint. And while sometimes it looks like that, often it doesn’t. The important thing is that it
show the team their actual progress towards their goal – and not in terms of how much time has
been spent so far (an irrelevant fact, as far as Scrum is concerned), but in terms of how much
work remains – what separates the team from their goal. If the curve is not tracking towards
completion at the end of the Sprint, then the team needs to either pick up the pace, or simplify
and pare down what it’s doing. While this chart this can be maintained electronically using
Excel, many teams find it’s easier and more effective to do it on paper taped to a wall in their
workspace, with updates in pen; this low-tech solution is fast, simple, and often more visible than
an electronic one.
One of the core tenets of Scrum is that the duration of the Sprint is never extended – it ends on
the assigned date regardless of whether the team has completed the work it committed to or not.
If the team has not completed their Sprint Goal, they have to stand up at the end of the Sprint and
acknowledge that they did not meet their commitment. The idea is that this creates a very visible
feedback loop, and teams are forced to get better at estimating what they are capable of
accomplishing in a given Sprint, and then delivering it without fail. Teams will typically over-
commit in their first few Sprints and fail to meet their Sprint Goal; they might then
overcompensate and undercommit, and finish early; but by the third or fourth Sprint, teams will

typically have figured out what they’re capable of delivering, and they’ll meet their Sprint goals
reliably after that. Teams are encouraged to pick one duration for their Sprints (say, 2 weeks) and
not change it frequently – a consistent duration helps the team learn how much it can
accomplish, and it also helps the team achieve a rhythm for their work (this is often referred to as
the “heartbeat” of the team).
Figure 5. Daily Estimates and Burndown Chart
Sprint Review
After the Sprint ends, there is the Sprint Review, where the team demos what they’ve built
during the Sprint. Present at this meeting are the Product Owner, Team Members, and
ScrumMaster, plus customers, stakeholders, experts, executives, and anyone else interested. This
is not a “presentation” the team gives – there are no PowerPoints, and typically no more than 30
minutes is spent preparing for it – it’s literally just a demo of what’s been built, and anyone
present is free to ask questions and give input. It can last 10 minutes, or it can last two hours –
whatever it takes to show what’s been built and get feedback.

Sprint Retrospective
Following the Sprint Review, the team gets together for the Sprint Retrospective. This is a
practice that some teams skip, and that’s unfortunate because it’s one of the most important tools
for making Scrum successful. It’s an opportunity for the team to discuss what’s working and
what’s not working, and agree on changes to try. The Scrum Team, the Product Owner, and the
ScrumMaster will all attend, and a neutral outsider will facilitate the meeting; a good approach is
for ScrumMasters to facilitate each others’ retrospectives, which enables cross-pollination
among teams.
A simple way to structure the Sprint Retrospective is to hang two sheets of poster-sized paper
labeled “What’s Working Well” and “What’s Not Working, or Could Work Better” – and then
have each person add several items to either. As items are repeated, check-marks are added next
to them, so the common items become clear. Then the team looks for underlying causes, and
agrees on changes to make in the upcoming Sprint, along with a commitment to review the
results at the next Sprint Retrospective. Another useful practice is for the team to label each of
the items in each column with either a “C” if it is caused by Scrum, or a “V” if it is made visible
by Scrum (in other words, it would be happening with or without Scrum, but Scrum makes it
known to the team). The team may find a lot of C’s on the “What’s Working” side of the board,
and a lot of V’s on the “What’s Not Working”; this is good news, even if the “What’s Not
Working” list is a long one, because the first step to solving underlying issues is making them
visible, and Scrum is a powerful catalyst for that.
Starting the Next Sprint

Following the Sprint Review Meeting, the Product Owner takes all the input, as well as all new
priorities that have appeared during the Sprint, and incorporates them into the Product Backlog;
new items are added, and existing ones are modified, reordered, or deleted. Once this updating of
the Product Backlog is complete, the cycle is ready to begin all over again, with the next Sprint
Planning Meeting.
One practice many teams find useful is to hold a Prioritization Meeting toward the end of each
Sprint, to review the Product Backlog for the upcoming Sprint with the Product Owner. In
addition to giving the team an opportunity to suggest items the Product Owner may not be aware
of – technical maintenance, for example – this meeting also kicks off any preliminary thinking
that’s required before the Sprint Planning Meeting.
There’s no downtime between Sprints – teams will often go from a Sprint Review one afternoon
into the next Sprint Planning Meeting the following morning. One of the values of Agile
development is “sustainable pace”, and only by working regular hours at a reasonable level of
intensity can teams continue this cycle indefinitely.
Release Planning
Sprints continue until the Product Owner decides the product is almost ready for release, at
which point there may be a “Release Sprint” to do final integration and testing in preparation for

launch. If the team has followed good development practices along the way, with continuous
refactoring and integration, and effective testing during each Sprint, there should be little cleanup
required.
A question that’s sometimes asked is how, in an iterative model, long-term release planning
takes place. At the beginning of a project the team will do high-level release planning; since they
cannot possibly know everything up front, the focus is on creating a rough plan to give the
project broad direction, and clarify how tradeoff decisions will be made (scope versus schedule,
for example). Think of this as the roadmap guiding you towards your final destination; which
exact roads you take and the decisions you make during the journey will be determined en route.
Some releases are date-driven; for example: “We will release version 2.0 of our project at a
trade-show on November 10.” In this situation, the team will complete as many Sprints (and
build as many features) as is possible in the time available. Other products require certain
features to be built before they can be called complete and the product will not launch until these
requirements are satisfied, however long that takes. Since Scrum emphasizes producing
potentially shippable code each Sprint, teams may choose to start doing interim releases, to allow
the customer to reap the benefits of completed work sooner.
Most Product Owners will choose one release approach, but inform it with the other – for
example, they’ll decide a release date, but they’ll work with the team to come up with a rough
estimate of the Backlog items that will be completed by that date. In situations where a “fixed
price / fixed date / fixed deliverable” commitment is required – for example, contract
development – at least one of those variables must have a built-in buffer to allow for uncertainty
and change; in that respect, Scrum is no different from other development methodologies.
Common Challenges
Scrum tends to make visible a lot of issues that exist within the team, particularly at the
beginning. For example, most teams are not good at estimating how much they can get done in a
certain period, and so will fail to deliver what they committed to in the first Sprint. To the team,
this feels like failure, and could cause them to question their adoption of Scrum. In reality, this
experience is the necessary first step toward becoming better at estimating, and with the support
of an experienced Scrum practitioner, the team can be helped to see it this way. Another
difficulty a team might have is around the Daily Standup Meeting – getting all team members to
commit to gather at a specified time, on time and every day without fail, may require the team to
operate at a higher level than it’s accustomed to, and some teams will be unable to do this.
One very common mistake teams make, when presented with a Scrum practice that challenges
them, is to change the practice, not change themselves. For example, teams that have trouble
delivering on their Sprint commitment might decide to make the Sprint duration extendable, so
they never run out of time – and in the process, ensure they never have to learn how to do a
better job of estimating and managing their time. In this way, without training and the support of
an experienced Scrum coach, teams can morph Scrum into just a mirror image of their own
weaknesses and dysfunction, and undermine the real benefit that Scrum offers: Making visible
the good and the bad, and giving the team the choice of elevating itself to a higher level.
Another common mistake is to assume that a practice is discouraged or prohibited just because
Scrum doesn’t specifically require it. For example, Scrum doesn’t specifically require the

Product Owner to set a long-term strategy for his or her product; nor does it require engineers to
seek advice from more experienced engineers (for example, their managers) about complex
technical problems. Scrum leaves it to the individuals involved to make the right decision; and
in most cases, both of these practices (along with many others) would be well-advised. To put it
another way: “Just because Scrum doesn’t say anything about breakfast doesn’t mean you have
to go hungry!”
Something else to be wary of is managers imposing Scrum on their teams; Scrum is about giving
a team space and tools to self-organize, and having this dictated from above is not a recipe for
success. A better approach might begin with a team learning about Scrum from a peer or
manager, getting comprehensively educated in professional training, and then making a decision
as a team to follow the practices faithfully for a defined period (say, 90 days); at the end of that
period, the team will evaluate its experience, and decide whether to continue.
The good news is that while the first Sprint is often very challenging to the team, the benefits of
Scrum tend to be visible by the end of it, leading many new Scrum teams to exclaim: “Scrum is
hard, but it sure is a whole lot better than what we were doing before!”
Results From Scrum

The benefits of Scrum reported by teams come in various aspects of their experience. At Yahoo!,
we have migrated nearly 50 projects to Scrum in the last 18 months, totaling almost 600 people,
and the list of teams using it is quickly growing. These projects have ranged from consumer-
facing, design-heavy websites like Yahoo! Photos, to the mission-critical back-end infrastructure
of services like Yahoo! Mail, which serves hundreds of millions of customers; they range from
entirely new products like Yahoo! Podcasts, which used Scrum from ideation through launch
(and won a Webby Award for best product in its category that year), to more incremental
projects, which included work on new features as well as bug fixes and other maintenance; and
we’ve used Scrum for distributed projects, where the team is on separate continents. Once each
quarter, we survey everyone at Yahoo! that is using Scrum (including Product Owners, Team
Members, ScrumMasters, and the functional managers of those individuals) and ask them to
compare Scrum to the approach they were using previously. An in-depth white paper on
Yahoo!’s results is being prepared, but some early data is presented here:
• Productivity: 68% of respondents reported Scrum is better or much better (4 or 5 on a 5-

point scale); 5% reported Scrum is worse or much worse (1 or 2 on a 5-point scale); 27%
reported Scrum is about the same (3 on a 5-point scale).
• Team Morale: 52% of respondents reported Scrum is better or much better; 9% reported
Scrum is worse or much worse; 39% reported Scrum is about the same.
• Adaptability: 63% of respondents reported Scrum is better or much better; 4% reported
• Accountability: 62% of respondents reported Scrum is better or much better; 6%
reported Scrum is worse or much worse; 32% reported Scrum is about the same.
• Collaboration and Cooperation: 81% of respondents reported Scrum is better or much
better; 1% reported Scrum is worse or much worse; 18% reported Scrum is about the
same.

• Team productivity increased an average of 36%, based on the estimates of the
Product Owners.
• 85% of team-members stated that they would continue using Scrum if the decision
were solely up to them.
© 2006 Pete Deemer and Gabrielle Benefield

Reproduced with permission from the authors.

Rolling out Agile at a large Enterprise
Gabrielle Benefield
Senior Director of Agile Development at Yahoo! Inc., 2007
Abstract
Yahoo! is a $50B company that has one of the largest Agile implementations in the world. The
adoption of Scrum and Agile practices has been steadily growing over the past two years, and
now encompasses more than 150 Yahoo! teams and more than 1500 people in the United States,
Europe, and Asia-Pacific. The projects range from new product development such as Yahoo!
Autos to heavy-duty infrastructure work on Yahoo! Mail which serves 250 million users each
month around the globe.
Introduction
In the highly competitive Internet space, getting products to market quickly while being both
flexible and adaptive to change is critical. Yahoo! needed a process that supported an Internet
start-up culture within the structure of providing products and services to more than 500 million
users worldwide.
Background
Yahoo! went from being a small start-up and grew to a large enterprise company quickly. The
company still seems like a large start-up with the good and bad that comes with it. The things
people liked about being a start-up was working closely with a small set of people, being able to
quickly get products to market, the code base was relatively small and simple to work within and
technical debt had not built up in it. The interdependencies between products are small and
scaling for a small set of users is easy to deal with on the backend and in the application layer.
Standardization in brand, user interface and tools is fairly straightforward. If you need
something you usually know who to go to and how to find them to get things done. It is also
very exciting as you ramp up quickly and the money starts rolling in. As a company grows it
needs to deal with the complexity of many moving pieces, more people who you don’t know, the
logistics of seating, feeding and making thousands of employees happy. You are under intense
public scrutiny, particularly if you are a Public company. Legal concerns heighten and if
anything goes wrong the effects can be massive and broad reaching. No longer can you simply
launch a product, multiple stakeholders need to be involved in the decision making, multiple
properties may be affected if interconnected in the Yahoo! portal.
It is very hard to track down information as the size of the company grows. It appears to be a
natural trend for start-ups that grow into large companies to hire in people with big company
experience. Sometimes these people can add a lot of value, sometimes they can bring in overly
bureaucratic processes that are at odds with the “just-get-things-done” start-up culture which
drew in the employees in the first place. People who build systems from the ground up have a
lot of passion and ownership and aren’t always ready to share and the systems, platforms and
tools put in place to service one set of needs, that of a small company no longer service hundreds

of thousands, sometimes millions of users. In an attempt to make sense of scaling pain the knee
jerk reaction is to often put in processes to manage and control software development. Often
these look excellent on paper and in theory should work, unfortunately they can be at odds with
the main ingredient in organizations, people.
Yahoo! attempted to control the software development process and released a waterfall process
called the “Product Development Process” in 2002. The process was rolled out globally and the
use of it was mandatory. Unfortunately for the creators (or perhaps fortunately) a lot of teams
simply ignored the process, or where they couldn’t ignore it paid lip service and made it look like
they adhered to the steps. The teams that did follow it found it was heavy, slowed them down
and added little real value. Management felt like they were in control but the teams rebelled.
There were some grass roots efforts in 2004 to try out some Agile practices such as Extreme
Programming and Scrum. This was led by the team members or in one case by a smaller
company (Stata Labs) that Yahoo! acquired. Tobias Mayer, an engineer on a team started a
small grass roots movement to spread the word and found his way to the VP of Product
Development at the time, Pete Deemer. Pete to his credit realized that the heavy weight process
he had helped rollout was not succeeding as he had hoped and was curious about Agile. Tobias
asked Ellen Salisbury, an engineering leader from Stata Labs to give an internal tech talk on her
experiences. The talk was inspirational and piqued peoples interests. In a lucky confluence of
events, Pete happened to contact Jeff Sutherland and Ken Schwaber when he was in the Bay
Area (where Yahoo!’s main headquarters are located) on the same day as the executive team had
an offsite dinner. Pete invited Jeff to be a guest speaker at the dinner to share his experiences
with Scrum. The executive team was very inspired upon hearing his research that they decided
to sponsor a pilot program on the spot. This led to the official rollout of the Scrum pilot program
in February of 2005.
Pete and Tobias evangelized the benefits of Scrum amongst their contacts and managed to get
four teams to volunteer to try Scrum for two months and participate in a survey to gather data
about their experiences. The teams covered a broad set of products and services including the
new Yahoo! Photos 3.0, a new backend for Yahoo! Mail, internal tools for managing small
business sites and a media site re-design. A subset of team leads were sent to a Certified Scrum
Master class with Ken Schwaber. The teams used a very standard out-of-the-box Scrum
framework to address prioritization concerns, self-organization and teamwork, greater customer
involvement and incremental product releases. At this stage little attention was put on technical
practices as Scrum was seen as an easy first step to test the waters. At the end of their first month
of using Scrum all the team members and their managers were invited to participate in an online
survey to anonymously gather their feedback. The responders’ received a custom printed Scrum
t-shirt for participating, which also served a dual purpose to promote Scrum. The overall
response rate was 71% (~85% for Scrum Pilot team members). The questions asked sought to
track and collate information in the areas that Yahoo! wanted to improve upon which were
mainly qualitative and focused on the human aspects of software development. The questions
asked people to rate their experiences against their previous process.

the feedback was positive; the teams liked the process and experience, and management saw
positive results. Two years later we have over one hundred teams spread around the globe and
we continue to grow rapidly.
The culture at Yahoo is very much like a large start-up. There is a constant stream of innovative
ideas from every department, new product features are constantly released, and the company
strives to be the first to market with new services, while meeting the needs of our users. The
founders still work at the company, and have remained involved in day-to-day activities. They
continue to instill in all employees a love for the culture and work. The company is committed to
preserving the things that make Yahoo! great while putting in some process and practices to help
teams deliver better products faster; this was the greatest challenge we faced when introducing
Agile.
We started with Scrum, implementing small, cross-functional teams to address organizational

issues, highlight business priorities and most importantly, create a collaborative environment.
Next we added in Agile engineering practices and Lean fundamentals to deliver greater business
value.
Kick-off
To kick off the program, we focused on building excitement internally, and motivating
employees to get involved. We invited guest speakers like Ken Schwaber and Jeff Sutherland,
the inventors of Scrum to address employees, and had a new employee give a talk on her
experiences with Scrum at a previous company. We were also fortunate to have the VP of
Product Development evangelize the benefits of Scrum to the top executives in the language that
they could relate to.
Once we laid the ground work for the pilot program we experimented with an engagement model
that allowed us to coach multiple teams efficiently. Where we had the bandwidth we would work
closely with teams to get them up and running.

We found it worked well to start by spending some time preparing the Product Backlog with the
Product Owner, then train the whole team together. A coach was assigned to lead the first Sprint
planning meeting, stand-up, Sprint review and retrospective. For the second sprint we let the
Scrum Master lead the team and we shadowed them. After that, depending on the team, we
would do drop-ins and try to keep a good connection with the Product Owner and Scrum Master
to help guide them through the early iterations. At this point, some teams were off and running,
while others with difficult issues needed some additional coaching. We also found that teams
unlearnt things over extended periods of time so we stayed in contact and continually re-engaged
to lead some master retrospectives and give them objective advice and coaching.
We found an evolutionary approach to be far more successful than a revolutionary one for long
term good results. At times we needed to push the envelope and take risks but trying to rip the
band-aid off too quickly is dangerous and can lead to ultimate failure. For us, organizational
change meant getting buy in at all levels, having people see results and be able to learn in a safe
environment.
Management Support
Agile is all about experimentation and the ability to inspect and adapt as an empirical approach.
At the end of the day, nothing really mattered but what employees actually experienced and all
ideas were useless unless we executed on them and could prove that they worked. One thing that
was key to the ongoing funding and success of the program was a quarterly survey. The survey
was distributed to all the team members and managers, and we used the data to help us improve,
worked with teams that most needed it, and distributed the data back to management (see
examples of the questions and responses at the end of this report). Over time the survey became
less useful as a way to compare Scrum with the old process as people only had Scrum experience
at the company. We also found that after a couple of surveys people didn’t want to do anymore,
so long term data gathering is not effective using a survey alone.
Employee Support
We decided to keep the program voluntary and still do. It was agreed that for a process to be
truly successful it needed to stand on its own merits. Although Agile was bought in from the top-
down, the fact that the program was never mandated meant it had bottom-up support.
While we built relationships at all levels and marketed the successes to the management team,
the real driving force was letting the word spread virally. The teams using Scrum spread the
word about the process and people moving throughout the company seeded new teams. We
leveraged the experiences of the people in the trenches to create a very effective promotion
engine.
Feedback
We tried to keep a lot of transparency around the process and feedback we were receiving. We
gained far more credibility by being open and letting people know that the process is not a silver
bullet and acknowledging that change is hard. By being upfront with the challenges we were able
to confront difficult issues and improve. We had panel discussions and “tech talks” from

different Scrum teams to share their experiences, the challenges of transitioning to Agile and
how they dealt with issues. A core tenet of Agile is around transparency and we felt this should
also go for the methodology itself.
Top-down mandates that tried to enforce Scrum practices in a by-the-book fashion always
backfired for us, as did teams that followed the practices so zealously they lost the forest for the
trees. We did have teams that simply weren’t ready or willing to use Agile and we had to respect
that; we didn’t want to force our coaches to be the process police and become part of the
problem.
We didn’t get too hung up on having the perfect tracking tools, training materials, coaching
program, etc. in place. We made a lot of mistakes but we also improved quickly based on
iterative feedback. Our philosophy dictated that it was better to make the flight than have our
bags packed perfectly and still be waiting on the ground.
Roll-out
To kickoff an enterprise Agile rollout, we found it really helped to have people with real
experience in the field. The foundation of the overall strategy was built on lessons learned and
understanding how to deal with change. We built out a centralized team of coaches who were
passionate and good at building relationships.
The team had a mixture of skills including Product management, QA, Design, Extreme
Programming, Scrum and Lean. It was useful to have people with specializations in additional to
generalized coaching so they could build bridges into different functional groups. Personality
was also key. We needed people with passion and enthusiasm, as we were only as good as the
relationships we could form. Having people who were overly zealous or abrasive would have
quickly bought the program to a halt. One important aspect of hiring was to find people with
strong skills in collaboration and building consensus.
Next we found that the best Agile champions were the people already in the teams, from all
levels and disciplines. These people knew the context and the challenges of their particular
situation and could adjust the process to meet their team’s needs. Finding good people who really
get it and training them up to help their own team is one of the differentiators, and is the only
way to scale effectively in a large organization.
Challenges
Managers often feel left out when the team becomes more self-organized and don’t know how to
transition from the traditional command and control model to one of a strategic and supportive
leader. They sometimes lash out or subvert the process out of fear. Where we came across people
who are anti-Agile, we tried to get them to understand their changed role and to give them some
responsibility. Training and coaching these people is worth the investment. We also did have to
deal with the reality that not everyone is willing or can change and ultimately the new
environment may no longer be a good fit for them anymore.

We wanted people to participate in the surveys but after a couple of rounds they were bored with
having to fill out survey information all over again. So we offered free t-shirts. These weren’t
any old t-shirt, they were cool. We didn’t even brand them as the sheer ambiguity was very
appealing and people would be curious to know what the t-shirts meant and this would open
some doors for us. We had people signing up to try Scrum just to get a t-shirt.
Another challenge we had and still have to some extent is to keep to the key tenets of Scrum
while adapting to different contexts. Scrum provides an extremely flexible framework and how
you apply it is an open-ended question.
We have a very strong and brilliant design group at Yahoo! Our products are heavily consumer
focused so design is very important to us. The Designers initial reaction to Agile was similar to
the way engineering architects react when faced with the idea that you don’t design everything
up front that you constantly re-factor and that requirements will change.
Working to understand the challenges and finding common ground helped improve the situation.
There are some things that do fit with design thinking. Designers do want to adapt and work
incrementally. User stories that are focused on the customer are also warmly greeted. Lean
thinking in keeping the features to a minimum and doing them well also strikes a positive note.
We tried to be flexible, to listen to the design viewpoint and to help the whole team find a way to
work together in a way that made sense to them. If the whole team was not able to find common
ground that worked at the expense of key members the holistic team did not succeed.
We allowed the teams to find a way to work within the Agile framework that suited their context
and needs, using Scrum as a flexible framework. We have teams that do overlap work within
each iteration, where some design is done looking forward to the next iteration, some user testing
of work completed in a previous iteration, and handoffs during the iteration occur. This may not
look like pure Scrum but it works for teams developing consumer facing products and Ken
Schwaber has always tried to get across the values rather than the rules of Scrum so we took this
very much to heart. If we tried to enforce only working on tasks for the current iteration during
the current iteration the designers would have mutinied and the team collaboration would have
suffered. Again, Scrum is adaptive and if it works for people they keep doing it.
We have found though that the number one reason designers like Agile is the collaboration
aspect. If the team spirit is strong and collaboration between team members is working they can
overcome the logistical difficulties as they work together.
Training
It was and still is extremely challenging to get executives and senior managers to Scrum training
due to their busy schedules, but it is also invaluable and worth the investment. One General
Manager took the two day Certified Scrum Master class with his team and said it was a great
experience. He got to hear the tough challenges and issues the team were facing while the team
got a lot of insight into the business challenges and vision. This established a healthy base for
ongoing two-way conversations. The manager bought a lot of credibility for investing time to sit
and learn in the same room as everyone else.

Even though we had great internal coaches, we were very understaffed. We realized that having
great people like Mike Cohn, Ken Schwaber, Jeff Sutherland, Mary and Tom Poppendieck was
crucial to getting the program off the ground. It is worth spending the money on consultants if
you lack internal expertise as they can ultimately save the company money if you apply them
wisely.
Next time
The Agile Development team at Yahoo! approached the strategy with pragmatism and
adaptability, and has experienced great success with the program. Nevertheless, there are always
things you can change, including dedicating more resources and funding to the project, but until
we could prove the process worked the business was not going to invest a lot. The whole process
has been and still is all about learning and adapting as we go. The failures propelled us to new
levels and it was important to allow teams to understand that failure is in itself an effective
learning mechanism.
Fund the internal coaching team adequately

It would have been great to have the internal coaching team staffed adequately so we could get a
good scaling strategy in place earlier. I was working by myself for a period of time, and the
group only had two full time coaches consistently for the first year. We put new meaning to the
term “lean”. It took a long time to get more resources assigned to the central coaching team and
this only came after financial analysis helped to prove each coach’s value (around 1.4 million
dollars saved per year for each coach helping 10 teams be more productive). This would have
made a lot more teams a lot more successful and we could have scaled faster and better with
additional resources.
Encourage deeper engagement from coaching staff

It would have been preferable to coach teams more intensely rather than being so broadly
focused. Due to constrained resources and huge demand it was impossible to work as deeply
with teams as we would have liked, and this showed very clearly in the survey results. The teams
with adequate coaching showed productivity increases of 2-3 times more than teams trying to
work by themselves.
There are teams that kicked off by attending a public class and had no follow-up coaching, due to
either bandwidth restrictions or not realizing the value of it. We sometimes run into these teams
or hear about them through the grapevine and find out they are not really doing Scrum at all but a
hybrid that allows them to continue their dysfunctional practices while calling it Scrum. This is a
major problem so now when teach classes we stress the importance of coaching.
It would have also been great to have solid engineering coaches available from day one to work
with new teams, helping them set up build and test systems and introducing Agile engineering
concepts. It is very challenging to deliver incremental products without good engineering
discipline and this has definitely held back the productivity and quality of many of our teams.

Deeper management involvement
We could have pushed harder to get senior management to attend focused training. Having only
a shallow understanding has lead to misunderstandings that could have been avoided if we had
been able to do more of this. Again, in an ideal world with more resources we could have done
more targeted training for management, product management, QA and design to better integrate
them into the collective team spirit. We are currently working towards this.
Focus on the majority

It is a given that not everyone will be happy in every situation. We found in the process of
introducing Scrum to Yahoo! that we did have people who were very negative towards our
efforts. Fear, control and politics are constantly challenging and we had to simply realize we
weren’t always going to make friends. People will react to the changes and if they didn’t, you
would probably be telling them what they wanted to hear, not perhaps what they need to hear.
Conclusion
Although we have over 100 teams at Yahoo!, we still have a long way to go. Some days it feels
like we are winning and Agile is spreading its love over the whole company, other days teams
revert to bad practices and new blocks appear that feel impossible to break through. Some teams
are very Agile, others do mini-waterfalls and call it Agile. Change is difficult, and to change a
company as large as Yahoo! sometimes feels like trying to steer the Titanic with a small paddle.
We learned that patience is important, as is remembering that even the smallest of incremental
improvements have a massive payoff when you do them at large scale.
Results from Scrum

The benefits of Scrum reported by teams come in various aspects of their experience. Once each
quarter, we surveyed everyone at Yahoo! using Scrum (including Product Owners, Team
Members, ScrumMasters, and the functional managers of those individuals) and ask them to
compare Scrum to the approach they were using previously. Below are some results from our
previous surveys:
• Productivity: 68% of respondents reported Scrum is better or much better (4 or 5 on a 5-point

scale); 5% reported Scrum is worse or much worse (1 or 2 on a 5-point scale); 27% reported
Scrum is about the same (3 on a 5-point scale).
• Team Morale: 52% of respondents reported Scrum is better or much better; 9% reported
• Adaptability: 63% of respondents reported Scrum is better or much better; 4% reported Scrum
is worse or much worse; 33% reported Scrum is about the same.
• Accountability: 62% of respondents reported Scrum is better or much better; 6% reported
• Collaboration and Cooperation: 81% of respondents reported Scrum is better or much better;
1% reported Scrum is worse or much worse; 18% reported Scrum is about the same.
• Team productivity increased an average a 37% increase, based on the estimates of the
Product Owners.

• 86% of team-members stated that they would continue using Scrum if the decision were
solely up to them.
Contact: Gabrielle Benefield ([email protected])

Capturing Extreme Business Value: 1000% Annual Return
on Investment in Scrum Trainers
PatientKeeper, Inc., 2007
Abstract
In 2005, Jeff Sutherland worked together with Peter Deemer at Yahoo! to brief the Yahoo! senior
management team on Scrum. After senior management made the decision to move forward with
Scrum, a productivity analysis of rollout of Scrum at a previous large enterprise (IDX Systems,
now GE Healthcare) was used to calculate an annual ROI of 1000% on a three year rollout of
Scrum at Yahoo! After two years of deployment and Scrum rollout to over 100 teams the rate of
return at Yahoo! for investment in each internal Scrum trainer was $1.4 based on training of 10
teams annually per trainer, or roughly 1000% return on investment . Teams coached by a Scrum
trainer achieved 3-4 times the productivity gains of uncoached teams [27] .
Introduction
The internal rate of return on investment in Scrum training is quite high. Many companies have
doubled the rate of software production on the average for all teams measured. Recently a CMMI
Level 5 company cut the costs of software projects in half and reduced measured defects by 40%
while still maintaining CMMI Level 5 compliance for all projects [19]. Even the best companies
will radically improve performance by introducing Scrum and they will achieve far more than
1000% rate of return on investment in Scrum training.
This paper addresses the ROI on Scrum training for the average large company with thousands
of employees and hundreds or thousands of developers. These companies have established
heavyweight processes over many years that are bureaucratic and loaded with waste. While on
the surface it would appear easy to provide substantial gains by eliminating the most obvious
sources of inefficiency, introducing a radically new process company wide can be slow and
painful. Scrum has a systematic continuous quality improvement process that identifies and
prioritizes companywide impediments to progress.
IDX Systems (now GE Healthcare): Scaling Scrum for the First Time
During the summer of 1996, IDX Systems hired Jeff Sutherland as senior VP of engineering and
product development. IDX had over 4,000 customers and was one of the largest US healthcare
software companies, with hundreds of developers working on dozens of products. Here was an
opportunity to extend Scrum to large-scale development.
The approach at IDX was to organize the entire development group into an interlocking set of
Scrums. While this was the first large development team to try this approach, the strategy has
now been executed many times and documented by Ken Schwaber in “Scrum in the Enterprise”
[28]. Every part of the organization was team based, including the management team, which

included two vice presidents, a senior architect, and several directors. Front-line Scrums met
daily. A Scrum of Scrums, which included the team leaders of each Scrum in a product line, met
weekly, The management Scrum met monthly.
The key learning at IDX was that Scrum scales to any size. With dozens of teams in operation,
the most difficult problem was ensuring the quality of the Scrum process in each team,
particularly when the entire organization had to learn Scrum all at once. IDX was large enough
to bring in productivity experts to monitor throughput on every project. While most teams were
only able to double the industry average in function points per month delivered, several teams
moved into a hyperproductive state, producing deliverable functionality at four to five times the
industry average. These teams became shining stars in the organization and examples for the rest
of the organization to follow.
One of the most productive teams at IDX was the Web Framework team that built a web
frontend infrastructure for all products. The infrastructure was designed to host all IDX
applications, as well as seamlessly interoperate with end user or third party applications. The
Web Framework was created by a distributed team with developers in Boston, Seattle, and
Vermont who met by teleconference in a daily Scrum meeting. The geographic transparency of
this model produced the same high performance as co-located teams and has become the
signature of hyperproductive distributed/outsourced Scrums at Xebia in the Netherlands/India
and Exigen Services in United States/Russia [29].
The quality of software of many of the hyperproductive Scrum teams can be extraordinarily
high. The IDX Web Framework was first deployed in 1997 and in 2007 was selected as the core
web technology for GE Healthcare systems. However, very few of the IDX software teams
achieved the hyperproductive state. On the average, based on function point analysis by Capers
Jones company, Software Productivity Research, IDX only achieved average productive gains of
240%, primarily due to loss of production because of large Scrum teams, up to 15 people in size.
It is well understood today that these large teams cause significant loss in productivity and make
it impossible to achieve linear scalability, one of the key features of well-executed Scrum
implementations in the best Scrum companies.
Summary of Productivity Gains at IDX

The budget of the IDX development organization was almost $50M per year and this was
sufficient size to a detailed analysis of baseline productivity before Scrum and productive gains
generated by Scrum. An independent consulting firm, Software Productivity Research (SPR),
was hired to do function point analysis of every IDX software product. Jeff Sutherland had
worked with Capers Jones [30], the founder of SPR, during the original creation of Scrum and
wanted to compare the productivity goals designed into the Scrum process with actual
performance in the field. Some of the IDX products were well over 12000 function points, the
equivalent of over a million lines of Java or C# code and the smaller products were typically
5000-6000 function points. Applications were financial and clinical products for operating
hospitals and independent physician groups at thousands of sites. Implementation platforms
varied from Mumps to Cobol to Java and the latest Microsoft tools and languages available.

Function points were chosen as an industry standard measure that was independent of the
software development language and environment. This was to ensure realistic comparison of
productivity across technologies and development teams, as well as comparability with external
industry data. External professional experts were hired to calculate function points in order to
provide research quality data. Function points are not easy calculated by the average
development team and not recommended as an operational strategy. Story points have emerged
as industry best practice for measuring Agile development team velocity [31]. They are easily
calculated and useful for release planning. However, they are not comparable across teams or
across companies so were not suitable for research data on the first deployment of Scrum in a
large enterprise.
At every release point for every product, the number of function points was recalculated to
reflect the new features by Software Productivity Research consultations. The increase in
function points was divided by person months for the fully burdened development teams,
including design, coding, testing, administrative, and management staff. Initial velocity of all
development teams was industry average at 2-3 function points per staff month.
Some teams accelerated into a hyperproductive state using Scrum, achieving 5-10 times industry
average performance. These teams were about 10% of the organization and achieved an average
productivity increase of 666%. A small number of teams (less than 5%) experienced failures for
either personnel or technology reasons. Occasionally, a team would not be able to work together
effectively and was reorganized or disbanded, usually during the first Sprint. Less frequently, a
high performance team had taken a calculated risk on new technologies (always approved by
management) and technical failure of some Sprints was anticipated (even encouraged) in order to
gain a technology lead in the market. There were no failures of large projects, only failure to
delivery software in isolated Sprints or a short series of Sprints. In the case of team failures, the
teams were always reformed. In the case of technology failures, efforts were redoubled in
subsequent Sprints to overcome research and development challenges.
The remaining 85% of teams achieved an average velocity of 5-6 function points per staff month,
averaging a 100% gain. The net productivity gain of all teams combined was 240%. This was
viewed as a failure to achieve Toyota level performance. However, it was a good first start for
enterprise wide deployment of Scrum. Recently, development teams of comparable size have
achieved over 10 times industry average velocity consistently. All teams in an entire organization
have gone hyperproductive achieving the original design goal of Scrum.
It is important to note that these productivity gains were achieved at a sustainable pace with
increased employee retention and enhanced ability to hire the best people in the software
industry due to the high quality working environment provided by Scrum for developers. The
hyperproductive teams were always the most spirited teams who loved their jobs and worked
closely together like a professional sports team. Hyperproductivity is not achieved by working
harder, but only by working better through intense communication, mutual support, and an
“effortless” skill that makes hard things look easy. Think of Michael Jordan going up for a
basketball shot. The team has set him up and the shot often looks so smooth and easy it generates
exhilaration in both the players and the spectators.

Proposed ROI Strategy for Yahoo
To be completed …

Chapter 2: The First Scrum
Scrum was derived from best practices in the Japanese auto and consumer products industry at
Easel Corporation in 1993. The story of the first Scrum was published by the Cutter Agile
Advisory Service outlining lessons learned on the first Scrum [21].
Senior management support is very helpful in implementing Scrum. At Easel, Scrum was used
for the first time on the most critical project in the company. Scrum initial design was to take the
toughest project and turn it into a success. In that environment, it would never have begun
without the support of the Easel CEO. How do you sell Scrum to management without any past
experience or references to rely upon?
In 1993, the 1986 paper in the Harvard Business Review by Takeuchi and Nonaka [1] combined
with the Coplien paper on the development of Quattro for Windows at Borland [2] triggered the
first daily Scrum meetings and the monthly Sprint cycle. Japanese best practices in new product
development at Honda and Fuji-Xerox reminded Takeuichi and Nonaka of the Scrum formation
in Rugby. It would take more than these papers to convince a CEO under pressure to approve a
new process he had never seen before. What were the key arguments?
Today, there are ROI analyses, experience reports, success stories, and lean manufacturing
practices that help make a compelling case for selecting Scrum. Toyota has emerged as the
leading Japanese example of lean product development. Many publications document Toyota’s
process which achieves 4 times the productivity and 12 times the quality of a typical U.S.
competitor. This is what can be expected from a high quality Scrum implementation in software
development. As a result OpenView Ventures Partners asked senior management teams in all of
their portfolio companies to start learning about Scrum by reading the “Toyota Way” [13].

Agile Development: Lessons Learned from the First Scrum
Introduction
The first Scrum for software development teams was started Easel Corporation in 1993, where
we built the first object-oriented design and analysis (OOAD) tool that incorporated round-trip
engineering. In a Smalltalk development environment, code was auto-generated from a graphic
design tool and any changes to the code from the Smalltalk integrated development environment
(IDE) were immediately reflected back into design.
Since the product was directed toward enterprise software development, we spent a lot of time
analyzing best practices in software development methodologies.
Reviewing Software Development Processes

We realized we needed a development process that fit an enhanced version of rapid application
development, where visualization of design could result immediately in working code. This led
to an extensive review of both the literature and the real experience from leaders of hundreds of
software development projects.
There were some key factors that influenced the introduction of Scrum at Easel Corporation.
“Wicked Problems, Righteous Solutions” [32] reviewed the reasons why the waterfall approach
to software development does not work.
• Requirements are not fully understood before the project begins,

• Users know what they want only after they see an initial version of the software,
• Requirements change often during the software construction process,
• And new tools and technologies make implementation strategies unpredictable.
DeGrace and Stahl reviewed “All-at-Once” models of software development that uniquely fit
object-oriented implementation of software and help resolve these challenges.
“All-at-Once” models assume that the creation of software is done by simultaneously working on
requirements, analysis, design, coding, and testing, then delivering the entire system all at once.
The simplest “All-at-Once” model is a single super-programmer creating and delivering an
application from beginning to end. All aspects of the development process reside in one person’s
head. This is the fastest way to deliver a product that has good internal architectural consistency
and is the “hacker” model of implementation. For example, in a project before the first Scrum, a
single individual spent two years writing every line of code for the Matisse object database used
to drive $10B nuclear reprocessing plants worldwide. At less than 50,000 lines of code, the

nuclear engineers said it was the fastest and most reliable database ever benchmarked for nuclear
plants. Brooks has documented a variant of this approach called the Surgical Team, which IBM
has shown to be their most productive software development process [33].
The Surgeon or super-programmer approach has the fatal flaw that there are at most one or two
individuals even in a large company that can execute this model. For example, it took years for a
leading team of developers to understand the conceptual elegance of the Matisse object server
technology enough to maintain it. The single-programmer model does not scale well to large
projects.
The next level of “All-at-Once” development is handcuffing two programmers together, as in

pair programming in the eXtreme Programming paradigm [23]. Here, two developers working at
the same terminal deliver a component of the system together. This has been shown to deliver
better code (usability, maintainability, flexibility, extendibility) faster than two developers
working individually [34]. The challenge is achieve a similar productivity effect with more than
two people. What is the best way to work with multiple teams of people on large software
projects?
Our scalable, team-based “All-at-Once” model was motivated by the Japanese approach to new
product development. We were already using an iterative and incremental approach to building
software [35]. It was implemented in slices in which an entire piece of fully integrated
functionality worked at the end of an iteration. What intrigued us was Takeuchi and Nonaka’s
description of the team-building process for setting up and managing a Scrum [1]. The idea of
building a self-empowered team in which everyone had the global view of the product on a daily
basis seemed like the right idea. The approach to managing the team, which had been so
successful at Honda, Canon, and Fujitsu, also resonated with the systems thinking approach
promoted by Professor Senge at MIT [15].
We were prodded into setting up the first Scrum meeting after reading Coplien’s paper on
Borland’s development of Quattro Pro for Windows [2]. The Quattro team delivered one million
lines of C++ code in 31 months with a 4-person staff that later grew to 8. This was about 1,000
lines of deliverable code per person per week, the most productive software project ever
documented. The team attained this level of productivity by intensive interaction in daily
meetings with project management, product management, developers, documenters, and quality
assurance staff.
Why the Easel CEO Supported the First Scrum

The primary driver for beginning the first Scrum was absolute commitment to a date, where
failure would break the company. We had to guaranteed delivery of an innovative product to the
market that would achieve rapid adoption.
Meeting with the CEO, I pointed out that he had been given plans for years that were supported
by GANTT charts. He agreed no plan had ever delivered the required functionality on time.
Many delays had been extensive and hurt the company financially. Forecasted revenue on a
major new product upgrade was millions of dollars a month so every month late cost the

company millions in revenue. We could not afford late delivery again, as the company would
operate at a loss for a quarter or more and damage to the stock price would be significant.
Further, I pointed out that in the past, he had no visibility on where the software was in the
middle of the project. He had GANTT charts and reports that looked good on paper but never
delivered the software on time. He had never seen a promised delivery date met and worse, he
rarely discovered slippage until it was too late to reforecast company revenue.
I said to my CEO that if we adopt Scrum, we set the objectives at the beginning of a Sprint. It is
the team’s responsibility to figure out how to best meet those objectives. During the Sprint, no
one can bother the team members. At the end of a Sprint, I added, we will have working code
that can be demonstrated so he could see the progress being made. You can decide to ship
anytime or do another Sprint to get more functionality. Visible working code will give you more
confidence than extensive documentation with no operational system.
We committed to a fixed date six months out and planned for six monthly Sprints. The CEO
agreed to proceed with the first software development Scrum. It took him about 60 seconds to
decide. Little did he know how much of the future of global software development rested on that
decision!
Scrum Basics
The first Scrum started with a half-day planning session that outlined the feature set we wanted
to achieve in a six-month period, and then broke it into six pieces that were achievable in 30-day
Sprints. This was the Product Backlog. For the first Sprint, the Product Backlog was
transformed into development tasks that could be done in less than one day each, the first Sprint
Backlog.
Short daily meetings were essential to drive the project with common mindshare. The three
Scrum questions were used in the first Sprint. What did you do yesterday, what will you do
today, and what is getting in your way? Daily meetings at Easel were disciplined in the way that
we now understand as the Scrum pattern [25]. This radically altered the nature of the software
development process. It allowed sharing of the state of software components so that development
tasks, thought to take days, could often be accomplished in hours using someone else’s code as a
starting point.
One of the most interesting effect of Scrum on Easel’s development environment was an
observed “punctuated equilibrium” effect. This occurs in biological evolution when a species is
stable for long periods of time and then undergoes a sudden jump in capability. During the long
period of apparent stability, many internal changes in the organism are reconfigured that cannot
be observed externally. When all pieces are in place to allow a significant jump in functionality,
external change occurs suddenly. A fully integrated component design environment leads to
unexpected, rapid evolution of a software system with emergent, adaptive properties resembling
the process of punctuated equilibrium observed in biological species. Sudden leaps in
functionality resulted in earlier than expected delivery of software in the first Scrum [36].

This aspect of self-organization, the creators of Scrum now understand as a type of Set Based
Engineering that is practiced at Toyota. Different developers were working on many components
of the system and trying to evolve them as fast as possible. Decision on how to implement a task
from a Sprint Backlog was delayed until the last possible moment. The most evolved component
for the task was selected to absorb the new functionality.
By having every member of the team see every day what every other team member was doing,
we began to get comments from one developer that if he changed a few lines of code, he could
eliminate days of work for another developer. This effect was so dramatic that the project
accelerated to the point at which it had to be slowed down by outnumbering developers with
documentation and testing engineers. This hyperproductive state was seen in a many subsequent
Scrums, although never as dramatic as the first one at Easel. It was a combination of (1) the skill
of the team, (2) the flexibility of a Smalltalk development environment, and (3) the way we
approached production prototypes that rapidly evolved into a deliverable product.
For example, a key to entering a hyperproductive state was not just the Scrum organizational
pattern. We did constant component testing of topic areas, integration of packages, refactoring of
selected parts of the system, and multiple builds per day. These activities have become key
features of eXtreme Programming [37].
Adding the Set Based Engineering practice is now viewed as the “secret sauce” that
turbocharged the process. So we are still learning lessons from the first Scrum. The magic
happens when the Scrum process combines with good engineering practices and a sophisticated
approach to product evolution. Set Based Engineering caused punctuated equilibrium and
uncontrollably fast proliferation of functionality.
We held a demo every Friday during the first Scrum and brought development experts from
other companies in to look at the product. As a result, our developers had to do demos to peers in
other companies. This was one of the most powerful accelerators I have seen in software
development. The outside experts would say, "That sucks, look at Borland's Product X to see
how it should be done." Or "How could you possible have a dumb bug like that?"
The next week, everything would be fixed! The developers refused to be embarrassed again in
front of their peers. The total transparency encouraged by Scrum was extended outside the
company and MIT and Route 128 lead engineers self-organized to throw the Scrum into
overdrive. This was very challenging to the Scrum team. Every week they felt they were not
good enough and were depressed. I kept reminding them that to be world class, we had to
repeatedly face defeat and triumph over it. We now understand from the President of Toyota that
this repeated failure, along with inspecting and adapting, is a fundamental practice that allows
persons and teams to move to a higher level of practice.
At the end of each month, the CEO got his demo. He could use the software himself and see it
evolving. We then gave the software to the consulting group to use in prototyping consulting
projects. This gave us an incredible amount of feedback to incorporate into the Product Backlog,
a list of features that are desirable to have in the software. At the beginning of each Sprint, the
Product Backlog is reprioritized before transformation into development tasks. The Scrum

adaptability to change allowed the CEO and the Product Owner to steer product development
more effectively than other project management techniques.
Scrum Results
The CEO saw significant, step-by-step progress in each increment and agreed the software was
ready to ship in the fifth increment. It had more functionality than expected in some areas and
less in others. The sixth increment was primarily a packaging increment. We shipped on the day
the product was scheduled to be shipped.
We gave a money-back guarantee that this new software would double developer productivity in
the first month of use. It sold well until the Smalltalk market started to hit the wall in the mid-
1990s, and became a model for Rational Rose development. The first ScrumMaster, John
Scumniotales went on to lead the Rational Rose development team a few years later.
Everyone agreed that (1) Scrum could meet a deadline, (2) more functionality was achieved than
expected, and (3) there would never be a return to a waterfall-type mentality because (1)
waterfall could not predict, (2) it could not deliver on time, (3) it produced less functionality per
developer unit of time, and (4) user satisfaction was terrible when the product was delivered,
since waterfall approaches did not lend themselves to customer involvement or alteration of
specifications required by rapidly changing market conditions.
Over the last decade, Scrum has emerged from humble beginnings to a movement involving tens
of thousands of projects in hundreds of the leading software development companies worldwide.
The process model used in the first Scrum 1993 is essentially the same as taught in Certified
ScrumMaster courses in 2007.

The First Scrum: How Scrum provides energy, focus,
clarity, and transparency to project teams developing
complex systems
Interview with Dr. Jeff Sutherland, CTO of PatientKeeper, Inc.
EWork-Out Business Processes, May, 2006
“…The primary driver for beginning the first Scrum was absolute commitment to a date, where
failure would break the company. The task: guaranteed delivery of an innovative product to the
market that would achieve rapid adoption.
“In a meeting with the CEO, I noted that for years he had received project plans that were
supported by Gantt charts. The CEO agreed that no plan had ever delivered the required
functionality on time. Many delays had been extensive and hurt the company financially.
“Forcasted revenue on a major new product upgrade was millions of dollars a month, so every
month that a project was late cost the company millions in revenue. As the company would
operated at a loss for a quarter or more and damage to the stock price would be significant, we
could not afford to repeat this cycle.
“Further, I pointed out that the CEO had no view of the status of the software by the middle of
the project. He had Gantt charts and reports that looked solid on paper but failed to deliver the
software on time. He had never seen a promised delivery date met, and worse, he rarely
discovered slippage until it was too late to reforecast company revenue.
“I told the CEO that in adopting Scrum, we set the objectives at the beginning of what Scrum
refers to as a sprint. It is the teams responsibility to determine how to best meet those
objectives. During the sprint, no one can bother team members with requests. At the end of a
sprint, I added, working code that will be demonstrated, so you can see the progress made. You
can decide to ship anytime or do another Sprint to get more functionality. Visible working code
provides more confidence than extensive documentation with no operational system.
“In the case of this project, the date was six months out, and we established six sprints. The CEO
agreed to proceed with the first software development Scrum.
“The first Scrum started with a half day planning session that outlined the feature set we wanted
to achieve in a six month period. We then broke it into six pieces which were achievable in 30
day sprints. This was the product backlog. For the first sprint, the product backlog was
transformed into development tasks that could be done in less than a day.
“Daily meetings allowed everyone on the project team to see the status of all aspects of the
project in real time. This allowed the collective neural networks of the team's mind to fine-tune
or redirect efforts on a daily basis to maximize throughput. The result was radical alteration of
the software development process by allowing sharing of software resources. Development tasks
thought to take days could often be accomplished in hours using someone else's code as a
starting point.

…
“The meetings were kept short, typically under 30 minutes and discussion was restricted to the
three SCRUM questions:
What did you do yesterday?

What will you do today?
What obstacles got in your way?
“By having every member of the team see every day what every other team member was doing,
we could make progress by identifying work that could be improved by others. We received
comments from one developer, for example, that if he changed a few lines of code, he could
eliminate days of work for another developer. This effect was so dramatic that the project
accelerated to the point at which it had to be slowed down by outnumbering developers with
documentation and testing engineers. This hyperproductive state was seen in several subsequent
Scrums, although never as dramatically as the first at Easel.
…
“At the end of each month, the CEO got a demo. He could use the software himself and see it
work. We then gave the software to the consulting group to use in prototyping consulting
projects. This provided an incredible amount of feedback to incorporate into the Scrum product
backlog: a list of desirable features to include in the software. At the beginning of each sprint,
product backlog is reprioritized before transformation into development tasks. The Scrum
adaptability to change enabled the CEO to steer product development more effectively than other
project management techniques.
“The CEO saw significant, step by step progress in each increment and he agreed that the
product was ready to ship in the fifth increment. It had more functionality than expected in some
areas and less in others. We shipped on the day it was scheduled to be shipped.
“Everyone agreed that first, Scrum could meet a deadline; second, more functionality was
achieved than expected; and third, there would never be a return to [the old] mentality
“Over the past decade, Scrum has emerged from humble beginnings to a movement involving
tens of thousands of projects in hundreds of the leading software development companies
worldwide. Properly implemented, Scrum represents best business practice in some of the
world's leading corporations.
“It allows teams to operate close to the 'edge of chaos' to foster rapid system evolution,
enforcing a simple set of rules for self-organization of software teams to produce systems
with evolving architectures, aligning individual and organization objectives, creating a
culture driven by performance, supporting shareholder value creation, achieving stable
and consistent communication of performance at all levels, and enhancing individual
development and quality of life.”

Chapter 3: What is Scrum? The First Papers on the Scrum
Development Process
Note: Best practices in Scrum have evolved over the years and may not reflected in this early
paper. For example, at PatientKeeper, there is no closure phase to a project. All user
acceptance testing, documentation and training is done as part of a Sprint and software goes
into production at multiple large enterprises. This means the Sprint demo in advanced Scrum
implementations is a live system with thousands of users at the end of every Sprint.
Many of the core principles taught in the Certified ScrumMaster training course are the ones
created by the first Scrum team at Easel Corporation in 1993. Sprints, monthly iterations, daily
meetings with three questions, impediments, and backlogs. The first Scrum also implemented all
the eXtreme Programming engineering practices in some form two years before Kent Beck
codified XP.
From 1995-2000, Jeff Sutherland chaired a series of workshops at the annual OOPSLA
Conference on Business Object Design and Implementation [38]. Ken Schwaber agreed to write
the first paper on Scrum for the OOPSLA’95 Workshop [39]. He observed the first Scrum in
action and laid out the fundamental principles of operation. Over a decade later this paper is still
one of the most widely read papers from OOPSLA Conferences. It continues to get more hits
than any other paper at http://jeffsutherland.com/Scrum.
The concepts of complexity theory are introduced along with the important distinction between
empirical and predictive processes. Business enterprises are complex adaptive systems and the
software that runs them is rapidly becoming equally complex. Software systems evolve over time
like biological systems. Concepts from artificial life [6] are informative as systems that are
flexible evolve faster as flexibility increases up to the boundary of chaos. Using empirical
process control to prevent chaotic behavior was shown to be the essence of Scrum in this paper.
Mike Beedle [25] led the Scrum process through elaboration as an organizational pattern
supported by Ken Schwaber, Jeff Sutherland, and others (see Appendix I). Further information
on the essence of Scrum can be found in Agile Development with Scrum by Ken Schwaber and
Mike Beedle [17] with contributions from Jeff Sutherland that can be found in a paper later in
this volume – Agile Can Scale: Inventing and Reinventing Scrum in Five Companies [40].
Over the last 13 years, Jeff Sutherland has used five companies as research laboratories for
Scrum (Easel, VMARK, Individual, IDX, PatientKeeper) and Ken Schwaber has worked as a
consultant in the last three of these companies where they have tested and refined Scrum
together. Of continued interest is why a few Scrum teams enter the hyperproductive state like the
first Scrum. This is clearly affected by the stages of maturity of Scrum implementations [41]
along with the structure of deployment of Scrum teams [29].

Scrum Development Process
Ken Schwaber
Advanced Development Methods, Inc., 1995
_______________________________________________________________________
Abstract
The stated, accepted philosophy for systems development is that the development process is a
well understood approach that can be planned, estimated, and successfully completed. This has
proven incorrect in practice. SCRUM assumes that the systems development process is an
unpredictable, complicated process that can only be roughly described as an overall
progression. SCRUM defines the systems development process as a loose set of activities that
combines known, workable tools and techniques with the best that a development team can
devise to build systems. Since these activities are loose, controls to manage the process and
inherent risk are used. SCRUM is an enhancement of the commonly used iterative/incremental
object-oriented development cycle.
KEY WORDS: Scrum SEI Capability-Maturity-Model Process Empirical

________________________________________________________________________
1. Introduction
In this paper we introduce a development process, Scrum, that treats major portions of systems
development as a controlled black box. We relate this to complexity theory to show why this
approach increases flexibility and ability to deal with complexity, and produces a system that is
responsive to both initial and additionally occurring requirements.
Numerous approaches to improving the systems development process have been tried. Each has
been touted as providing “significant productivity improvements.” All have failed to produce
dramatic improvements [42]. As Grady Booch noted, “We often call this condition the software
crisis, but frankly, a malady that has carried on this long must be called normal.”[43]
Concepts from industrial process control are applied to the field of systems development in this
paper. Industrial process control defines processes as either “theoretical” (fully defined) or
“empirical” (black box). When a black box process is treated as a fully defined process,
unpredictable results occur [44]. A further treatment of this is provided in Appendix 1.
A significant number of systems development processes are not completely defined, but are
treated as though they are. Unpredictability without control results. The Scrum approach treats
these systems development processes as a controlled black box.
Variants of the Scrum approach for new product development with high performance small
teams was first observed by Takeuchi and Nonaka [1]3 at Fuji-Xerox, Canon, Honda, NEC,
Epson, Brother, 3M, Xerox, and Hewlett-Packard. A similar approach applied to software

development at Borland was observed by Coplien [2] to be the highest productivity C++
development project ever documented. More recently, a refined approach to the SCRUM process
has been applied by Sutherland [45] to Smalltalk development and Schwaber [46] to Delphi
development.
The Scrum approach is used at leading edge software companies with significant success. We
believe Scrum may be appropriate for other software development organizations to realize the
expected benefits from Object Oriented techniques and tools [47].
2. Overview
Our new approach to systems development is based on both defined and black box process
management. We call the approach the Scrum methodology (see Takeuchi and Nonaka, 1986),
after the Scrum in Rugby -- a tight formation of forwards who bind together in specific positions
when a Scrumdown is called.
As will be discussed later, Scrum is an enhancement of the iterative and incremental approach to
delivering object-oriented software initially documented by Pittman [48] and later expanded
upon by Booch [49]. It may use the same roles for project staff as outlined by Graham [50], for
example, but it organizes and manages the team process in a new way.
Scrum is a management, enhancement, and maintenance methodology for an existing system or

production prototype. It assumes existing design and code which is virtually always the case in
object-oriented development due to the presence of class libraries. Scrum will address totally
new or re-engineered legacy systems development efforts at a later date.
Software product releases are planned based on the following variables:
• Customer requirements - how the current system needs enhancing.

• Time pressure - what time frame is required to gain a competitive advantage.
• Competition - what is the competition up to, and what is required to best them.
• Quality - what is the required quality, given the above variables.
• Vision - what changes are required at this stage to fulfill the system vision.
• Resource - what staff and funding are available.
These variables form the initial plan for a software enhancement project. However, these
variables also change during the project. A successful development methodology must take these
variables and their evolutionary nature into account.
3. Current Development Situation
Systems are developed in a highly complicated environment. The complexity is both within the
development environment and the target environment. For example, when the air traffic control
system development was initiated, three-tier client server systems and airline deregulation did
not have to be considered. Yet, these environmental and technical changes occurred during the
project and had to be taken into account within the system being built.

Environmental variables include:
• Availability of skilled professionals - the newer the technology, tools, methods, and
domain, the smaller the pool of skilled professionals.
• Stability of implementation technology - the newer the technology, the lower the stability
and the greater the need to balance the technology with other technologies and manual
procedures.
• Stability and power of tools - the newer and more powerful the development tool, the
smaller the pool of skilled professionals and the more unstable the tool functionality.
• Effectiveness of methods - what modeling, testing, version control, and design methods
are going to be used, and how effective, efficient, and proven are they.
• Domain expertise - are skilled professionals available in the various domains, including
business and technology.
• New features - what entirely new features are going to be added, and to what degree will
these fit with current functionality.
• Methodology - does the overall approach to developing systems and using the selected
methods promote flexibility, or is this a rigid, detailed approach that restricts flexibility.
• Competition - what will the competition do during the project? What new functionality
will be announced or released.
• Time/Funding - how much time is available initially and as the project progresses? How
much development funding is available.
• Other variables - any other factors that must be responded to during the project to ensure
the success of the resulting, delivered system, such as reorganizations.
The overall complexity is a function of these variables:
complexity = f(development environment variables + target environment variables)
where these variables may and do change during the course of the project.
As the complexity of the project increases, the greater the need for controls, particularly the
ongoing assessment and response to risk.
Attempts to model this development process have encountered the following problems:

• Many of the development processes are uncontrolled. The inputs and outputs are either
unknown or loosely defined, the transformation process lacks necessary precision, and
quality control is not defined. Testing processes are an example.
• An unknown number of development processes that bridge known but uncontrolled

processes are unidentified. Detailed processes to ensure that a logical model contains
adequate content to lead to a successful physical model are one such process.
• Environmental input (requirements) can only be taken into consideration at the beginning
of the process. Complex change management procedures are required thereafter.
Attempts to impose a micro, or detailed, methodology model on the development process have
not worked because the development process is still not completely defined. Acting as though
the development process is defined and predictable, results in being unprepared for the
unpredictable results.
Although the development process is incompletely defined and dynamic, numerous

organizations have developed detailed development methodologies that include current
development methods (structured, OO, etc.). The Waterfall methodology was one of the first
such defined system development processes. A picture of the Waterfall methodology is shown
in Figure 1.
Planning
Analysis Design Development Implement
Figure 1 : Waterfall Methodology
Although the waterfall approach mandates the use of undefined processes, its linear nature has
been its largest problem. The process does not define how to respond to unexpected output from
any of the intermediate process.
Barry Boehm introduced a Spiral methodology to address this problem [51]. Each of the
waterfall phases is ended with a risk assessment and prototyping activity. The Spiral
methodology is shown in Figure 2.
The Spiral methodology “peels the onion,” progressing through “layers” of the development
process. A prototype lets users determine if the project is on track, should be sent back to prior
phases, or should be ended. However, the phases and phase processes are still linear.

Requirements work is still performed in the requirements phase, design work in the design phase,
and so forth, with each of the phases consisting of linear, explicitly defined processes.
Determine objectives,
Evaluate alternatives;
alternatives,
Identify, resolve risks
constraints
Prototype
Concept
Plan next phases
Requirements Develop next level product
Design
Implement
Figure 2 : Spiral Methodology
The Iterative methodology improves on the Spiral methodology. Each iteration consists of all of
the standard Waterfall phases, but each iteration only addresses one set of parsed functionality.
The overall project deliverable has been partitioned into prioritized subsystems, each with clean
interfaces. Using this approach, one can test the feasibility of a subsystem and technology in the
initial iterations. Further iterations can add resources to the project while ramping up the speed
of delivery. This approach improves cost control, ensures delivery of systems (albeit
subsystems), and improves overall flexibility. However, the Iterative approach still expects that
the underlying development processes are defined and linear. See Figure 3.

Detail Design Coding
Preliminary
Module Test
Design
Requirements
System Test
Analysis
Figure 3 : Iterative Methodology
Given the complex environment and the increased reliance on new "state-of-the-art" systems, the
risk endured by system development projects has increased and the search for mechanisms to
handle this risk has intensified.
One can argue that current methodologies are better than nothing. Each improves on the other.
The Spiral and Iterative approaches implant formal risk control mechanisms for dealing with
unpredictable results. A framework for development is provided.
However, each rests on the fallacy that the development processes are defined, predictable
processes. But unpredictable results occur throughout the projects. The rigor implied in the
development processes stifles the flexibility needed to cope with the unpredictable results and
respond to a complex environment.
Despite their widespread presence in the development community, our experience in the industry
shows that people do not use the methodologies except as a macro process map, or for their
detailed method descriptions.
The following graph demonstrates the current development environment, using any of the
Waterfall, Spiral or Iterative processes. As the complexity of the variables increase even to a
moderate level, the probability of a “successful” project quickly diminishes (a successful project
is defined as a system that is useful when delivered). See Figure 4.

0.9
Inflexible response to
unpredictability (internal & external)
causes sharp drop in p(Success)
probability(Success) 0.5 as complexity increases
0.1
Low Medium High
Complexity
Figure 4: Defined Process Risk/Complexity Graph
4. Scrum Methodology
The system development process is complicated and complex. Therefore maximum flexibility
and appropriate control is required. Evolution favors those that operate with maximum exposure
to environmental change and have maximized flexibility. Evolution deselects those who have
insulated themselves from environmental change and have minimized chaos and complexity in
their environment.
An approach is needed that enables development teams to operate adaptively within a complex
environment using imprecise processes. Complex system development occurs under rapidly
changing circumstances. Producing orderly systems under chaotic circumstances requires
maximum flexibility. The closer the development team operates to the edge of chaos, while still
maintaining order, the more competitive and useful the resulting system will be. Langton has
modeled this effect in computer simulations [6] and his work has provided this as a fundamental
theorem in complexity theory.
Methodology may well be the most important factor in determining the probability of success.
Methodologies that encourage and support flexibility have a high degree of tolerance for changes
in other variables. With these methodologies, the development process is regarded as
unpredictable at the onset, and control mechanisms are put in place to manage the
unpredictability.

If we graph the relationship between environmental complexity and probability of success with a
flexible methodology that incorporates controls and risk management, the tolerance for change is
more durable. See Figure 5.
Figure 5: Risk/Complexity Comparison Graph
Figures 4 and 5 reflect software development experiences at ADM, Easel, VMARK, Borland and
virtually every other developer of “packaged” software. These organizations have embraced risk
and environmental complexity during development projects. Increased product impact,
successful projects, and productivity gains were experienced. The best possible software is built.
Waterfall and Spiral methodologies set the context and deliverable definition at the start of a
project. Scrum and Iterative methodologies initially plan the context and broad deliverable
definition, and then evolve the deliverable during the project based on the environment. Scrum
acknowledges that the underlying development processes are incompletely defined and uses
control mechanisms to improve flexibility.
The primary difference between the defined (waterfall, spiral and iterative) and empirical
(Scrum) approach is that the Scrum approach assumes that the analysis, design, and development
processes in the Sprint phase are unpredictable. A control mechanism is used to manage the
unpredictability and control the risk. Flexibility, responsiveness, and reliability are the results.
See Figure 6.

Figure 6 : Scrum Methodology
Characteristics of Scrum methodology are :
• The first and last phases (Planning and Closure) consist of defined processes, where all
processes, inputs and outputs are well defined. The knowledge of how to do these
processes is explicit. The flow is linear, with some iterations in the planning phase.
• The Sprint phase is an empirical process. Many of the processes in the sprint phase are
unidentified or uncontrolled. It is treated as a black box that requires external controls.
Accordingly, controls, including risk management, are put on each iteration of the Sprint
phase to avoid chaos while maximizing flexibility.
• Sprints are nonlinear and flexible. Where available, explicit process knowledge is used;
otherwise tacit knowledge and trial and error is used to build process knowledge. Sprints
are used to evolve the final product.
• The project is open to the environment until the Closure phase. The deliverable can be
changed at any time during the Planning and Sprint phases of the project. The project
remains open to environmental complexity, including competitive, time, quality, and
financial pressures, throughout these phases.
• The deliverable is determined during the project based on the environment.
The table in Figure 7 compares the primary Scrum characteristics to those of other
methodologies.

Waterfall Spiral Iterative Scrum
Defined Required Required Required Planning &
processes Closure only
Final product Determined Determined Set during Set during project
during planning during planning project
Project cost Determined Partially variable Set during Set during project
during planning project
Completion date Determined Partially variable Set during Set during project
during planning project
Responsiveness Planning only Planning At end of Throughout
to environment primarily each
iteration
Team flexibility, Limited - Limited - Limited - Unlimited during
creativity cookbook cookbook cookbook iterations
approach approach approach
Knowledge Training prior to Training prior to Training Teamwork
transfer project project prior to during project
project
Probability of Low Medium low Medium High
success
Figure 7 : Methodology Comparison
4.1 Scrum Phases
Scrum has the following groups of phases:
4.1.1. Pregame
• Planning : Definition of a new release based on currently known backlog, along with an
estimate of its schedule and cost. If a new system is being developed, this phase consists
of both conceptualization and analysis. If an existing system is being enhanced, this
phase consists of limited analysis.
• Architecture : Design how the backlog items will be implemented. This phase includes
system architecture modification and high level design.
4.1.2. Game
• Development Sprints : Development of new release functionality, with constant respect to

the variables of time, requirements, quality, cost, and competition. Interaction with these
variables defines the end of this phase. There are multiple, iterative development sprints,
or cycles, that are used to evolve the system.
4.1.3. Postgame

• Closure : Preparation for release, including final documentation, pre-release staged
testing, and release.
Scrum Methodology
Pregame
• Planning
• System Architecture/High Level
Design
Game
• Sprints (Concurrent Engineering)
• Develop
(Analysis,Design,Develop)
• Wrap
• Review
• Adjust
Postgame
• Closure
Figure 8: Scrum Methodology
4.2 Phase Steps
Each of the phases has the following steps:
4.2.1. Planning
• Development of a comprehensive backlog list.

• Definition of the delivery date and functionality of one or more releases.
• Selection of the release most appropriate for immediate development.
• Mapping of product packets (objects) for backlog items in the selected release.
• Definition of project team(s) for the building of the new release.
• Assessment of risk and appropriate risk controls.
• Review and possible adjustment of backlog items and packets.
• Validation or reselection of development tools and infrastructure.
• Estimation of release cost, including development, collateral material, marketing,
training, and rollout.
• Verification of management approval and funding.

4.2.2. Architecture/High Level Design
• Review assigned backlog items.

• Identify changes necessary to implement backlog items.
• Perform domain analysis to the extent required to build, enhance, or update the domain
models to reflect the new system context and requirements.
• Refine the system architecture to support the new context and requirements.
• Identify any problems or issues in developing or implementing the changes
• Design review meeting, each team presenting approach and changes to implement each
backlog item. Reassign changes as required.
4.2.3. Development (Sprint)
The Development phase is an iterative cycle of development work. The management determines
that time, competition, quality, or functionality are met, iterations are completed and the closure
phase occurs. This approach is also known as Concurrent Engineering. Development consists of
the following macro processes:
• Meeting with teams to review release plans.

• Distribution, review and adjustment of the standards with which the product will
conform.
• Iterative Sprints, until the product is deemed ready for distribution.
A Sprint is a set of development activities conducted over a pre-defined period, usually one to
four weeks. The interval is based on product complexity, risk assessment, and degree of
oversight desired. Sprint speed and intensity are driven by the selected duration of the Sprint.
Risk is assessed continuously and adequate risk controls and responses are put in place. Each
Sprint consists of one or more teams performing the following:
• Develop: Defining changes needed for the implementation of backlog requirements into
packets, opening the packets, performing domain analysis, designing, developing,
implementing, testing, and documenting the changes. Development consists of the micro
process of discovery, invention, and implementation.
• Wrap: Closing the packets, creating a executable version of changes and how they
implement backlog requirements.
• Review: All teams meeting to present work and review progress, raising and resolving
issues and problems, adding new backlog items. Risk is reviewed and appropriate
responses defined.
• Adjust: Consolidating the information gathered from the review meeting into affected
packets, including different look and feel and new properties.
Each Sprint is followed by a review, whose characteristics are :
• The whole team and product management are present and participate.
• The review can include customers, sales, marketing and others.

• Review covers functional, executable systems that encompass the objects assigned to that
team and include the changes made to implement the backlog items.
• The way backlog items are implemented by changes may be changed based on the
review.
• New backlog items may be introduced and assigned to teams as part of the review,
changing the content and direction of deliverables.
• The time of the next review is determined based on progress and complexity. The Sprints
usually have a duration of 1 to 4 weeks.
4.2.4. Closure
When the management team feels that the variables of time, competition, requirements, cost, and
quality concur for a new release to occur, they declare the release “closed” and enter this phase.
The closure phase prepares the developed product for general release. Integration, system test,
user documentation, training material preparation, and marketing material preparation are among
closure tasks.
4.3. Scrum Controls
Operating at the edge of chaos (unpredictability and complexity) requires management controls
to avoid falling into chaos. The Scrum methodology embodies these general, loose controls,
using OO techniques for the actual construction of deliverables.
Risk is the primary control. Risk assessment leads to changes in other controls and responses by
the team.
Controls in the Scrum methodology are :
• Backlog: Product functionality requirements that are not adequately addressed by the
current product release. Bugs, defects, customer requested enhancements, competitive
product functionality, competitive edge functionality, and technology upgrades are
backlog items.
• Release/Enhancement: backlog items that at a point in time represent a viable release

based on the variables of requirements, time, quality, and competition.
• Packets: Product components or objects that must be changed to implement a backlog

item into a new release.
• Changes: Changes that must occur to a packet to implement a backlog item.
• Problems: Technical problems that occur and must be solved to implement a change.
• Risks: risks that affect the success of the project are continuously assessed and responses
planned. Other controls are affected as a result of risk assessment.

• Solutions: solutions to the problems and risks, often resulting in changes.
• Issues: Overall project and project issues that are not defined in terms of packets, changes
and problems.
These controls are used in the various phases of Scrum. Management uses these controls to
manage backlog. Teams use these controls to manage changes, problems. Both management
and teams jointly manage issues, risks, and solutions. These controls are reviewed, modified, and
reconciled at every Sprint review meeting.
4.4 Scrum Deliverables
The delivered product is flexible. Its content is determined by environment variables, including
time, competition, cost, or functionality. The deliverable determinants are market intelligence,
customer contact, and the skill of developers. Frequent adjustments to deliverable content occur
during the project in response to environment. The deliverable can be determined anytime during
the project.
4.5 Scrum Project Team
The team that works on the new release includes full time developers and external parties who
will be affected by the new release, such as marketing, sales, and customers. In traditional
release processes, these latter groups are kept away from development teams for fear of over-
complicating the process and providing “unnecessary” interference. The Scrum approach,
however, welcomes and facilitates their controlled involvement at set intervals, as this increases
the probability that release content and timing will be appropriate, useful, and marketable.
The following teams are formed for each new release:
Management: Led by the Product Manager, it defines initial content and timing of the release,
then manages their evolution as the project progresses and variables change. Management deals
with backlog, risk, and release content.
Development teams: Development teams are small, with each containing developers,
documenters and quality control staff. One or more teams of between three and six people each
are used. Each is assigned a set of packets (or objects), including all backlog items related to
each packet. The team defines changes required to implement the backlog item in the packets,
and manages all problems regarding the changes. Teams can be either functionally derived
(assigned those packets that address specific sets of product functionality) or system derived
(assigned unique layers of the system). The members of each team are selected based on their
knowledge and expertise regarding sets of packets, or domain expertise.
4.6 Scrum Characteristics
The Scrum methodology is a metaphor for the game of Rugby. Rugby evolved from English
football (soccer) under the intense pressure of the game:

Rugby student William Webb Ellis, 17, inaugurates a new game whose rules will
be codified in 1839. Playing soccer for the 256-year-old college in East
Warwickshire, Ellis sees that the clock is running out with his team behind so he
scoops up the ball and runs with it in defiance of the rules.
The People's Chronology, Henry Holt and Company, Inc. Copyright © 1992.
Scrum projects have the following characteristics:
• Flexible deliverable - the content of the deliverable is dictated by the environment.
• Flexible schedule - the deliverable may be required sooner or later than initially planned.
• Small teams - each team has no more than 6 members. There may be multiple teams
within a project.
• Frequent reviews - team progress is reviewed as frequently as environmental complexity

and risk dictates (usually 1 to 4 week cycles). A functional executable must be prepared
by each team for each review.
• Collaboration - intra and inter-collaboration is expected during the project.
• Object Oriented - each team will address a set of related objects, with clear interfaces and
behavior.
The Scrum methodology shares many characteristics with the sport of Rugby:
• The context is set by playing field (environment) and Rugby rules (controls).
• The primary cycle is moving the ball forward.
• Rugby evolved from breaking soccer rules - adapting to the environment.
The game does not end until environment dictates (business need, competition, functionality,
timetable).
5. Advantages of the Scrum Methodology
Traditional development methodologies are designed only to respond to the unpredictability of

the external and development environments at the start of an enhancement cycle. Such newer
approaches as the Boehm spiral methodology and its variants are still limited in their ability to
respond to changing requirements once the project has started.

The Scrum methodology, on the other hand, is designed to be quite flexible throughout. It
provides control mechanisms for planning a product release and then managing variables as the
project progresses. This enables organizations to change the project and deliverables at any point
in time, delivering the most appropriate release.
The Scrum methodology frees developers to devise the most ingenious solutions throughout the
project, as learning occurs and the environment changes.
Small, collaborative teams of developers are able to share tacit knowledge about development
processes. An excellent training environment for all parties is provided.
Object Oriented technology provides the basis for the Scrum methodology. Objects, or product
features, offer a discrete and manageable environment. Procedural code, with its many and
intertwined interfaces, is inappropriate for the Scrum methodology. Scrum may be selectively
applied to procedural systems with clean interfaces and strong data orientation.
6. Scrum Project Estimating
Scrum projects can be estimated using standard function point estimating. However, it is
advisable to estimate productivity at approximately twice the current metric. The estimate is
only for starting purposes, however, since the overall timetable and cost are determined
dynamically in response to the environmental factors.
Our observations have led us to conclude that Scrum projects have both velocity and
acceleration. In terms of functions delivered, or backlog items completed:
• initial velocity and acceleration are low as infrastructure is built/modified

• as base functionality is put into objects, acceleration increases
• acceleration decreases and velocity remains sustainably high
Further development in metrics for empirical processes is required.
7. System Development Methodologies : Defined or Empirical
System development is the act of creating a logical construct that is implemented as logic and
data on computers. The logical construct consists of inputs, processes, and outputs, both macro
(whole construct) and micro (intermediate steps within whole construct). The whole is known as
an implemented system.
Many artifacts are created while building the system. Artifacts may be used to guide thinking,
check completeness, and create an audit trail. The artifacts consist of documents, models,
programs, test cases, and other deliverables created prior to creating the implemented system.
When available, a metamodel defines the semantic content of model artifacts. Notation describes
the graphing and documentation conventions that are used to build the models.

The approach used to develop a system is known as a method. A method describes the activities
involved in defining, building, and implementing a system; a method is a framework. Since a
method is a logical process for constructing systems (process), it is known as a metaprocess (a
process for modeling processes).
A method has micro and macro components. The macro components define the overall flow
and time-sequenced framework for performing work. The micro components include general
design rules, patterns and rules of thumb. General design rules state properties to achieve or to
avoid in the design or general approaches to take while building a system. Patterns are solutions
that can be applied to a type of development activity; they are solutions waiting for problems that
occur during an activity in a method. Rules of thumb consist of a general body of hints and tips.
Figure 9: Relationship between the Method, the Artifacts, and the System
Applying concepts from industrial process control to the field of systems development, methods
can be categorized as either “theoretical” (fully defined) or “empirical” (black box).
Correctly categorizing systems development methods is critical. The appropriate structure of a

method for building a particular type of system depends on whether the method is theoretical or
empirical.
Models of theoretical processes are derived from first principles, using material and energy
balances and fundamental laws to determine the model. For a systems development method to
be categorized as theoretical, it must conform to this definition.

Models of empirical processes are derived categorizing observed inputs and outputs, and
defining controls that cause them to occur within prescribed bounds. Empirical process
modeling involves constructing a process model strictly from experimentally obtained
input/output data, with no recourse to any laws concerning the fundamental nature and properties
of the system. No a priori knowledge about the process is necessary (although it can be helpful);
a system is treated like a black box.
Primary characteristics of both theoretical and empirical modeling are detailed in Figure 10.
Theoretical Modeling Empirical Modeling

1. Usually involves fewer measurements; Requires extensive measurements, because
requires experimentation only for the it relies entirely on experimentation for the
estimation of unknown model parameters. model development.
2. Provides information about the internal Provides information only about that
state of the process. portion of the process which can be
influenced by control action.
3. Promotes fundamental understanding of Treats the process like a “black box.”
the internal workings of the process.
4. Requires fairly accurate and complete Requires no such detailed knowledge; only
process knowledge. that output data be obtainable in response
to input changes.
5. Not particularly useful for poorly Quite often proves to be the only
understood and/or complex processes. alternative for modeling the behavior of
poorly understood and/or complex
processes.
6. Naturally produces both linear and Requires special methods to produce
nonlinear process models. nonlinear models.
Figure 10: Theoretical vs. empirical modeling
Upon inspection, we assert that the systems development process is empirical:
• Applicable first principles are not present

• The process is only beginning to be understood
• The process is complex
• The process is changing
Most methodologists agree with this assertion; “...you can’t expect a method to tell you
everything to do. Writing software is a creative process, like painting or writing or architecture...
... (a method) supplies a framework that tells how to go about it and identifies the places where
creativity is needed. But you still have to supply the creativity....”[52]
Categorizing the systems development methods as empirical is critical to the effective

management of the systems development process.

If systems development methods are categorized as empirical, measurements and controls are
required because it is understood that the inner workings of the method are so loosely defined
that they cannot be counted on to operate predictably.
In the past, methods have been provided and applied as though they were theoretical. As a
consequence, measurements were not relied upon and controls dependent upon the
measurements weren’t used.
Many of the problems in developing systems have occurred because of this incorrect
categorization. When a black box process is treated as a fully defined process, unpredictable
results occur. Also, the controls are not in place to measure and respond to the unpredictability.

Chapter 4: Scrum on Large Projects: Distributed,
Outsourced Scrum
First Scrum Scaling at IDX Systems 1996-2000
In 1996, Jeff Sutherland become SVP of Product Development at IDX Systems Corporation with
a development team that grew to almost 600 people by 2000. Scrum was introduced immediately
for all developers. A new organizational structure was created to coordinate Scrum teams called
a Scrum of Scrums. Previous management positions were eliminated and all managers became
team leaders. Typically, Directors of Engineering became Scrum of Scrum leaders. Ken
Schwaber provided consulting experience in various parts of the development organization
during 1996-2000 to help introduce Scrum to the organization.
Business units with product portfolios typically had less than 100 developers. A Scrum of
Scrums was able to manage this size group effectively. At this level a lead architect became the
Product Owner of the architecture backlog for the business unit and was the single person
responsible for the architecture for that unit.
At a SVP level, there was a team of Directors and VPs that met periodically to coordinate
activities across business units. The SVP, having s systems architecture background, led a virtual
team of all the business unit architects to develop a global architecture backlog for all business
units. Sprint planning involved the entire business unit once a month and the business unit
architect was responsible for getting a commitment from each Scrum team to devote 10% of the
resources in every Scrum team to addressing the global architecture backlog. This drove all
business units incrementally towards a common architectural framework.
A similar virtual team strategy was used for software integration across business units and for
common quality assurance processes and procedures. The virtual team strategy allowed all senior
developers to be working with a Scrum team during every Sprint. The approach was designed to
avoid separate specialized teams and to get everyone into front line production.
The SVP team worked well to coordinate Scrum of Scrums teams across business units.
However, it was not optimal for driving a global Product Backlog across business units. Today,
best practices are to implement a MetaScrum above the Scrum of Scrums teams. The MetaScrum
is lead by the Chief Product Owner and incorporates all stakeholders in the company. Product
release strategy and the state of every Sprint is reviewed at MetaScrum meetings. All decisions
to start, stop, or change Sprints are made there. Often, the CEO is the ScrumMaster for the
MetaScrum.
Linear Scalability in Large, Distributed, and Outsourced Scrums

Two case studies published in 2007 demonstrate for the first time that a software development
process can scale linearly across both development team size and geographies. The
SirsiDynix/StarSoft Development Labs project delivered over a million lines of code with teams
distributed across the U.S., Canada, and Russia [18]. When the development team doubled in

size by bringing on engineers in St. Petersburg, Russia, the velocity of software delivery more
than doubled. A similar effect was noted by a CMMI Level 5 Scrum implementation at
Systematic Software Engineering in Denmark [19]. Introducing Scrum increased productivity is
small teams only slightly as they were already Agile. On larger projects, they consistently
achieved the same level of productivity per developer as on small teams. After achieving at 80%
reduction in planning costs, a 50% reduction in total project costs, along with significantly
increased user and employee satisfaction, they converted their company-wide CMMI Level 5
process documentation, training, and implementation to Scrum.
There are many moving parts in an enterprise wide implementation of Scrum. While a clear,
consistent model is achievable for any company, it must be localized into a specific company
structure through inspection and adaptation, the hallmark of Scrum. This immediately leads to
questions on how to best organize the company to take advantage of Scrum. It is strongly
recommended to use expertise from an experienced Scrum Trainer who has led multiple
enterprise Scrum implementations to work through best strategies for implementing enterprise
Scrum in a specific company.

Agile Can Scale: Inventing and Reinventing Scrum in Five
Companies
Introduction
Agile development is focused on delivering maximum business value in the shortest possible
time. It is well known that over half the requirements on a typical software project change during
development and about half the features in delivered software are never used by customers.
Adaptive planning and self-organizing teams are used to embrace changing requirements and
avoid building features with low return on investment. The result is faster delivery, better
quality, higher user satisfaction, and a more creative and enjoyable working environment for
developers.
Scrum derives from Japanese best practices in lean manufacturing and lean product development
[1, 13]. The goal of Scrum is to achieve the “Toyota Effect”, deliver four times as much software
with twelve times the quality within a series of short time boxes called “Sprint's,” which last 30
days or less. Scrum is characterized by 15 minute, intensive, daily meetings of a
multidisciplinary software delivery team, usually including product marketing, testers, software
analysts, designers, and coders, and even deployment support staff. Prior to each iteration a list
of features to be delivered called the Product Backlog is reprioritized so that maximum value
features are developed first. Most tasks required to complete Product Backlog features are
defined and estimated at the start of the Sprint by the multidisciplinary development team to
product the Sprint Backlog. Some tasks require further design or definition during a Sprint so the
Sprint Backlog will dynamically evolve, yet a skilled team will estimate the amount of work
required to complete a Sprint with less than 20% error using improve estimating techniques
based on the latest research on planning [31].
A lean production environment is based on a “pull” system [8] where team members pull
inventory into a work environment “just in time.” A the start of a Sprint the team pulls the
amount of Product Backlog into the Sprint that they can commit to complete during the Sprint.
Loading of team members during the Sprint is dynamic, in the sense that team members chose
their own tasks and “pull” from the Sprint Backlog when they are ready to start another task.
Good teams “minimize the work in progress” by keeping as few tasks open as possible to avoid
integration problems and missed dates at the end of a Sprint.
Within a Sprint, problems and challenges are evaluated in the daily Scrum meeting and the team
self organizes in real time to maximize the number of production ready features delivered by the
team at the end of each Sprint. The daily meetings focused on answers to three questions by each
member of the team – What did I do yesterday? What will I do today? What blocks, problems, or
impediments are getting in my way? A well functioning team will dynamically recalculate the
plan daily through a brief discussion and provide enough information for a ScrumMaster, the
team leader, to calculate a “Burndown Chart” of work to be completed. Team efforts to

accelerate or decelerate the downward velocity of the burndown graph allow a team to “fly” the
project into a fixed delivery date.
A typical Burndown Chart is illustrated below. It consists of the cumulative time it takes to
complete outstanding tasks for deliverable software for a Scrum sprint. Each developer breaks
down tasks into small pieces and enters into a backlog system work remaining on each active
task every day. This could be as simple as task updates on a white board or an Excel
spreadsheet, or could be a specialized tool for support of distributed teams. The remaining work
for each task is added up to generate the cumulative backlog. Best current methods require only
one minute of each developers time each day to update two data items for active tasks (percent
complete and time remaining) allowing an automated system to produce the Burndown Chart,
showing how fast outstanding backlog is decreasing each day. In the daily Scrum meetings, the
team members use information sharing to determine what actions taken that day will maximize
the download movement of the cumulative backlog. Experience has shown that Scrum project
planning will consistently produce a faster path to the end goal than any other form of project
planning reported to date, with less administrative overhead than any previously reported
approach.
Burndown Chart
50
45
40
35
Days Remaining
30
Desired Trend
25
Days Remaining
20
15
10
0
1 2 3 4 5 6 7 8 9 10
Scrum Meeting (2 week Sprint)
Figure 1: Burndown chart
Details of the Scrum approach have been carefully documented elsewhere [22]. Scrum is the
only Agile methodology that has been formalized and published as an organizational pattern for
software development [25]. The process assumes that requirements will change during the period
between initial specification and delivery of a product. It supports Humphrey’s Requirements
Uncertainty Principle [53], which states that for a new software system, the requirements will not
be completely know until after the users have used it. Scrum allows for Ziv’s Uncertainty
Principle in software engineering, which observes that uncertainty is inherent and inevitable in
software development processes and products [54]. And it accounts for Wegner’s mathematical
proof (lemma) that it is not possible to completely specify an interactive system [55]. Most

software systems built today are object-oriented implementations, and most of those object-
oriented systems depend on environmental inputs to determine process outputs (i.e., they are
interactive systems).
Traditional, heavyweight software methodologies assume that requirements can be specified in

advance, that they will not change during development, that the users know what they want
before they see it, and that software development is a predictable, repeatable process. These
assumptions are fundamentally flawed and inconsistent with the mathematical lemmas and
principles cited above. As a result, 31% of software projects, usually driven by a variant of the
waterfall methodology, are terminated before completion [56].
This article serves as a short retrospective on the origins of Scrum, its evolution in five
companies, and a few key learnings along the way. It will provide a reference point for further
investigation and implementation of Scrum for those interested in using a proven, scalable,
lightweight development process that supports the principles of the Agile Alliance as outlined in
the “Manifesto for Agile Software Development” (see www.agilealliance.org).
Easel Corporation: The First Scrum

Scrum was started in 1993 for software teams at Easel Corporation, where I was VP of object
technology. In the initial Scrum, we built the first object-oriented design and analysis tool that
incorporated round-trip engineering. The second Scrum implemented the first product to
completely automate object-relational mapping in enterprise development environment. I was
assisted by two world-class developers--Jeff McKenna, now a Scrum and eXtreme Programming
consultant, and John's Scumniotales, now a development leader for object-oriented design tools
at Rational Corporation.
In 1995, Easel was acquired by VMARK. Scrum continued there until I joined Individual in
1996 as VP of engineering to develop Personal NewsPage (now www.office.com). I asked Ken
Schwaber, CEO of Advanced Development Methodologies, to help me incorporate Scrum into
Individual’s development process. On the same year, I took Scrum to IDX Systems when I
assumed the position of senior VP of engineering and product development and CTO. IDX, one
of the largest US healthcare software companies, was the proving ground for multiple team
Scrum implementations. At one point, almost 600 developers were working on dozens of
products. In 2000, Scrum was introduced to PatientKeeper, a mobile/wireless healthcare
platform where I became CTO. So I have experienced Scrum in five companies that varied
widely in size. They were proving grounds for Scrum in all phases of company growth from
startup, to initial IPO, to mid-sized, and then to a large company delivering enterprise systems to
the marketplace.
“All-at-Once” Software Development

There were some key factors that influenced the introduction of Scrum at Easel Corporation.
The book Wicked Problems, Righteous Solutions [32] by Peter DeGrace and Leslie Hulet Stahl
reviewed the reasons why the waterfall approach to software development does not work for
software development today. Requirements are not fully understood before the project begins.

The users know what they want only after they see an initial version of the software.
Requirements change during the software construction process. And new tools and technologies
make implementation strategies unpredictable. DeGrace and Stahl reviewed “All-at-Once”
models of software development, which uniquely fit object-oriented implementation of software
and help to resolve these challenges.
“All-at-Once” models of software development assume that the creation of software is done by
simultaneously working on requirements, analysis, design, coding, and testing and then
delivering the entire system all at once. The simplest All-at-Once model is a single super-
programmer creating and delivering an application from beginning to end. All aspects of the
development process reside in a single person’s head. This is the fastest way to deliver a product
that has good internal architectural consistency, and it is the “hacker” mode of implementation.
The next level of approach to All-at-Once development is handcuffing two programmers
together, as in the XP practice of pair programming [23]. Two developers deliver the entire
system together. This is been shown to deliver better code (in terms of usability, maintainability,
flexibility, and extendability) faster than work delivered by larger teams. The challenge is to
achieve a similar productivity effect in the large with an entire team and then with teams of
teams.
Our team based All-at-Once model was based on both the Japanese approach to new product
development, Sashimi, and Scrum. We were already using production prototyping to build
software. It was implemented in slices (Sashimi) where an entire piece of fully integrated
functionality worked at the end of an iteration. What intrigued us was Hirotaka Takeuchi and
Hujiro Nonaka’s description of the team-building process in setting up and managing a Scrum
[1]. The idea of building a self-empowered team in which everyone had a global view of the
product on a daily basis seemed like the right idea. This approach to managing the team, which
had been so successful at Honda, Canon, and Fujitsu, resonated with the systems thinking
approach being promoted by Peter Senge at MIT [15].
We were also impacted by recent publications in computer science. As I alluded above, Peter
Wagner at Brown University demonstrated that it was impossible to fully specify or test an
interactive system, which is designed to respond to external inputs (Wegner's lemma) [55]. Here
was mathematical proof that any process that assumed known inputs, as does the waterfall
method, was doomed to failure when building an object-oriented system.
We were prodded into setting up the first Scrum meeting after reading James Coplien's paper on
Borland's development of Quattro Pro for Windows [2]. The Quattro team delivered one million
lines of C++ code in 31 months, with a four person staff growing to eight people later in the
project. This was about a thousand lines of deliverable code per person per week, probably the
most productive project ever documented. The team attained this level of productivity by
intensive interaction in daily meetings with project management, product management,
developers, documenters, and quality assurance staff.
Software Evolution and Punctuated Equilibrium

Our daily meetings at Easel were disciplined in a way that we now understand as the Scrum
pattern [25]. The most interesting effect of Scrum on Easel's development environment was an
observed "punctuated equilibrium effect." A fully integrated component design environment
leads to rapid evolution of a software system with emergent, adaptive properties, resembling the
process of punctuated equilibrium observed in biological species.
Evolutionary biologists have noticed that change occurs sharply at intervals separated by long
periods of apparent stagnation, leading to the concept of punctuated equilibrium [6]. Computer
simulations of this phenomenon suggest that periods of equilibrium are actually periods of
ongoing genetic change of an organism. The effects of that change are not apparent until several
subsystems evolve in parallel to the point where they can work together to produce a dramatic
external effect [10]. This punctuated equilibrium effect has been observed by teams working in
a component-based environment with adequate business process engineering tools, and the
Scrum development process accentuates the effect.
By having every member of the team see every day what every other team member was doing,
we began to see how we could accelerate each other's work. For instance, one developer
commented that if he changed a few lines in code, he could eliminate days of work for another
developer. This effect was so dramatic that the project accelerated the point where it had to be
slowed down. This hyperproductive state was seen in several subsequent Scrum's, but never
went so dramatic as the one at Easel.
Stimulating Software Evolution with SyncSteps
The first Scrum worked from a unique view of the software system. A project domain can be
viewed as a set of packages that will form a release. Packages are what the user perceives as
pieces of functionality, and they evolve out of work on topic areas (see Figure 2). Topic areas
are business object components. Changes are introduced into the system by introducing a unit of
work that alters a component. Refactoring often causes a single change to ripple throughout the
system. This unit of work in the initial Scrum was called a SynchStep.
Draft 10/14/2007 ©Jeff Sutherland 1993-2007 - not for distribution

83
Figure 2: View of a software systems as seen by the first Scrum
System evolution proceeds in SyncSteps (see Figure 3). After one or more SyncSteps have gone
to completion and forced some refactoring throughout the system, a new package of
functionality emerges that is observable to the user. These SyncSteps are similar to genetic
mutations. Typically, several interrelated components must mutate in concert to produce a
significant new piece of functionality. This new functionality appears as the punctuated
equilibrium effect to builders in the system. For a period of time, the system is stable with no
new behavior. Then when a certain (somewhat unpredictable) SyncStep completes, the whole
system pops up to a new level of functionality, often surprising the development team.

84
Figure 3: Firing a SyncStep causes a ripple through the system that triggers emission of a
package of functionality visible to the user.
This aspect of self-organization is now understood as a type of Set-Based Concurrent

Engineering (SBCE) which is practiced at Toyota [57]. Developers consider sets of possible
solutions and gradually narrow the set of possibilities to converge on a final solution. Here,
decisions on how and where to implement a feature in a set of components was delayed until the
last possible moment. The most evolved component is selected “just in time” to absorb new
functionality, resulting in minimal coding and a more elegant architecture. Thus emergent
architecture, a core principle in all Agile processes, is not random evolution. Properly
implemented, it is an SBCE technique viewed as a best business practice in some of the world’s
leading corporations.
Achieving a Sustainable HyperProductive State
The key to entering a hyperproductive state was not just the Scrum organizational pattern. It
was a combination of (1) the skill of the team, (2) the flexibility of a Smalltalk development
environment, (3) the implementation of what are now know as XP engineering practices, and (4)
the way we systematically stimulated production prototypes that rapidly evolved into a
deliverable product.
Furthermore, in the hyperproductive state, the initial Scrum entered what professional athletes
and martial artists call "the zone." No matter what happened or what problems arose, the
response of the team always was far better than the response of any individual. It was
reminiscent of the Celtics basketball team at their peak, when they could do no wrong. The
impact of entering the zone was not just hyperproductivity. Peoples personal lives were
changed. Team members said they would never forget working on the project, and they would
always be looking for another experience like it. It induced open, team oriented, fun-loving

85
behavior in unexpected persons. Those individuals who could not function well in an open,
hyperproductive environment self-selected themselves out of the team by finding other jobs.
This reinforced positive team behavior similar to biological systems, which select for fitness to
the environment, resulting in improved performance of individual organisms.
VMARK: The First Senior Management Scrum

When Easel Corporation was acquired by VMARK (subsequenctly Informix, now Ascension
Software) the original Scrum team continued its work on the same product. The VMARK senior
management team was intrigued by Scrum and asked me to run a weekly senior management
team Scrum to drive all the company’s products to the Internet. These meetings started in 1995,
and within a few months, the team had caused the introduction of two new Internet products and
repositioned current products as Internet applications. Some members of this team left VMARK
to become innovators in emerging Internet companies, so Scrum had an early impact on the
Internet.
It was also at VMARK that Ken Schwaber was introduced to Scrum. Ken and I had worked
together on and off for years. I showed him Scrum and he agreed it worked better that other
project management approaches and was similar to how he built project management software in
his company. He quickly sold off the project management software business and worked on
bringing Scrum to the software industry at large. His work has had an incredible effect on
deploying Scrum worldwide.
Individual: The First Internet Scrum

In the spring of 1996, I returned to Individual, Inc., a company I co-founded as VP of
Engineering in 1988. Much of the Scrum experience at Individual has been documented by Ken
Schwaber [17]. The most impressive thing to me about Scrum at Individual was not that the
team delivered two new Internet products – and multiple releases of one of the products – in a
single quarter. It was the fact that Scrum eliminated several hours a day of senior management
meeting time starting the day that Scrum began, within a week of my arrival at the company.
Because Individual had just gone public at the beginning of the Internet explosion, there were
multiple competing priorities and constant revision of market strategy. As a result, the
development team was constantly changing priorities and unable to deliver product. The
management team was meeting daily to determine status of priorities that were viewed
differently by every manager. These meetings were eliminated and the Scrum meetings became
the focus for all decisionmaking.
It was incredibly productive to force all decisions to occur in the daily Scrum meeting. If anyone
wanted to know the status of specific project deliverables or wanted to influence any priority, he
or she could only do it in the daily Scrum meeting. I remember the senior VP of marketing sat in
on every meeting for a couple of weeks sharing her desperate concern about meeting Internet
deliverables and timetables. The effect on the team was not to immediately respond to her
despair. Over a period of two weeks, the team self-organized around a plan to meet her priorities
with achievable technical delivery dates. When she agreed to the plan, she no longer had to

86
attend any Scrum meetings. The Scrum reported status on the Web with green lights, yellow
lights, and red lights for all pieces of functionality. In this way, the entire company knew status
in real time, all the time. This transparency of information has become a key characteristic of
Scum.
IDX Systems: The First Scrum in the Large

During the summer of 1996, IDX Systems hired me as senior VP of engineering and product
development. IDX had over 4,000 customers and was one of the largest US healthcare software
companies, with hundreds of developers working on dozens of products. Here was an
opportunity to extend Scrum to large-scale development.
The approach at IDX was to turn the entire development group into an interlocking set of
Scrums. Every part of the organization was team based including the management team, which
included two vice presidents, a senior architect, and several directors. Front-line Scrums met
daily. A Scrum of Scrums, which included the team leaders of each Scrum in a product line, met
weekly, The management Scrum met monthly.
The key learning at IDX was that Scrum scales to any size. With dozens of teams in operation,
the most difficult problem was ensuring the quality of the Scrum process in each team,
particularly when the entire organization had to learn Scrum all at once. IDX was large enough
to bring in productivity experts to monitor throughput on every project. While most teams were
only able to double the industry average in function points per month delivered, several teams
moved into a hyperproductive state, producing deliverable functionality at four to five times the
industry average. These teams became shining stars in the organization and examples for the rest
of the organization to follow.
One of the most productive teams at IDX was the Web Framerwork team that built a web
frontend infrastructure for all products. The infrastructure was designed to host all IDX
applications, as well as seamlessly interoperate with end user or third party applications. It was
a distributed team with developers in Boston, Seattle, and Vermont who met by teleconference
in a daily Scrum meeting. The geographic transparency of this model produced the same high
performance as co-located teams and has become the signature of hyperproductive
distributed/outsourced Scrums [29].
PatientKeeper Scrum: The First Scrum Company

In early 2000, I joined PatientKeeper, Inc. as chief technology officer and began introducing
Scrum into a startup company. I was the 21st employee, and we grew the development team from
a dozen people to 45 people in six months. PatientKeeper deploys mobile devices in healthcare
institutions to capture and process financial and clinical data. Server technology synchronizes
the mobile devices and moves data to and from multiple back-end legacy systems. A robust
technical architecture provides enterprise application integration to hospital and clinical systems.
Data is forward-deployed from these systems in a PatientKeeper clinical repository. Server
technologies migrate changes from our clinical repository to a cache and then to data storage on

87
the mobile device. PatientKeeper proves that Scrum works equally well across technology
implementations.
The key learning at PatientKeeper has involved the introduction of eXtreme Programming
techniques as a way to implement code delivered by a Scrum organization. While all teams seem
to find it easy to implement a Scrum organizational process, they do not always find it easy to
introduce XP. We were able to do some team programming and constant testing and refactoring,
particularly as we migrated all development to Java and XML. It was more difficult to introduce
these ideas when developers were working in C and C++. After a year of Scrum meetings in all
areas of development, processes matured enough to capitalize on Scrum project management
techniques, which were fully automated.
Complete automation and transparency of data allowed PatientKeeper to multithread Sprints

through multiple teams. That in combination with implementing a MetaScrum of senior
stakeholders in the company allowed PatientKeeper to run from top to bottom as a Scrum and
become the first Scrum company to enter the hyperproductive state, delivering over 45
production releases a year of a large enterprise software platform. This became the prototype for
the All-at-Once, or Type C Scrum, implemented in at least five companies by 2006 [41].
Conclusions
After introducing Scrum into five different companies of different sizes and with different
technologies, I can confidently say that Scrum works in any environment and can scale into
programming in the large. In all cases, it will radically improve communication and delivery of
working code. The next challenge for Scrum, in my view, is to provide a tight integration of the
Scrum organizational pattern with XP programming techniques, combined with innovative
approaches to distributed teams and stimulation of rapid software system evolution. I believe
these strategies can generate a hyperproductive Scrum on a predictable basis. The first Scrum
did this intuitively and that was its key to extreme performance and a life-changing experience.
In addition, the participation of Scrum leaders in the Agile Alliance [58], a group which has
absorbed all leaders of well-known Agile development processes, will facilitated wider use of
Scrum and its adoption as an enterprise standard development process.

88
Distributed Scrum: Agile Project Management with
Outsourced Development Teams
Jeff Sutherland, Ph.D. Anton Viktorov Jack Blount Nikolai Puntikov

Patientkeeper StarSoft Dev. Labs SirsiDynix StarSoft Dev. Labs
Newton, MA, US St. Petersburg, Russia Provo, UT, USA St. Petersburg, Russia
[email protected] [email protected] [email protected] [email protected]
Abstract
Agile project management with Scrum derives from best business practices in companies like
Fuji-Xerox, Honda, Canon, and Toyota. Toyota routinely achieves four times the productivity
and 12 times the quality of competitors. Can Scrum do the same for globally distributed teams?
Two Agile companies, SirsiDynix and StarSoft Development Laboratories achieved comparable
performance developing a Java application with over 1,000,000 lines of code. During 2005, a
distributed team of 56 Scrum developers working from Provo, Utah; Waterloo, Canada; and St.
Petersburg, Russia, delivered 671,688 lines of production Java code. At 15.3 function points per
developer/month, this is the most productive Java project ever documented. SirsiDynix best
practices are similar to those observed on distributed Scrum teams at IDX Systems, radically
different than those promoted by PMBOK, and counterintuitive to practices advocated by the
Scrum Alliance. This paper analyzes and recommends best practices for globally distributed
Agile teams.
Introduction
Scrum is an Agile software development process designed to add energy, focus, clarity, and
transparency to project teams developing software systems. It leverages artificial life research [6]
by allowing teams to operate close to the edge of chaos to foster rapid system evolution. It
capitalizes on robot subsumption architectures [5] by enforcing a simple set of rules that allows
rapid self-organization of software teams to produce systems with evolving architectures. A
properly implemented Scrum was designed to increase speed of development, align individual
and organization objectives, create a culture driven by performance, support shareholder value
creation, achieve stable and consistent communication of performance at all levels, and enhance
individual development and quality of life.
Scrum for software development teams began at Easel Corporation in 1993 and was used to
build the first object-oriented design and analysis (OOAD) tool that incorporated round-trip
engineering. In a Smalltalk development environment, code was auto-generated from a graphic
design tool and changes to the code from the Smalltalk integrated development environment
(IDE) were immediately reflected back into design.

89
A development process was needed to support enterprise teams where visualization of design
immediately generated working code. This led to an extensive review of the computer science
literature and dialogue with leaders of hundreds of software development projects. Key factors
that influenced the introduction of Scrum at Easel Corporation were fundamental problems
inherent in software development:
• Uncertainty is inherent and inevitable in software development processes and products -

Ziv’s Uncertainty Principle [54]
• For a new software system the requirements will not be completely known until after the
users have used it - Humphrey’s Requirements Uncertainty Principle [59]
• It is not possible to completely specify an interactive system – Wegner’s Lemma [55]
• Ambiguous and changing requirements, combined with evolving tools and technologies
make implementation strategies unpredictable.
• “All-at-Once” models of software development uniquely fit object-oriented
implementation of software and help resolve these challenges. They assume the creation
of software involves simultaneously work on requirements, analysis, design, coding, and
testing, then delivering the entire system all at once [32].
“All-at-Once” Development Models

The simplest “All-at-Once” model is a single super-programmer creating and delivering an
application from beginning to end. This can be the fastest way to deliver a product that has good
internal architectural consistency and is the “hacker” model of implementation. For example, in a
“skunk works” project prior to the first Scrum, a single individual surrounded by a support team
spent two years writing every line of code for the Matisse object database [60] used to drive
$10B nuclear reprocessing plants worldwide. At less than 50,000 lines of code, the nuclear
engineers said it was the fastest and most reliable database ever benchmarked for nuclear plants.
IBM documented a variant of this approach called the Surgical Team and considered it the most
productive approach to software development [33]. The Surgical Team concept has a fatal flaw
in that there are at most one or two individuals even in a large company that can execute this
model. For example, it took three years for a competent team of developers to understand the
conceptual elegance of the Matisse object server well enough to maintain it. The single-
programmer model does not scale well to large projects.
The next level of “All-at-Once” development is handcuffing two programmers together. Pair
programming, an eXtreme Programming practice [23], is an implementation of this. Here, two
developers working at the same terminal deliver a component of the system together. This has
been shown to deliver better code (usability, maintainability, flexibility, extendibility) faster than
two developers working individually [34]. The challenge is to achieve a similar productivity
effect with more than two people.
Scrum, a scalable, team-based “All-at-Once” model, was motivated by the Japanese approach to
team-based new product development combined with simple rules to enhance team self-
organization (see Brooks’ subsumption architecture [5]). At Easel, the development team was
already using an iterative and incremental approach to building software [35]. Features were
implemented in slices where an entire piece of fully integrated functionality worked at the end of
90
an iteration. What intrigued us was Takeuchi and Nonaka’s description of the team-building
process for setting up and managing a Scrum [1]. The idea of building a self-empowered team in
which a daily global view of the product caused the team to self-organize seemed like the right
idea. This approach to managing the team, which had been so successful at Honda, Canon, and
Fujitsu, also resonated with research on systems thinking by Professor Senge at MIT [15].
Hyperproductivity in Scrum
Scrum was designed to allow average developers to self-organize into high performance teams.
The first Scrum achieved a hyperproductive state in 1993-1994 because of three primary factors.
The first was the Scrum process itself, characterized by 15 minute daily meetings where each
person answers three questions – what did you accomplish yesterday, what will you do today,
and what impediments are getting in your way? This is now part of the standard Scrum
organizational pattern [25]. Second, the team implemented all XP engineering processes [23]
including pair programming, continuous builds, and aggressive refactoring. And third, the team
systematically stimulated rapid evolution of the software system.
One of the most interesting complexity phenomena observed in the first Scrum was a
“punctuated equilibrium” effect [36]. This phenomenon occurs in biological evolution when a
species is stable for long periods of time and then undergoes a sudden jump in capability. Danny
Hillis simulated this effect on an early super-computer, the Connection Machine [61].
“The artificial organisms in Hillis’s particular world evolved not by steady progress of hill
climbing but by the sudden leaps of punctuated equilibrium… With artificial organisms Hillis
had the power to examine and analyze the genotype as easily as the realized phenotypes… While
the population seemed to be resting during the periods of equilibrium … the underlying genetic
makeup was actively evolving. The sudden increase in fitness was no more an instant occurrence
than the appearance of a newborn indicates something springing out of nothing; the population
seemed to be gestating its next jump. Specifically, the gene pool of the population contained a set
of epistatic genes that could not be expressed unless all were present; otherwise the alleles for
these genes would be recessive.” [62]
Using Scrum with a fully integrated component design environment leads to unexpected, rapid
evolution of a software system with emergent, adaptive properties resembling the process of
punctuated equilibrium. Sudden leaps in functionality resulted in earlier than expected delivery
of software in the first Scrum. Development tasks, originally planned to take days, could often be
accomplished in hours using someone else’s code as a starting point.
This aspect of self-organization is now understood as a type of Set-Based Concurrent

Engineering (SBCE) practiced at Toyota [57]. Developers consider sets of possible solutions
and gradually narrow the set of possibilities to converge on a final solution. Decisions on how
and where to implement a feature is delayed until the last possible moment. The most evolved
component is selected “just in time” to absorb new functionality, resulting in minimal coding and
a more elegant architecture. Thus emergent architecture, a core principle in all Agile processes,
is not random evolution. Properly implemented, it is an SBCE technique viewed as a best
business practice in some of the world’s leading corporations.
91
Distributed, Outsourced Scrum
Scrum was designed to achieve a hyperproductive state where productivity increases by an order
of magnitude over industry averages. Many small, collocated teams have achieved this effect.
The question for this paper is whether a large, distributed, outsourced team can achieve the
hyperproductive state.
U.S., European, or Japanese companies often outsource software development to Eastern Europe,
Russia, or the Far East. Typically, remote teams operate independently and communication
problems limit productivity. While there is a large amount of published research on project
management, distributed development, and outsourcing strategies as isolated domains, there are
few detailed studies of best project management practices on large systems that are both
distributed and outsourced.
Current recommended Scrum practice is for local Scrum teams at all sites to synchronize once a
day via a Scrum of Scrums meeting. Here we describe something rarely seen. At SirsiDynix, all
Scrum teams consist of developers from multiple sites. While some Agile companies have
created geographically transparency on a small scale, SirsiDynix uses fully integrated Scrum
teams with over 50 developers in the U.S., Canada, and Russia. This strategy helped build a new
implementation of platform and system architecture for a complex Integrated Library System
(ILS). The ILS system is similar to a vertical market ERP system with a public portal interface
used by more than 200 million people.
Best practices for distributed Scrum seen on this project consist of (1) daily Scrum team
meetings of all developers from multiple sites, (2) daily meetings of Product Owner team (3)
hourly automated builds from one central repository, (4) no distinction between developers at
different sites on the same team, (5) and seamless integration of XP practices like pair
programming with Scrum. While similar practices have been implemented on small distributed
Scrum teams [40] this is the first documented project that demonstrates Scrum hyperproductivity
for large distributed/outsourced teams building complex enterprise systems.
Distributed Team Models
Here we consider three distributed Scrum models commonly observed in practice:
• Isolated Scrums - Teams are isolated across geographies. In most cases off-shore teams
are not cross-functional and may not be using the Scrum process.
• Distributed Scrum of Scrums – Scrum teams are isolated across geographies and
integrated by a Scrum of Scrums that meets regularly across geographies.
• Totally Integrated Scrums – Scrum teams are cross-functional with members distributed
across geographies. In the SirsiDynix case, the Scrum of Scrums was localized with all
ScrumMasters in Utah.
Most outsourced development efforts use a degenerative form of the Isolated Scrums model
where outsourced teams are not cross-functional and not Agile. Requirements may be created in

92
the U.S. and developed in Dubai, or development may occur in Germany and quality assurance
in India. Typically, cross-cultural communication problems are compounded by differences in
work style in the primary organization vs. the outsourced group. In the worst case, outsourced
teams are not using Scrum and their productivity is typical of waterfall projects further delayed
by cross-continent communications lag time. Implementations of Scrum in a data rich CMMI
Level 5 company simultaneously running waterfall, incremental, and iterative projects, showed
productivity of Scrum teams was at least double that of waterfall teams, even with CMMI Level
5 reporting overhead [63]. Outsourced teams not using Scrum will typically achieve less than
half the velocity of the primary site using Scrum.
Isolated Scrums Teams
Distributed Scrum of Scrums
Integrated Scrums
Figure 2: Strategies for distributed Scrum teams [11].
The latest thinking in the Project Management Institute Guide to the Project Management Body
of Knowledge (PMBOK) models is a degenerative case of isolated non-Scrum teams [64]. This
is a spiral waterfall methodology which layers the Unified Modeling Language (UML) and the
Rational Unified Process (RUP) onto teams which are not cross-functional [65]. It partitions
work across teams, creates teams with silos of expertise, and incorporates a phased approach
laden with artifacts that violate the principles of lean development [12].
Best practice recommended by the Scrum Alliance is a Distributed Scrum of Scrums model. This
model partitions work across cross-functional, isolated Scrum teams while eliminating most
dependencies between teams. Scrum teams are linked by a Scrum-of-Scrums where
ScrumMasters (team leaders/project managers) meet regularly across locations. This encourages
communication, cooperation, and cross-fertilization and is appropriate for newcomers to Agile
development.
An Integrated Scrums model has all teams fully distributed and each team has members at
multiple locations. While this appears to create communication and coordination burdens, the
daily Scrum meetings help to break down cultural barriers and disparities in work styles. On
large enterprise implementations, it can organize the project into a single whole with an
integrated global code base. Proper implementation of this approach provides location
transparency and performance characteristics similar to small co-located teams. A smaller, but
similar, distributed team at IDX Systems Corporation during 1996-2000 achieved almost ten
times industry average performance [40]. The SirsiDynix model approached this level of
performance for distributed/outsourced Integrated Scrums. It appears to be the most productive

93
distributed team ever documented for a large Java enterprise system with more than one million
lines of code. This Integrated Scrums model is recommended for experienced Agile teams at
multiple locations.
SirsiDynix Case Study
SirsiDynix has approximately 4,000 library and consortia clients, serving over 200 million
people through over 20,000 library outlets in the Americas, Europe, Africa, the Middle East and
Asia-Pacific. Jack Blount, President and CEO of Dynix and now CTO of the merged SirsiDynix
company, negotiated an outsource agreement with StarSoft who staffed the project with over 20
qualified engineers in 60 days. Significant development milestones were completed in a few
weeks and joint development projects are efficiently tracked and continue to be on schedule.
StarSoft Development Labs, Inc. is a software outsourcing service provider in Russia and
Eastern Europe. Headquartered in Cambridge, Massachusetts, USA, StarSoft operates
development centers in St. Petersburg, Russia and Dnepropetrovsk, Ukraine, employing over 450
professionals. StarSoft has experience handling development efforts varying in size and duration
from just several engineers working for a few months to large-scale projects involving dozens of
developers and spanning several years. StarSoft successfully uses Agile development and
particularly XP engineering practices to maintain CMM Level 3 certification.
Hidden Costs of Outsourcing
The hidden costs of outsourcing are significant, beginning with startup costs. Barthelemy [66]
surveyed 50 companies and found that 14% of outsourcing operations were failures. In the
remainder, costs of transitioning to a new vendor often canceled out anticipated savings from
low labor costs. The average time from evaluating outsourcing to beginning of vendor
performance was 18 months for small projects. As a result, the MIT Sloan Management Review
advises readers not to outsource critical IT functions.
The German Institute for Economic Research analyzed 43,000 German manufacturing firms
from 1992-2000 and found that outsourcing services led to poor corporate performance, while
outsourcing production helped [67]. While this is a manufacturing study rather than software
development, it suggests that outsourcing core development may provide gains not seen
otherwise.

94
Figure 3 - SirsiDynix lines of new Java code in thousands from 2003-2006.
Large software projects are very high risk. The 2003 Standish Chaos Report show success rates
of only 34%. 51% of projects are over budget or lacking critical functionality. 15% are total
failures [68].
SirsiDynix sidestepped many of the hidden costs, directly outsourced primary production, and
used Integrated Scrums to control risk. The goals of increasing output per team member and
linearly increasing overall output by increasing team size were achieved. Production velocity
more than doubled when the 30 person North American development team was augmented with
26 Russians from StarSoft in December 2005.
Intent of the Integrated Scrums Model

An Agile company building a large product and facing time-to-market pressure needs to quickly
double or quadruple productivity within a constrained budget. The local talent pool is not
sufficient to expand team size and salary costs are much higher than outsourced teams. On the
other hand, outsourcing is only a solution if Agile practices are enhanced by capabilities of the
outsourced teams. The primary driver is enhanced technical capability resulting in dramatically
improved throughput of new application functionality. Cost savings are a secondary driver.
Context
Software complexity and demands for increased functionality are exponentially increasing in all
industries. When an author of this paper flew F-4 aircraft in combat in 1967, 8% of pilot
functions were supported by software. In 1982, the F16 software support was 45%, and by 2000,
the F22 augmented 80% of pilot capabilities with software [64]. Demands for ease of use,
scalability, reliability, and maintainability increase with complexity.

95
SirsiDynix was confronted with the requirement to completely re-implement a legacy library
system with over 12,500 installed sites. Large teams working over many years in a changing
business environment faced many new requirements in the middle of the project. To complicate
matters further, the library software industry was in a consolidating phase. Dynix started the
project in 2002 and merged with Sirsi in 2005 to form SirsiDynix.
Fortunately, Dynix started with a scalable Agile process that could adapt to changing
requirements throughout the project. Time to market demanded more than doubling of output.
That could only happen by augmenting resources with Agile teams. StarSoft was selected
because of their history of successful XP implementations and their experience with systems
level software.
The combination of high risk, large scale, changing market requirements, merger and acquisition
business factors, and the SirsiDynix experience with Scrum combined with StarSoft success with
XP led them to choose an Integrated Scrums implementation. Jack Blount's past experience with
Agile development projects at US Data Authority, TeleComputing and JD Edwards where he had
used Isolated Scrums and Distributed Scrum of Scrums models did not meet his expectations.
This was a key factor in his decision to structure the project as Integrated Scrums.
Forces
Complexity Drivers
The Systems and Software Consortium (SSCI) has outlined drivers, constraints, and enablers that
force organizations to invest in real-time project management information systems. Scalable
Scrum implementations with minimal tooling are one of the best real-time information
generators in the software industry.
SSCI complexity drivers are described as [64]:
• Increasing problem complexity shifting focus from requirements to objective capabilities

that must be met by larger teams and strategic partnerships.
• Increasing solution complexity which shifts attention from platform architectures to
enterprise architectures and fully integrated systems.
• Increasing technical complexity from integrating stand alone systems to integrating
across layers and stacks of communications and network architectures.
• Increasing compliance complexity shifting from proprietary to open standards.
• Increasing team complexity shifting from a single implementer to strategic teaming and
mergers and acquisitions.
SirsiDynix faced all of these issues. Legacy products were difficult to sell to new customers.
They needed a new product with complete functionality for the library enterprise based on new
technologies that were highly scalable, easily expandable, and used the latest computer and
library standards,

96
The SirsiDynix Horizon 8.0 architecture supports a wide variety of users from publication
acquisition to cataloging, searching, reserving, circulating, or integrating information from local
and external resources. The decision was made to use Java with J2EE, a modular design,
database independency, maximum use of free platforms and tools, and wide support of
MARC21, UNIMARC, Z39.50 and other ILS standards.
The project uses a three-tier architecture and Hibernate as a database abstraction layer. Oracle
10g, MS SQL, and IBM DB2 support is provided. The JBoss 4 Application server is used with a
Java GUI Client with WebStart bootstrap. It is a cross-platform product supporting MS Windows
2000, XP, 2003, Red Hat Linux, and Sun Solaris. Built-in, multi-language support has on-the-fly
resource editing for ease of localization. Other key technologies are JAAS, LDAP, SSL,
Velocity, Xdoclet, JAXB, JUnit, and Jython.
Top Issues in Distributed Development
The SSCI has carefully researched top issues in distributed development [64], all of which had to
be handled by SirsiDynix and StarSoft.
• Strategic: Difficult leveraging available resources, best practices are often deemed
proprietary, are time consuming and difficult to maintain.
• Project and process management: Difficulty synchronizing work between distributed
sites.
• Communication: Lack of effective communication mechanisms.
• Cultural: Conflicting behaviors, processes, and technologies.
• Technical: Incompatible data formats, schemas, and standards.
• Security: Ensuring electronic transmission confidentiality and privacy.
The unique way in which SirsiDynix and StarSoft implemented an Integrated Scrums model
carefully addressed all of these issues.
Solution: Integrated Scrums

There are three roles in a Scrum: the Product Owner, the ScrumMaster, and the Team.
SirsiDynix used these roles to solve the strategic distribution problem of building a high velocity,
real-time reporting organization with an open source process that is easy to implement and low-
overhead to maintain.
For large programs, a Chief ScrumMaster to run a Scrum of Scrums and a Chief Product Owner
to centrally manage a single consolidated and prioritized product backlog is essential. SirsiDynix
located the Scrum of Scrums and the Product Owner teams in Utah.
Team Formation

97
The second major challenge for large projects is process management, particularly synchronizing
work between sites. This was achieved by splitting teams across sites and fine tuning daily
Scrum meetings.
PO PO PO
SM SirsiDynix
Dev Provo, Utah
Dev
Dev
T Ld
Dev
Dev StarSoft
Dev St. Petersburg,
Catalogue Serials Circulation Search Reporting
Figure 3 Scrum teams split across sites. PO=Product Owner, SM=ScrumMaster,

TLd=Technical Lead.
Teams at SirsiDynix were split across the functional areas needed for an integrated library
system. Half of a Scrum team is typically in Provo, Utah, and the other half in St. Petersburg.
There are usually 3-5 people on the Utah part of the team and 4 or more on the St. Petersburg
portion of the team. The Search and Reporting Teams are smaller. There are smaller numbers of
team members in Seattle, Denver, St. Louis, and Waterloo, Canada.
Scrum Meetings
Teams meet across geographies at 7:45am Utah time which is 17:45 St. Petersburg time. Teams
found it necessary to distribute answers to the three Scrum questions in writing before the Scrum
meeting. This shortens the time needed for the join meeting teleconference and helps overcome
any language barriers. Each individual reports on what they did since the last meeting, what they
intend to do next, and what impediments are blocking their progress.
Email exchange on the three questions before the daily Scrum teleconference was used
throughout the project to enable phone meetings to proceed more smoothly and efficiently. These
daily team calls helped the people in Russia and the U.S. learn to understand each other. In
contrast, most outsourced development projects do not hold formal daily calls and the
communication bridge is never formed.

98
St. Petersburg, Russia 17:45pm
Local Team
Meeting
7:45am Provo, Utah
Scrum Team Meeting
Figure 4 – Scrum Team meetings
Local sub-teams have an additional standup meeting at the beginning of the day in St.
Petersburg. Everyone uses the same process and technologies and daily meetings coordinate
activities within the teams.
ScrumMasters are all in Provo, Utah or Waterloo, Canada, and meet in a Scrum of Scrums every
Monday morning. Here work is coordinated across teams. Architects are directly allocated to
production Scrum teams and all located in Utah. An Architecture group also meets on Monday
after the Scrum of Scrums meeting and controls the direction of the project architecture through
the Scrum meetings. A Product Owner resident in Utah is assigned to each Scrum team. A chief
Product Owner meets regularly with all Product Owners to assure coordination of requirements.
SirsiDynix achieved strong central control of teams across geographies by centrally locating
ScrumMasters, Product Owners, and Architects. This helped them get consistent performance
across distributed teams.
Sprints
Sprints are two weeks long on the SirsiDynix project. There is a Sprint planning meeting similar
to an XP release planning meeting in which requirements from User Stories are broken down
into development tasks. Most tasks require a lot of questions from the Product Owners and some
tasks take more time than initial estimates.
The lag time for Utah Product Owner response to questions on User Stories forces multitasking
in St. Petersburg and this is not an ideal situation. Sometimes new tasks are discovered after
querying Product Owners during the Sprint about feature details.
Code is feature complete and demoed at the end of each Sprint. Up until 2006, if it met the
Product Owner’s functional requirement, it was considered done, although full testing was not
completed. It was not deliverable code until SirsiDynix strengthened its definition of “done” to
include all testing in 2006. Allowing work in progress to cross Sprint boundaries introduces wait
times and greater risk into the project. It violates the lean principle of reducing work in progress
and increases rework.
Product Specifications

99
Requirements are in the form of User Stories used in many Scrum and XP implementations.
Some of them are lengthy and detailed, others are not. A lot of questions result after receiving
the document in St. Petersburg which are resolved by in daily Scrum meetings, instant
messaging, or email.
Story for Simple Renewals Use Case:
Patron brings book or other item to staff to be renewed.
Patron John Smith checked out "The Da Vinci Code" the last time he was in the library. Today
he is back in the library to pick up something else and brings "The Da Vinci Code" with him. He
hands it to the staff user and asks for it to be renewed. The staff user simply scans the item
barcode at checkout, and the system treats it as a renewal since the item is already checked out to
John. This changes the loan period (extends the due date) for the length of the renewal loan. Item
and patron circulation history are updated with a new row showing the renewal date and new due
date. Counts display for the number of renewals used and remaining. The item is returned to
Patron John Smith.
Assumptions:
Item being renewed is currently checked out to the active patron
• No requests or reservations outstanding
• Item was not overdue
• Item does not have a problem status (lost, etc)
• No renew maximums have been reached
• No block/circulation maximums have been reached
• Patron's subscriptions are active and not within renewal period
• No renewal charges apply
• No recalls apply
Renewal is from Check Out (not Check In)
Staff User has renewal privileges
Verification (How to verify completion):
• Launch Check Out
• Retrieve a patron who has an item already checked out but not yet overdue
• Enter barcode for checked out item into barcode entry area (as if it is being checked out),
and press <cr>.
• System calculates new due date according to circulation rules and agency parameters.
• The renewal count is incremented (Staff renewal with item)
• If user views "Circulation Item Details", the appropriate Renewals information should be
updated (renewals used/remaining)
• Cursor focus returns to barcode entry area, ready to receive next scan (if previous
barcode is still displayed, it should be automatically replaced by whatever is entered next)
• A check of the item and patron circulation statistics screens show a new row for the
renewal with the renewal date/time and the new due date.

100
For this project, St. Petersburg staff likes a detailed description because the system is a
comprehensive and complex system designed for specialized librarians. As a result, there is a lot
of knowledge that needs to be embedded in the product specification.
The ways libraries work in St. Petersburg are very different than English libraries. Russian
libraries operate largely via manual operations. While processes look similar to English libraries
on the surface, the underlying details are quite different. Therefore, user stories do not have
sufficient detail for Russian programmers.
Testing
Developers write unit tests. The Test team and Product Owners do manual testing. An
Automation Test team in Utah creates scripts for an automated testing tool. Stress testing is as
needed.
During the Sprint, the Product Owner tests features that are in the Sprint backlog. Up until 2006,
testers received a stable Sprint build only after the Sprint demo. The reason for this was a lower
tester/developer ratio than recommended by the Scrum Alliance.
There are 30 team members in North America and 26 team members in St. Petersburg on this
project. The St. Petersburg team has one project leader, 3 technical team leaders, 18 developers,
1 test lead, and 3 testers. This low tester/developer ratio initially made it impossible to have a
fully tested package of code at the end of the Sprints.
The test-first approach was initially encouraged and not mandated. Tests were written
simultaneously with code most of the time. GUIs were not unit tested.
Test
Component Cases Tested
Acquisitions 529 384
Binding 802 646
Cataloging 3101 1115
Circulation 3570 1089
Common 0 0
ERM 0 0
Pac Searching 1056 167
Serials 2735 1714
Sub Total 11793 5115
Figure 5 – Test Cases Created vs. Tested (coded and working)
In the summer of 2006, a new CTO of SirsiDynix, Talin Bingham, took over the project and
introduced Test Driven Design. Every Sprint starts with the usual Sprint Planning meeting and
teams are responsible for writing functional tests before doing any coding. Once functional tests
are written and reviewed, coding starts. Test-first coding is mandated. When coding is complete,
developers run unit tests and manually pass all the functional tests before checking in changes to
the repository.

101
Functional Area Reserve Book Room
Task Description Check that items from Item List is placed under
Reserve with “Inactive” status
Condition 1. User has right for placing Items under Reserve
2. At least one Item List exists in the system
3. Default Reserve Item Status in Session
Defaults is set to ”Inactive”
Entry Point Launcher is opened
Test Data No specific data
Action 1. Reserve > Reserve Item
2. Select “Item Search” icon
3. Select “Item List” in the Combo box list of
search options and enter appropriate Item
list name
4. Press Enter
5. Select all Items which appear in the Item
Search combo box and press “OK”
Expected Results 1. Items that were in Item list should appear

in the list in Reserve Item
2. Status of all items that has been just added
should be shown as “Inactive”
3. Save button should be inactive
4. All corresponding Item should retain their
original parameters
Figure 6 – Functional Test Example
Automation testing is done using the Compuware TestPartner tool, but there is still room for
improvement of test coverage.
Configuration Management
SirsiDynix was using CVS as source code repository when the decision was made to engage an
outsourcing firm. At that time, SirsiDynix made a decision that CVS could not be used
effectively because of lack of support for distributed development, largely seen in long code
synchronization times. Other tools were evaluated and Perforce was chosen as the best solution.
StarSoft had seen positive results on many projects using Perforce. It is fast, reliable and offers
local proxy servers for distributed teams. Although not a cheap solution, it has been very
effective for the SirsiDynix project.
Automated builds run every hour with email generated back to developers. It takes 12 minutes to
do a build, 30 minutes if the database changes. StarSoft would like to see faster builds and true
concurrent engineering. Right now builds are only stable every two weeks at Sprint boundaries.
Pair Programming, Refactoring, and other XP practices

102
StarSoft is an XP company and tries to introduce XP practices into all their projects. Pair
programming is done on more complicated pieces of functionality. Refactoring was planned for
future Sprints and not done in every iteration as in XP. Some radical refactoring without loss of
functionality occurred as the project approached completion. Continuous integration is
implemented as hourly builds. On this project, these three engineering practices were used with
Scrum as the primary project management methodology.
Measuring Progress
The project uses the Jira <http://www.atlassian.com> issue tracking and project management
software. This gives everyone on the project a real-time view into the state of Sprints. It also
provides comprehensive management reporting tools. The Figure below shows the Sprint burn-
down chart, a snapshot of Earned Business, and a synopsis of bug status.
Figure 6 – SirsiDynix Horizon 8.0 Project Dashboard

Data from Jira can be downloaded into Excel to create any requested data analysis. High velocity
projects need an automated tool to track status across teams and geographies. The best tools
support bug tracking and status of development tasks in one system and avoid extra work on data
entry by developers. Such tools should track tasks completed by developers and work remaining.
They provide more detailed and useful data than time sheets, which should be avoided. Time
sheets are extra overhead that do not provide useful information on the state of the project, and
are de-motivating to developers.
Other companies like PatientKeeper [69] have found tools that incorporate both development
tasks and defects that can be packaged into a Sprint Backlog are highly useful for complex
development projects. Thousands of tasks and dozens of Sprints can be easily maintained and
reviewed real-time with the right tool.
Resulting Context with Integrated Scrums

Collaboration of SirsiDynix and StarSoft turned the Horizon 8.0 project into one of the most
productive Scrum projects ever documented. For example, data is provide in the table below on a
project that was done initially with a waterfall team and then re-implemented with a Scrum team
[70]. The waterfall team took 9 months with 60 people and generated 54000 lines of code. It was
103
re-implemented by a Scrum team of 4.5 people in 12 months. The resulting 50,803 lines of code
had more functionality and higher quality.
Scrum Waterfall SirsiDynix

Person Months 54 540 827
Java LOC 50,803 54000 671,688
Function Points 959 900 12673
FP per dev/month 17.8 2.0 15.3
Figure 7 – Function Points/Developer Month for collocated vs. distributed projects.

Capers Jones of Software Productivity Research has published extensive tables on average
number of function points per lines of code for all major languages [71]. Since the average lines
of code per function point for Java is 53, we can estimate the number of function points in the
Scrum application. The waterfall implementation is known to have fewer function points.
Distributed teams working on Horizon 8.0 generated 671,688 lines of code in 14.5 months with
56 people. During this period they radically refactored the code on two occasions and reduced
the code based by 275,000. They have not been penalized for refactoring as that is rarely done in
large waterfall projects in the database from which Capers derived his numbers. They have also
not been rewarded for refactoring even though reducing lines of code is viewed as important as
adding new code on well-run Agile projects.
Jones has also shown from his database of tens of thousands of projects that industry average
productivity is 12.5 function points per developer/month for a project of 900 function points and
that this drops to 3 for a project with 13000 function points [7]. Some of this is due to 4GL and
other code-automation tools used on small projects, many of which are not implemented in third
generation languages like Java.
The SirsiDynix project is almost as productive as the small Scrum project with a collocated team
of 4.5 people. For a globally dispersed team, it is one of the most productive projects ever
documented at a run rate of five times industry average.
Conclusions

104
This case study is a proof point for the argument that distributed teams and even outsourced
teams can be as productive as a small collocated team. This requires excellent implementation of
Scrum along with good engineering practices. The entire set of teams must function as a single
team with one global build repository, one tracking and reporting tool, and daily meetings across
geographies.
Outsourced teams must be highly skilled Agile teams and project implementation must enforce
geographic transparency with cross-functional teams at remote sites fully integrated with cross-
functional teams at the primary site. In the SirsiDynix case, the teams were all run from a central
site giving strong central control.
It is highly unlikely that distributed outsourced teams using current Agile Alliance best practices
of distributing work to independent Scrum teams across geographies could achieve the level of
performance achieved in this case study. Therefore, SirsiDynix sets a new standard of best
practices for distributed and outsourced teams with a previously demonstrated high level of
Agile competence.

105
Scrum and CMMI Level 5: The Magic Potion for Code
Warriors
Jeff Sutherland, Ph.D. Carsten Ruseng Jakobsen Kent Johnson
Patientkeeper Inc. Systematic Software Engineering AgileDigm Inc.
[email protected] [email protected] [email protected]
Abstract
Projects combining agile methods with CMMI1 are more successful in producing higher quality
software that more effectively meets customer needs at a faster pace. Systematic Software
Engineering works at CMMI level 5 and uses Lean product development as a driver for
optimizing software processes. Valuable experience has been gained by combining Agile
practices from Scrum with CMMI.
Early pilot projects at Systematic showed productivity on Scrum teams almost twice that of
traditional teams. Other projects demonstrated a story based test driven approach to software
development reduced defects found during final test by 40%.
We assert that Scrum and CMMI together bring a more powerful combination of adaptability
and predictability to the marketplace than either one alone and suggest how other companies
can combine them.
Introduction
One of the trends in the software industry is software projects are more comprehensive and
complex while customers at the same time request faster delivery and more flexibility.
Successful software development is challenged by the supplier’s ability to manage complexity,
technology innovation, and requirements change. Customers continually requests solutions
faster, better and more cost-effective. Agile and CMMI methods both address these challenges
but have very different approach and perspective in methods applied.
Management of complexity requires process discipline, and management of increased speed of

change requires adaptability. CMMI primarily provides process discipline and Scrum enhances
adaptability. This leads to the question, whether or not it is possible to integrate CMMI and agile
practices like Scrum to achieve the benefits from both – or even more?
This paper provides an analysis of the effect of introducing Agile practices like Scrum and story
based test driven software development and knowledge gained on what is required to be CMMI
compliant, while running an Agile company.
CMMI
The Capability Maturity Model (CMM) has existed since 1991, as a model based on best
practices for software development. It describes an evolutionary method for improving an
1
® Capability Maturity Model, CMM and CMMI are registered in the U.S. Patent and Trademark Office
106
organization from one that is ad hoc and immature to one that is disciplined and mature [72]. The
CMM is internationally recognized and was developed by the Software Engineering Institute at
Carnegie Mellon University, Pittsburgh, USA.
In 2002, a new and significantly extended version called CMMI was announced, where the ‘I’
stands for ‘Integration’ [73]. This model integrates software engineering, systems engineering
disciplines, and software acquisition practices into one maturity model. CMMI defines 25
process areas to implement. For each process area required goals, expected practices and
recommended sub-practices are defined. In addition a set of generic practices must be applied for
all processes.
The past 15 years of experience with CMM and CMMI, demonstrates that organizations
appraised to higher levels of CMM or CMMI improve the ability to deliver on schedule, cost,
and agreed quality. Increasingly, the industry requires suppliers to be appraised to CMM or
CMMI level 3 or higher [74]. A number of governmental organizations worldwide, have
established CMMI maturity requirements. Recently the Danish Minister of Science proposed
regulations to require public organizations to request documentation of their supplier’s maturity
[75].
Scrum
Scrum for software development teams began at Easel Corporation in 1993 [21] and emerged as
a formal method at OOPSLA’95 [22]. A development process was needed to support enterprise
teams where visualization of design immediately generated working code. Fundamental
problems inherent in software development influenced the introduction of Scrum:
• Uncertainty is inherent and inevitable in software development processes and products -

Ziv’s Uncertainty Principle [54]
• For a new software system the requirements will not be completely known until after the
users have used it - Humphrey’s Requirements Uncertainty Principle [59]
• It is not possible to completely specify an interactive system – Wegner’s Lemma [55]
• Ambiguous and changing requirements, combined with evolving tools and technologies
make implementation strategies unpredictable.
“All-at-Once” models of software development uniquely fit object-oriented implementation of

software and help resolve these challenges. They assume the creation of software involves
simultaneous work on requirements, analysis, design, coding, and testing, then delivering the
entire system all at once [32].
Sutherland and Schwaber, co-creators of Scrum joined forces with creators of other Agile
processes in 2001 to write the Agile Manifesto [58]. A common focus on working software, team
interactions, customer collaboration, and adapting to change were agreed upon as central
principles essential to optimizing software productivity and quality.

107
CMMI and Agile methods
Soon after publication of the Agile Manifesto in 2001, Mark Paulk principal contributor and
editor of Capability Maturity Model Version 1.0 [72], observed that Agile practices are intended
to maximize the benefits of good practice [76, 77]. “The SW-CMM tells what to do in general
terms, but does not say how to do it; agile methodologies provide a set of best practices that
contain fairly specific how-to information – an implementation model – for a particular kind of
environment.” However, Paulk noted that aligning the implementation of agile methods with the
interests of the customer and other stakeholders in a government contracting environment for
software acquisition might be an impossible task, where high customer interaction is difficult.
Surdu [78] and McMahon [79] reported positive experiences in 2006 using agile processes on
government contracts while noting the need for process discipline, good system engineering
practices, and development of self-motivated teams. Collaboration with customers was achieved
through agile education and negotiation. These studies provide practical confirmation of Paulk’s
analysis of the applicability of agile practices in a CMM environment.
Paulk [77] points out that “When rationally implemented in an appropriate environment, agile
methodologies address many CMM level 2 and level 3 practices.” Similarly Kane and Ornburn
present a mapping of Scrum and XP to CMMI [80] demonstrating that a majority of the CMMI
process areas related to Project Management can be addressed with Scrum and the majority of
process areas related to software engineering can be addressed with XP. CMMI expects that
processes are optimized and perhaps replaced over time and states: “Optimizing processes that
are agile and innovative depends on the participation of an empowered workforce aligned with
the business values and objectives of the organization.” [73] (page 49).
We agree with the authors above that Agile methodologies advocate good engineering practices
that can be integrated in the CMMI framework, and consider the largest drawback of most Agile
methodologies is a limited concept of institution-wide deployment. Institutionalization is key to
implementation of all processes in CMMI, and is strongly supported by a set of Generic
Practices. It is our belief that these practices could be used to ensure that Agile methodologies
are institutionalized in any organization.
Agile methods like Scrum and XP are practical methods that can support different parts of
CMMI. Combining Scrum and CMMI practices can produce a more powerful result than either
alone and can be done in way where CMMI compliance is maintained. A more detailed analysis
of a full implementation of the Scrum development process along with some XP engineering
practices used at Systematic shows quantitative results of introducing good agile practices and
how to maintain CMMI compliance in an Agile company.
Scrum and CMMI: a magic potion

Systematic was established in 1985 and employs 371 people worldwide with offices in Denmark,
USA and the UK. It is an independent software and systems company focusing on complex and
critical IT solutions within information and communication systems. Often these systems are
mission critical with high demands on reliability, safety, accuracy and usability.

108
Customers are typically professional IT-departments in public institutions and large companies
with longstanding experience in acquiring complex software and systems. Solutions developed
by Systematic are used by tens of thousands of people in the defense, healthcare, manufacturing,
and service industries. Systematic was appraised 11 November 2005 using the SCAMPISM2
method and found to be CMMI level 5 compliant.
Working at CMMI level 5 brings many advantages. Systematic has first hand experience of
reduction in rework by 38% to 42% over earlier levels, estimation precision deviation less than
10%, and 92% of all milestones delivered early or on time. At the same time, extra work on
projects has been significantly reduced.
More importantly, Systematic has transformed over twenty years of experience into a unified set
of processes used by all software projects. Historical data are systematically collected and
analyzed to continuously provide insight into the capability and performance of the organization.
The use of a shared common process makes it easier for people to move from project to project
and share experiences and lessons learned between projects. Insight into the capability and
performance of processes makes it possible to evaluate performance of new processes to
performance of existing processes. And this forms the foundation for continuous improvement.
CMMI 1
Project effort 100 % Rework
100%
Work
90%
Proces focus
80%
CMMI
50 % 69 %
70%
9%
SCRUM
60%
a
50%
40% 35 %
4%
30%
50 % 50 %
20% 25 %
10%
10 % 6%
CMMI 1 CMMI 5 CMMI 5
SCRUM
Figure 1: CMMI and Scrum Productivity Gains
In short, Systematic is able to deliver what the customer has ordered on schedule, cost and
quality using 69% effort compared to a CMMI Level 1 company [81, 82]. This benefit comes at
the minimal cost of 9% process focus in project management and engineering. CMMI Level 5 is
increasingly a requirement from customers and key to obtaining large contracts, especially
within defence and healthcare. Customers recognize that CMMI Level 5 gives high predictability
and better-engineered product for scalability, maintainability, adaptability, and reliability.
2 SM
Capability Maturity Model Integration, and SCAMPI are service marks of Carnegie Mellon University
109
Early results indicate that when CMMI traditional processes are optimized using Scrum, the
productivity for large projects is doubled and the amount of rework is reduced an additional 40%
over that of CMMI Level 5 companies. It is important to note that the optimized process is a
mixed process, using traditional CMMI processes to establish a project baseline expressed as a
product backlog combined with Scrum as the preferred way to implement the project in iterations
of one month Sprints. The combination of the CMMI and Scrum into the optimized CMMI
Scrum process includes the proper activities to establish sufficient planning needed by both
customer and supplier, and at the same time the flexibility and adaptability provided by Scrum.
This combined process is treated similarly to any other process in CMMI.
CMMI provides insight into what processes are needed to maintain a disciplined mature
organization capable of predicting and improving performance of the organization and projects.
Scrum provides guidance for efficient management of projects in a way that allows for high
flexibility and adaptability. When mixing the two, a magic potion emerges, where the mindset
from Scrum ensures that processes are implemented efficiently while embracing change, and
CMMI ensures that all relevant processes are considered.
Individually CMMI and Scrum has proven benefits but also pitfalls. An Agile company may
implement Scrum correctly but fail due to lack of institutionalization, (see section 0) or
inconsistent or insufficient execution of engineering or management processes. CMMI can help
Agile companies to institutionalize Agile methods more consistently and understand what
processes to address.
A company can comply with CMMI, but fail to reach optimal performance due to inadequate
implementation of processes. Scrum and other Agile methodologies can guide such companies
towards more efficient implementation of CMMI process requirements.
How Systematic adopted Scrum

Here we describe the generic steps of the process Systematic executed that resulted in the
adoption of Scrum, early testing, and story based development.
Identify Business Objectives and Needs. CMMI states [73] (page 55) that “successful process-
improvement initiatives must be driven by the business objectives of the organization”. Business
objectives and needs are addressed by the strategy of the organization.
Systematic made a strategic decision to use Lean as the dominant paradigm for future
improvements. Lean has demonstrated notable results for many years in domains such as auto
manufacturing, and due to it’s popularity, has been adapted to other domains, including product
and software development. It was expected that adoption of a Lean mindset would facilitate a
more efficient implementation of CMMI.
The strategic decision to use Lean as a dominant tool for optimization of processes, is input to
CMMI Organizational Process Focus (OPF) and driven by an organizational shared function for
process improvements.
110
Analysis. Different Lean dialects were evaluated and Lean Software Development [8] was
identified as the dialect most relevant to Systematic. Lean Software Development is an agile
toolkit. A careful interpretation of the Agile Manifesto shows that this is not necessarily in
conflict with CMMI Level 5.
The Agile Manifesto recognizes that processes, tools, comprehensive documentation, contract
negotiation and following a plan have value, but emphasizes people, interactions, working
software, customer collaboration and responding to change to have more value. “The agile
methodology movement is not anti-methodology; in fact, many of us want to restore credibility to
the word. We also want to restore a balance: We embrace modeling, but not merely to file some
diagram in a dusty corporate repository. We embrace documentation, but not to waste reams of
paper in never-maintained and rarely used tomes. We plan, but recognize the limits of planning
in a turbulent environment.” [58]
Successful application of Lean Software Development - an agile toolkit, depends on the adoption
of an agile mindset to supplement the process focus. Systematic values are consistent with the
Agile Manifesto, and special focus was placed on the following aspects for new improvements:
Individuals and interactions. Empowerment: the person executing the process is also
responsible for updating the process.
Working software over documentation. Critical evaluation of what parts of the documentation
or process can be removed or refined to increase the customers perceived value of the activities
is essential.
Responding to change. Determining how the process could be improved to support reduced
cycle time drove customer value.
Lean competences were established, through handout of books, formal and informal training, and
walk-the-talk activities. Project Managers were trained in Lean Software Development, and
Mary Poppendieck [8] visited Systematic for a management seminar on Lean Software
Development.
This seminar established an understanding of the Agile and Lean mindset. Based on this training,
causal dependencies between the principles and tools in Lean Software Development were
analyzed, and as a result test practices and reduced cycle time were identified as good candidates
for initial activities. They represented a good starting point for implementing Lean and at the
same time focused on processes where improvements would have significant effect on
efficiency.
Pilot. Lean advocates that the people performing a process should be responsible and
empowered to maintain that process. In the introduction to the CMMI OPF process area CMMI
says the same thing.
111
An analysis of the causal dependencies in Lean Software Development led to the decision to
seek improvements based on the principles of Amplify Learning, Deliver Fast, and Build
Integrity In.
Selected projects were asked if they would like to pilot improved processes related to test and
reduced cycle time respectively. Project staff were trained in the Lean mindset and asked to
suggest how to adopt Lean into their processes. The result was a selection of Scrum and early
testing based on story-based development.
The result of the pilots were two-fold: it confirmed the general idea of using Lean mindset as
source for identification of new improvements, and secondly it provided two specific successful
improvements showing how agile methods can be adopted while maintaining CMMI
compliance.
Implementation. It was decided to adopt Scrum and story based software development in the
organization. Process Action Teams (PATs) were formed to integrate the experience and
knowledge gained from the pilots, into the processes shared by all projects in the organization.
The PATs were staffed with people that would be expected to execute the new process when
released.
The largest change to project planning is that features and work are planned in sufficient detail as
opposed to a complete initial detailed analysis. The result is a Scrum Product Backlog with a
complete prioritized list of features/work for the project. All features have a qualified estimate,
established with a documented process and through the use of historical data, but the granularity
of the features increase as the priority falls. The uncertainty that remains is handled through risk
management activities.
The primary change to project execution processes, was to integrate Scrum as method for
completing small iterations (Sprints), on a selected subset of the work with highest priority.
This work verified that Scrum could be adopted in project management while maintaining
CMMI compliance. This is important because, one of the first steps to embrace change is to
ensure that project management processes support and allow agility. In addition the people
executing the process were trained and certified as ScrumMasters by Jeff Sutherland [83], who
also did a management seminar on Scrum at Systematic. Concurrent to the above pilots, Lean
was considered by all projects and shared functions as one of several ways to identify possible
improvements.
Result. The first step for Systematic in adapting a Lean mindset resulted in the adoption of
Scrum and story based development as the recommended way of working. Systematic provides a
default implementation of a Projects Defined Process (PDP) known as PDP Common. The PDP
Common has been updated to integrate Scrum and story based development into the relevant
processes.

112
The apparent result of adopting agile methods into existing CMMI compliant processes, has led
to integration of processes or process areas that initially were implemented separately. The new
processes are more efficient, and the changes have improved quality, customer and employee
satisfaction.
Risk. Some of the risks of applying Agile mindset to a CMMI organization include:
• Degrading CMMI compliant processes to non-compliance.

• Local optimizations increasing project efficiency at the expense of inefficiency at the
organizational level, e.g. due to lack of organizational coordination.
These risks were handled by a central process team that kept the organization on track with
respect to the risks and change management. The process team was responsible for:
• Build and share competencies on Lean, Agile and Scrum with the organization.
• Define and communicate vision, constraints and measures for adoption of a Lean mindset.
• Encourage and empower different parts of the organization to challenge current process
implementations with a Lean mindset, in search of improvement opportunities.
• Collect experiences from the organization and consolidate improvements at the
organizational level.
The dominant risk for failure in adapting Lean is insufficient understanding or adoption of Lean
or Agile mindset. Systematic has addressed this risk by inviting Jeff Sutherland and Mary
Poppendieck to Systematic to establish a good understanding of Lean, Scrum and Agile.
Systematic experience from pilots

In a period of approximately 4 months, two small projects piloted Scrum and early testing in
story based development.
Scrum. The first pilot was initiated on a request for proposal, where Systematic inspired by
Lean principles suggested a delivery plan with bi-weekly deliveries and stated explicit
expectations to customer involvement and feedback.
One of the main reasons that Systematic was awarded the contract was the commitment to
deliver working code bi-weekly and thereby providing a very transparent process to the
customer. During project execution, a high communication bandwidth was kept between the
team, the customer and users. This was identified as one of the main reasons for achieving high
customer satisfaction.
The delivery plan and customer involvement resulted in early detection of technological issues.
Had a traditional approach been used these issues would have been identified much later with
negative impacts on cost and schedule performance.

113
However, productivity of this small project was at the expected level compared to the
productivity performance baseline for small projects. Another small project using Scrum shows a
similar productivity and the same indications on high quality and customer satisfaction.
At Systematic, productivity for a project is defined as the total number of lines of code produced
divided by the total project effort spent in hours. These numbers are gathered from the
configuration and version control system. Data are attributed with information related to
programming language, type of code: new, reuse or test. This definition of productivity has been
chosen because it provides sufficient insight and is very simple and efficient to collect.
Systematic has established and maintains a productivity performance baseline (PPB) for
productivity compared to project size estimated in hours, from data collected on completed
projects [84]. The data shows that productivity is high on small projects and declines with the
size of the project with traditional CMMI Level 5. The productivity performance baseline in
Systematic is divided into two groups: small projects less than 4000 hours and large projects
above 4000 hours. Productivity of small projects is 181% the productivity of large projects.
When comparing the projects using Scrum to the current productivity baseline it is seen that
productivity for small projects is insignificantly changed, but the productivity for large projects
shows 201% increase in productivity. As mentioned above, the large projects did additional
improvements, and it is therefore not possible to attribute the benefit solely to Scrum. However
the people involved all agree that Scrum was a significant part of this improvement.
There is a strong indication that large projects in Systematic using Scrum will double
productivity going forward. Small projects in Systematic already show a high productivity. We
believe that this is because small projects in Systematic always have been managed in a way
similar to Scrum. However quality and customer satisfaction seems to be improved and we
believe this is because Scrum has facilitated a better understanding of how small projects are
managed efficiently.
Early testing. Two different approaches were used to facilitate early test. One large project
decided to use a story based approach to software development and another project decided to
focus on comprehensive testing during development.
The idea of story-based development was to subdivide features of work, typically estimated to
hundreds of hours of work into smaller stories of 20-40 hours of work. The implementation of a
story followed a new procedure, where the first activity would be to decide how the story could
be tested before any code was written. This test could then be used as the exit criteria for
implementation of the story.
In order to ensure that the new procedure was followed, the procedure included a few
checkpoints where an inspector would inspect the work produced, and decide whether or not the
developer could proceed to the next activity in the procedure. These inspections are lightweight,
and could typically be done in less than 5 minutes.

114
Many benefits from story-based development were immediately apparent. The combination of a
good definition of when a story was complete, and early incremental testing of the features,
provided a very precise overview of status and progress for both team and other stakeholders.
Developing a series of small stories rather than parts of a big feature is more satisfactory, and
creates a better focus on completing a feature until it fulfills all the criteria for being “done”.
This project finished early, and reduced the number of coding defects in final test by 38%
compared to previous processes.
The project using comprehensive testing placed test specialists together with the developers. As
in the story based approach, this caused discussion and reflection between testers, developers,
user experience engineers and software architects, before or very early in the development of
new functionality. As a consequence the amount of remaining coding defects in final test were
reduced by 42%.
Based on these two projects test activities should be an integrated activity through out the
projects lifetime. Scrum inherently supports this, through cross-functional teams and frequent
deliveries to the customer.
Real needs. A customer sent a request for proposal on a fixed set of requirements. When
Systematic responded, we expressed our concern that the scope and contents expressed in the
requirements were beyond the customer’s real needs.
Systematic decided to openly share the internal estimation of the requirements with the customer,
for the purpose of narrowing scope by removing requirements not needed or too expensive
compared to the customers budget. The customer agreed to re-evaluate the requirement
specification, and the result was that requirements and price were reduced by 50%.
This experience supports results in a Standish Group Study reported at XP2002 by chairman Jim
Johnson, showing that 64% of features in a fixed price contract are never or rarely used by end-
users.
We believe that this illustrates how important it is to have a high communication bandwidth with
the customer, in order to find out what the real needs are. Success is not achieved by doing the
largest project, but by doing the project that provides the most value for the customer, leaving
time for software developers to work with other customers with real needs. It gives motivation to
developers to provide solutions to real need, which in turn benefits dedication and productivity.
Even though this experience is related to activities before the project is started, the challenge of
maintaining close communication with the customer, to ensure that the project delivers the most
value within the agreed constraints, continues and is strongly supported by Scrum.
Guide for mixing CMMI and Agile

The previous section has described how Systematic, an organization appraised to CMMI Level 5,
has adopted agile methods. This section presents our advice to the agile organizations on how to
adopt the concept of institutionalization.

115
How CMMI can improve Agile
Our focus is on using CMMI to help an organization institutionalize Agile Methods. We have all
heard Agile Methods described by some as just another disguise for undisciplined hacking and of
some individuals who claim to be Agile just because they “don’t document.” We believe the
value from Agile Methods can only be obtained through disciplined use. CMMI has a concept of
Institutionalization that can help establish this needed discipline.
Institutionalization is defined in CMMI as “the ingrained way of doing business that an

organization follows routinely as part of its corporate culture.” Others have described
institutionalization as simply “this is the way we do things around here.” Note that
institutionalization is an organizational-level concept that supports multiple projects.
CMMI supports institutionalization through the Generic Practices (GP) associated with all
process areas. For the purposes of our discussion, we will look at the 12 generic practices
associated with maturity levels 2 and 3 in the CMMI [73] pp. 39-44 and how they might help an
organization use Agile Methods. We have paraphrased the generic practices (shown in bold
text below) to match our recommended usage with Agile Methods. In CMMI terms, the projects
in an organization would be expected to perform an activity that accomplished each of these
generic practices. We have used Scrum as the example Agile Method to describe some of the
activities that relate to these practices.
Establish and maintain an organizational policy for planning and performing Agile
Methods (GP 2.1). The first step toward institutionalization of Agile Methods is to establish
how and when they will be used in the organization. An organization might determine that Agile
Methods will be used on all projects or some subset of projects based on size, type of product,
technology, or other factors. This policy is a way to clearly communicate the organization’s
intent regarding Agile Methods. In keeping with the Agile Principle of face-to-face conversions
at “all hands meeting” or a visit by a senior manager during a project’s kick off could be used to
communicate the policy.
Establish and maintain the plan for performing Agile Methods (GP2.2). This practice can
help ensure that Agile Methods do not degrade into undisciplined hacking. The expectation is
that Agile Methods are planned and that a defined process exists and is followed. The defined
process should include a sequence of steps capturing the minimum essential information needed
to describe what a project really does. The plan would also capture the essential aspects of how
the other 10 generic practices are to be implemented in the project. In Scrum, some of this
planning is likely to be captured in a product backlog and/or sprint backlog, most likely within a
tool as opposed to a document.
Provide adequate resources for performing Agile Methods (GP2.3). Every project wants,
needs, and expects competent professionals, adequate funding, and appropriate facilities and
tools. Implementing an activity to explicitly manage these wants and needs has proved useful. In
Scrum, for example, these needs may be reviewed and addressed at the Sprint Planning Meeting
and reconsidered when significant changes occur.
116
Assign responsibility and authority for performing Agile Methods (GP 2.4). For a project to
be successful, clear responsibility and authority need to be defined. Usually this includes a
combination of role descriptions and assignments. The definitions of these roles identify a level
of responsibility and authority. For example, a Scrum Project would assign an individual or
individuals to the roles of Product Owner, ScrumMaster, and Team. Furthermore, the roles in
the Team are likely to include a mix of domain experts, system engineers, software engineers,
architects, programmers, analysts, QA experts, testers, UI designers, etc. Expertise in the Team is
likely to include a mix of domain experts, system engineers, software engineers, architects,
programmers, analysts, QA experts, testers, UI designers, etc. Scrum assigns the team as a whole
the responsibility for delivering working software. The Product Owner is responsible for
specifying and prioritizing the work. The ScrumMaster is responsible for assuring the Scrum
process is followed. Management is responsible for providing the right expertise to the team.
Train the people performing Agile Methods (GP 2.5). The right training can increase the
performance of competent professionals and supports introducing new methods into an
organization. People need to receive consistent training in the Agile Method being used in order
to ensure institutionalization. This practice includes determining the individuals to train,
defining the exact training to provide, and performing the needed training. Training can be
provided using many different approaches, including programmed instruction, formalized on-
the-job training, mentoring, and formal and classroom training. It is important that a mechanism
be defined to ensure that training has occurred and is beneficial.
Place designated work products under appropriate level of configuration management (GP
2.6). The purpose of a project is to produce a deliverable product (or products). This product is
often a collection of a number of intermediate or supporting work products (code, manuals,
software systems, build files, etc.). Each of these work products has a value and often goes
through a series of steps that increase their value. The concept of configuration management is
intended to protect these valuable work products by defining the level of control, for example,
version control or baseline control and perhaps multiple levels of baseline control to use within
the project.
Identify and involve the relevant stakeholders as planned (GP 2.7). Involving the customer
as a relevant stakeholder is a strength of Agile Methods. This practice further identifies the need
to ensure that the expected level of stakeholder involvement occurs. For example, if the project
depends on customer feedback with each increment, build, or sprint, and involvement falls short
of expectations it is then necessary to communicate to the appropriate level, individual, or group
in the organization to allow for corrective action. This is because corrective action may be
beyond the scope of the project team. In advanced Scrum implementations, this is often
formalized as a MetaScrum [41] where stakeholders serve as a board of directors for the Product
Owner.
Monitor and control Agile Methods against the plan and take appropriate corrective action
(GP 2.8). This practice involves measuring actual performance against the project’s plan and
117
taking corrective action. The direct day-to-day monitoring is a strong feature of the Daily Scrum
Meeting. Further, examples of this can be seen in Scrum with the use of the Product Burndown
Chart showing how much work is left to do at the beginning of each Sprint and the Sprint
Burndown Chart showing the total task hours remaining per day. Scrum enhances the
effectiveness of the plan by allowing the Product Owner to inspect and adapt to maximize ROI,
rather than merely assuring plan accuracy.
Objectively evaluate adherence to the Agile Methods and address noncompliance (GP2.9).
This practice is based on having someone not directly responsible for managing or performing
project activities evaluate the actual activities of the project. Some organizations implement this
practice as both an assurance activity and coaching activity. The coaching concept matches
many Agile Methods. The ScrumMaster has primary responsibility for adherence to Scrum
practices, tracking progress, removing impediments, resolving personnel problems, and is
usually not engaged in implementation of project tasks. The Product Owner has primary
responsibility for assuring software meets requirements and is high quality.
Review the activities, status, and results of the Agile Methods with higher-level
management and resolve issues (GP2.10). The purpose of this practice is to ensure that higher-
level management has appropriate visibility into the project activities. Different managers have
different needs for information. Agile Methods have a high level of interaction, for example,
Scrum has a Sprint Planning Meeting, Daily Scrum Meetings, a Sprint Review Meeting, and a
Sprint Retrospective Meeting. Management needs are supported by transparency of status data
produced by the Scrum Burndown Chart. This, in combination with defect data can be used to
produce a customized management dashboard for project status. Management responsibilities are
to (1) provide strategic vision, business strategy, and resources, (2) remove impediments
surfaced by Scrum teams that the teams cannot remove themselves, (3) ensure growth and career
path of staff, and (4) challenge the Scrum teams to move beyond mediocrity. The list of
impediments generated by the Scrum teams is transparent to management and it is their
responsibility to assure they are removed in order to improve organizational performance.
Establish and maintain the description of Agile Methods (GP 3.1). This practice is a
refinement of GP2.2 above. The only real difference is that description of Agile Methods in this
practice is expected to be organization-wide and not unique to a project. The result is that
variability in how Agile Methods are performed would be reduced across the organization; and
therefore more exchange between projects of people, tools, information and products can be
supported.
Collect the results from using Agile Methods to support future use and improvement the
organization’s approach to Agile Methods (GP 3.2). This practice supports the goal of
learning across projects by collecting the results from individual projects. The Scrum Sprint
Retrospective Meeting could be used as the mechanism for this practice.
All of these generic practices have been useful in organizations implementing other processes.
We have seen that a number of these generic practices have at least partial support in Scrum or
118
other Agile Methods. We believe that implementing these practices can help establish needed
discipline to any Agile Method.
Critiques of CMM
In research funded by the Danish government, Rose et. al. surveyed the literature on critiques of
CMM [85]. They observed that the chief criticism of CMM is not the process itself, but the
effects of focus on process orientation. While side effects of process focus may be viewed as
simply poor CMM implementation, organizations with heavyweight processes are highly prone
to poor execution.
As with any other model, good and bad implementations of CMM exist. We believe that bad
implementations are one of the main reasons for the existence of many negative criticisms of
CMM. Such implementations are often characterized as in the table below, whereas many good
CMM implementations address most of the criticism.
One way to enhance chances for a good CMM or CMMI implementation is to use Scrum.
Applying Scrum and agile mindset while implementing CMMI will help to recognize that CMMI
addresses people and technology aspects, in a way fully supportive of an agile mindset.
More importantly, the work in this paper has shown that the mix of CMMI and Scrum blends a
magic potion for software development that is even better than the sum of the two alone.
We acknowledge that the CMM criticism listed in the table below exist, but from our knowledge
of CMMI we consider it to be incorrect. But a bad implementation of CMMI may be perceived
this way. Even though good CMMI implementations can be done without agile methods, the
table shows that Scrum will contribute with a beneficial focus on issues stemming from “bad”
CMMI implementation.
CMM criticism Scrum support

CMM reveres process but ignores people. Scrum is the first development process to treat
people issues the same as other project
management issues [86].
Does not focus on underlying organizational A primary responsibility of the ScrumMaster is
problems that should be solved. to maintain and resolve an impediment list that
contains organizational issues, personal issues,
and technical problems.
Ignores quality in the software product The Scrum Product Owner is responsible for
assuming an unproven link between quality in continuously reprioritizing the Product
the process and quality in the resulting product. Backlog to maximize business value in current
Differing project and organizational context.
circumstances may mean that a process that
delivers a good product in one context delivers
a poor product in another context.
Lack of business orientation The primary focus of Scrum is on delivering
business value.

119
Poor awareness of organizational context. Creation and prioritization of features, tasks,
and impediments is always done in
organizational context by inspected and
adapting.
Ignores technical and organizational Daily inspection and adaptation in Scrum
infrastructures. meetings focuses on technical and
organizational issues.
Encourages an internal efficiency focus and Focus is on delivering business value. Type C
thus market and competition blindness. Scrum allows an entire company to dominate a
market segment through inspecting and
adapting in real time to competition [41].
Conclusions
This paper shows that CMMI and Scrum can be successfully mixed. The mix results in
significantly improved performance while maintaining compliance to CMMI Level 5 as
compared to performance with either CMMI or Scrum alone.
Scrum pilot projects showed significant gains in productivity and quality over traditional
methods. These results led to an ROI based decision to more widely introduce Scrum and
consider other Agile practices in Systematic. Scrum now reduces every category of work
(defects, rework, total work required, and process overhead) by almost 50%.
This paper has outlined how Systematic adopted Scrum and story based development into its
CMMI processes inspired from a strategic focus on Lean. For Agile companies the article has
presented how Generic Practices from CMMI can be used to institutionalize agile practices.
Furthermore the article has presented Lean Software Development [8] as an operational tool to
identify improvement opportunities in a CMMI 5 company.
We think companies in defense, aerospace, and other industries that require high maturity of
processes, should carefully consider introducing Agile practices into the workplace and all
software companies should consider introducing CMMI practices into their environment.
Our recommendation to the Agile community is to use the CMMI generic practices from CMMI
Level 3 to amplify the benefits from Agile methods. The efficiencies of agile practice can lower
the cost of CMMI process improvements making the benefits more widely available. Our
recommendation to the CMMI community is that Agile methods can fit into your CMMI
framework and will provide exciting improvements to your organization.

120
Chapter 5: Scrum Metrics
From the beginning, Scrum was designed to provide transparency to both team members and
those outside the team. The formation of the Scrum Alliance and the biannual Scrum Gatherings
provides a forum for the Scrum community to further develop and refine approaches for
management reporting.
A typical Scrum maintains a Scrum board showing columns of user stories, development tasks
relating to each story, and the state of each development task and tests associated with each
story. When a task is started a card is moved across the board into an open column. When code is
complete it is move to the verification column and when fully tested is moved to a done column.
An updated Scrum burndown chart along with a priotized list of impediments is often posted as
well. Some sites even have Lava lamps that change color depending on the state of the build or
state of the Sprint.
A manager can walk by a Scrum board and see the state of the team in a few seconds. If an
impediments list is posted, a manager can add relevant items to his or her work list. This
eliminates the need for most status reporting, particularly if critical information is put online on a
web page, a wiki, or a reporting tool.
Managing and tracking costs to evaluate actual Return on Investment (ROI) can provide an
additional feedback loop. Earned Value Management (EVM) may be of interest to management.
Sulaiman, Barton, and Blackburn [87] provide detailed calculations for EVM and evaluation of
its effectiveness in an Agile 2006 research paper.

121
Reporting Scrum Project Progress to Executive
Management through Metrics
Brent Barton, Ken Schwaber, Dan Rawsthorne
Contributors: Francois Beauregard, Bill McMichael, Jean McAuliffe, Victor Szalvay
Scrum Alliance, 2005
Introduction
The interest in Agile software methodologies is not surprising. Agile methods are presenting an
opportunity to develop software better and this is being noticed in the business community.
Scrum is particularly of interest partly because of its ROI focus and quick implementation.
While the efforts of innovators and early adopters have helped us assert that Agile is better than
traditional methods, improving the reporting capability would help. Even better would be able to
report project progress to executive management in a more compelling way. At a Scrum
Gathering, white papers were submitted and discussed. This is a summary of those discussions
and the integration of the contributions of many people. Visibility into project progress and
project “health” is a consistent theme executive management desires.
Transparency into Projects

Executive Management needs transparency into all operations by viewing important indicators
quickly: This is especially true of software projects. They want no surprises because in software
a surprise is rarely a pleasant one. It is worth mentioning, however, that bad things do happen;
executives know this and so does everyone else. It is always a surprise the first time one hears
bad news. In contrast, the kind of surprise executives hate the most, have significant impact and
were known much earlier than when they were finally informed. The negative emotional
response to the surprise is reinforced by the realization that decisions were made on faulty
information and this was preventable.
There are many techniques and practices for assessing the progress and probable success of
projects. Scrum provides four simple and effective artifacts for managing and monitoring project
performance: Product Backlog, Product Burndown, Sprint Backlog and Sprint Burndown.
Building on these, we are integrating a Functional Work Breakdown Structure and a technique
for measuring Earned Business Value.
Stakeholders and executives often have particular interest in certain areas of projects. The
grouping nature of a Work Breakdown Structure (WBS) affords the opportunity to present
progress at a mid-level: not a single view like a burndown and not at a detail level like a backlog.
By combining a WBS, transparency can be attained quickly with a few simple, graphical reports
on an executive dashboard

122
Executive Dashboard
The Executive Dashboard presented here is easily read, interpreted and provides the ability to
reference additional material if desired (see Figure 2: Executive Dashboard).
ATM Project Dashboard
Parking Lot Product Burndown

26% KS 17% BB 12% DR
Login Withdraw Cash Deposit Check 3000
(5) (4) (3) 2500
2000
( 100% ) ( 40% ) ( 32% )
1500
January 2005 February 2005 March 2005
Work
1000
25% KS 17% DR 2% BB 500

Business Support 0
Transfer Funds Buy Stamps
Activities
0 1 2 3 4 5
-500
(6) (7) (2) Estimated Completion
-1000 May 3, 2005
( 17% ) ( 0% ) ( 0% )
Sprint
Planned Completion March 2005 March 2005
LEGEND Work Breakdown Structure

BVI Owner
Feature Group, ATM
Use Case Package Completed
or Use Case Attention 3 - Product 0 - Team 1 - Business
# features or stories In Process 1- Function 0 - Structure Management 2 - Sales

Not Started
% complete 15 - Login Conversions Team Training 1 - Marketing
Progress Bar 10 - Withdraw Rewrites Dev/SCM/Tes 1 - User

Planned Completion
7 - Deposit Refactoring Dev Process 2 - User Docs
10 - Transfer App 1 - Business
1 - Buy Tools 1 - Adapt

Business Value Maintenance
$600,000
$500,000
Space for:
Links
Business Value
$400,000 Risks
Issues
Highlights
$300,000
$200,000
$100,000
$-
0 1 2 3 4 5
Sprint
Figure 4: Executive Dashboard

123
The contents of this dashboard report include:
Parking Lot: This is a pictorial that statuses groups of features or use cases. This has been
adopted from reports found in Feature Driven Development (FDD). With the addition of a
Business Value Index (described later), one can see the progress and value of this area to the
business. At a glance, the colors show where progress is made, areas of concern are and items
not started. The BVI represents the total value of the project and the owner’s initials describe
who is responsible for the groups. The legend is included (see Figure 3: Parking Lot).
26% KS 17% BB 12% DR
Login Withdraw Cash Deposit Check
(5) (4) (3)
( 100% ) ( 40% ) ( 32% )
January 2005 February 2005 March 2005
25% KS 17% DR 2% BB
Business Support
Transfer Funds Buy Stamps
Activities
(6) (7) (2)
( 17% ) ( 0% ) ( 0% )
Planned Completion March 2005 March 2005
LEGEND
BVI Owner
Feature Group,
or Use Case Attention
# features or stories In Process
Not Started
% complete
Progress Bar
Planned Completion
Figure 5: Parking Lot
Product Burndown: The burndown in work budgeted and planned compared as decreased by
work completed across time. Based upon this, an estimated completion date can be determined
as the trend line crosses the x-axis (see Figure 4: Product Burndown).
bbb
Figure 6: Product Burndown

124
Earned Business Value Graph: This presents the Business Value earned compared to the
Planned Business Value. Variance can be quickly estimated from the graph to assess the correct
prioritization and progress of the project Figure 5: Business Value Burnup).
Business Value
$600,000
$500,000
Business Value
$400,000
$300,000
$200,000
$100,000
$-
0 1 2 3 4 5
Sprint
Figure 7: Business Value Burnup
Graphical Work Breakdown Structure: This visual representation provides a concise, high-level
presentation of the project work items (see Figure 6: Functional Work Breakdown Structure).
Space for links, highlights, issues and risks. Every project and customer has its own specific
needs. This space is intended for a few bullet points.
Work Breakdown Structure
Dan Rawsthorne introduced a functional Work Breakdown Structure which provides us a

structure for reporting key areas within a project and also measuring Earned Business Value. A
Work Breakdown Structure provides “’A deliverable-oriented grouping of project elements
which organizes and defines the total scope of the project.’ [88]
b
Many think of Gantt charts and Microsoft Project Plans when they hear the term Work
Breakdown Structure. This visually appealing format allows anyone to quickly see the salient
work required to accomplish the project (Project at a Glance). This sample software project’s
WBS looks like the following, representing a fictitious ATM development project (see Figure 6:
Functional Work Breakdown Structure).

125
Figure 8: Functional Work Breakdown Structure
Each of bottom nodes in the Functional leg is a use case. A use case is not required but scenarios
of use cases and stories align well and help produce useful software. Using other forms of
requirements does not invalidate this structure.
Notice the numbers in the nodes that represent tangible things that can be valued. The other
items are necessary to deliver the required results but do not have direct business value.
Earned Business Value
In order to represent the Earned Business Value (EBV) of a project and its components, an
additive weight needs to be assigned. Total Business Value is determined by some ROI
calculation or equivalent. Business Value becomes earned only when the items are done. In
Scrum terms this means it is an “increment of potentially shippable functionality.” Thus, only
items of direct business value, such as functionality and training should be assigned weights
other than zero. The other items are the cost of doing business. By calling them “orphans” they
need to be adopted by items that do have value (Note: This is useful because it addresses total
cost, not just cost-per-feature of a project and makes visible the cost of doing business in
software. Also, the software team is reminded the difference between important work and
business value of the output).
Detailed calculation of EBV is was published in the Agile 2006 proceedings . Here, only a brief
overview is provided for one calculation [87]. In order to apply Business Value (BV) to a
project, we need to calculate the Business Value Indexes. The Business Value Index (BVI) of
the entire ATM project equals 1. For each level in the WBS, the index is 1: This is an
intermediate value that will be used to calculate the BVI. To calculate the BVI of subsequent
levels, you must also multiply the BVI of the all nodes above the level you are considering.
The next part is a bit more complex but the pattern is easy once you understand it. Remember,
the BVI of the next level is 1*1=1, the index of the Project times the index of the next lower
level. Because the functional leg has a weight of 3 and the business leg has a weight of 1, the
sum of the additive weights of this level is 3+1=4. Thus, the Index of the functional leg is 3/4
and the business leg 1/4: This was obtained by dividing the weight of the leg by the sum of the
126
weights on this leg. This is why additive weights must be used. So in order to calculate the BVI,
multiply the index of the target leg by the index of all the legs above it. Since the only leg above
the functional leg is the ATM project, 3/4 * 1 = 3/4. the weight of the leg by the sum of the
weights of all legs. Thus the BVI for the functional leg is 3/4 and for the business leg is 1/4.
Using the same method, the BVI of the login use case is 1 * 3/4 * 1 * 15/43 = 45/172. If the
Total Business Value of the project is $500,000, then the Earned Business Value (EBV) realized
by completing the Login Use case would be $130,814.
The Underlying Data
Managing and reporting effectively is a lot of work. The validity of the reports is only as good
as the validity of the data. Figure 7: WBS in Product Backlog Format captures the WBS and
calculates the BVI for each level. Note that each use case has been broken into stories
(scenarios) and the weights and BVI have been calculated as well. For each item marked as
done, the EBV is calculated. Notice the bottom row is done but there is no EBV because this
represents a cost of doing business.
Area Sub-Level Use Cases Value BVI Stories or FeaturesConcerns and
Gather Stakeholder Value BVI Estimate Done Sprint EBV
Product Function Login 15 26.2% determine postconditions 0 0% 8 1 1 $ -
Product Function Login Determine Main Success Scenario 0 0% 16 1 1 $ -
Product Function Login Codey up Main Success
y Scenario 10 20% 80 1 1 $ 100,626
Product Function Login Extensions 0 0% 30 1 1 $ -
Product Function Login Code Up "3 Strikes and You're Out" 3 6% 16 1 1 $ 30,188
Product Function Withdraw Cash 10 17.4% determine postconditions 0 0% 12 1 2 $ -
Product Function Withdraw Cash Determine Main Success Scenario 0 0% 20 1 2 $ -
Product Function Withdraw Cash Codey up Main Success
y Scenario 7 12% 120 1 2 $ 61,047
Product Function Withdraw Cash Extensions 0 0% 30 0 2 $ -
Product Function Withdraw Cash Code Up "Quick Cash Options" 3 5% 20 0 2 $ -
Product Function Deposit Check 7 12.2% determine postconditions 0 0% 20 1 2 $ -
Product Function Deposit Check Determine Main Success Scenario 0 0% 20 1 2 $ -
Product Function Deposit Check Codey up Main Success
y Scenario 10 10% 100 0 2 $ -
Product Function Deposit Check Extensions 0 0% 30 1 2 $ -
Product Function Deposit Check Code Up "Deposit Foreign Currency" 2 2% 50 0 2 $ -
Product Function Transfer Funds 10 17.4% 300 0 3 $ -
Product Function Buy Stamps 1 1.7% 100 0 3 $ -
Product Structure Conversions 0.0% 200 0
Product Structure Rewrites 0.0% 300 0
Product Structure Refactoring 0.0% 250 0
Team Team Management 0.0% 120 0
Team Team Team Training 0.0% 40 1 1
Figure 9: WBS in Product Backlog Format
So, how does this information produce a dashboard?
Putting the Data into the Report
Sprint Work Left The Product Burndown is automatic, using the columns, Estimate, Done and
0 2362 Sprint. Rather than complete automation, the auto-filter feature in Excel is
1 2052 applied for each Sprint and this can then be easily tabulated into this table
2 1152 (see Figure 7: Product Burndown). This is found in the header section of the
3 Product Backlog in the attached spreadsheet. Using the subtotal function in
4 Excel provides quick calculations from filtered data. Applying no filter
5
Figure 9: Product 127
Burndown
calculates the Work Left before the start of Sprint 1. Sprint 1 was found by using a custom filter
where Sprint “is not equal to 1”. Sprint 2 was found by applying the filter where Sprint “is
greater than 1.” As long as any items that are dropped form an active Sprint is reflected in the
Product Backlog as planned for the next Sprint (unless the Product Owner changes that to a
Sprint further out).
Creating the Parking Lot
The Parking Lot is created using the filters as well. For the “Deposit Check” use case, filter on
Deposit Check in the use case column. The number of stories equals 5 so this will go in that
middle of the diagram (see Figure 8: Deposit Check Progress Indicator). The total estimated
time to completion is 220 and the total estimate of time where “Done” is filtered to 1 is 70.
Thus, the percentage complete is 70/220 ~32%. Using the elements in the “Parking Lot
Construction” tab in the spreadsheet, you can copy and paste the progress bar, so it graphically
approximates this completion value. Since Scrum uses forward looking estimates, it is possible
for a use case or feature group to show negative progress compared to a previous report as new
information causes estimates to increase. The date is formatted in month and year but can be
reformatted to reflect the lengths of iterations.
LEGEND
BVI Owner
12% DR Feature Group,
Deposit Check or Use Case Attention
# features or stories In Process
(3) Not Started
% complete
( 32% ) Progress Bar
Planned Completion
March 2005
Figure 10: Deposit Check Progress Indicator
Earned Business Value
Earned Business Value is easily calculated by filtering on the Done field for a value of 1. Then
filter on the Sprints up to and including the one being calculated using the custom filter “is less
than or equal to.” This yields the data in the EBV column for each Sprint (see Figure 9: Earned
Business Value Data). This is located in the “EBV” tab in the accompanying spreadsheet. Note
the Planned Business Value is calculated initially. Value is typically realized using an ‘S’ curve.
The sample here has only 5 data points so the smoothing feature was used for the Planned
Business Value. If the project is highly emergent (little planning beyond the current Sprint) the
planned business value will only be one row ahead of the earned business value.

128
Conclusions
By taking advantage of the visibility that Agile methodologies provide, we can deliver
meaningful information all the way up to Executive Management using graphical
Sprint Planned Business Value Earned Business Value
representations. The addition of a functional work 0 $ - $ -
breakdown structure provides the ability to view the project 1 $
2 $
100,000 $
150,000 $
130,814
191,861
at multiple levels of granularity. 3 $ 300,000
4 $ 475,000
5 $ 500,000
Note:
Notes This column represents skewed
data because of graph
smoothing. Because there are
only 5 data points, a standard S-
shaped Business Value is not
Many other metrics can be derived from the core data, represented well without
smoothing
including early warning indicators and schedule variances.
How weights are applied so they are additive can be Figure 11: Earned Business Value Data
debated or can be simple. This should be coordinated with the business and financial people to
use their calculators to help derive this.

129
Chapter 6: Scrum Tuning
There is only one Scrum and one set of core processes. The inspect and adapt rules of Scrum
allow it to be fine tuned for any organization. However, one of the goals of the first Scrum was
to multiple productivity by an order of magnitude. The Smalltalk development environment
created by the first Scrum at Easel Corporation had a money back guarantee to double
productivity the first month of use. This may be the only money back guarantee of its type seen
in the software industry.
There are some leaders in the Agile community that view productivity increases as marginal -
zero to 25%. They do Agile development to improve quality and the lifestyle of the development
team. While these goals are laudable, they are a prescription for getting your development team
outsourced. There is a Scrum team in St. Petersburg, Russia, that can give you a velocity of
seven times a typical waterfall team and developer salaries are much lower than in the U.S. or
Europe. Scrum was designed to enable teams in the U.S. and Europe to produce software at half
the cost of a waterfall team in India, Dubai, Russia, or China. Keeping a job is the first step to
job satisfaction.
One of the motivating factors in creating Scrum was the Borland Quattro Pro project documented
in detail by James Coplien when he was at ATT Bell Labs [2]. Each developer on this project
generated 1000 lines of production C++ code every week for three years. In comparison,
productivity on the recent Microsoft Vista project was 1000 lines of code per developer per year.
The Borland Quattro Project was 52 times as productive as the Microsoft Vista project measured
by lines of code.
Mike Cohn, in his book on User Stories [70], documents a well running Scrum of 6 people that
reimplemented an application with 1/10 of the resources of a team that had used traditional
Waterfall practices to implement that same functionality. The first Scrum at Easel Corporation
achieved the same level of productivity. More recently, the SirsiDynix project demonstrated that
an outsourced, distributed Scrum on a large project with over one million lines of Java code
could achieve about the same high level of productivity [29] as small collocated teams.
It is therefore surprising when consulting with the many companies using Scrum that some get
little increase in productivity, some double productivity, some quadruple it, but few achieve the
order of magnitude increase for which Scrum was designed. Just like Rugby teams, some
companies implement well and win all the time. Others do not do as well. When you look under
the hood you find that the implementation of play is done poorly for some teams.
In an attempt to illustrate good, better, and best implementations of Scrum, implementations of

Scrum were classified into Type A, Type B, and Type C. People thought these were different
Scrums with different core practices. The core practices are the same. However, the style of
some implementations are better than others with improved velocity and quality achievements.
We wanted to analyze them so that Scrum practitioners could see the effect of rigorously

130
implementing the basic Scrum practices and doing it in ways that radically improved
development. This is Scrum tuning, or ways of implementing the core practices of Scrum at
different levels of rigor and adding some special sauce that takes the Scrum implementation from
“good” to “great.”
A better way to describe the evolution of Scrum styles is to show how focus on impediments in
Scrum implementations paved the way for more rigorous implementations of Scrum. Initially we
were concerned about making a team successful, then it became apparent the making the product
successful was a higher goal. Once that was achieved, we noticed the best product doesn’t
always win. The company as a whole must inspect and adapt and become Agile along with the
software developers.

131
Type B Scrum Continuous Flow: Advancing the State of the
Art3
Patientkeeper, Inc., Brighton, MA, 2005.
Introduction
Scrum is a process designed to add energy, focus, clarity, and transparency to software product
development. It can increase speed of development, align individual and corporate objectives,
create a culture driven by performance, support shareholder value creation, achieve consistent
communication of performance at all levels, and enhance individual development and quality of
life. Scrum adds value to the working life of software developers, while demonstrating to
management that radically increased throughput is possible and sustainable over time.
Background on Scrum
In 1983 I began work on better software development processes at Mid-Continent Computer
Services in Denver doing data processing for over 150 banks across the United States. This work
was continued at Saddlebrook Corporation, Graphael, Inc., and Object Databases on the MIT
campus in Cambridge, MA during 1986-1993. The goal was to deliver better software sooner
while improving the working environment for developers. Software teams are often late and
over budget and developers are usually blamed for failure. They are often punished and forced to
endure “death marches” where they regularly burn out and quit as quality deteriorates, morale
degenerates, and people are forced to work harder to produce less. Professor Peter Senge of MIT
views this phenomenon as generic to U.S. management whose leadership is dedicated to
mediocrity and destroys the spirit of the people [89].
In fact, as extensive data shows from an advanced Scrum implementation at PatientKeeper, Inc.
(see later chapter), late projects are usually the result of management failure. Resources are
continually diverted to non-critical activities instead of being fully allocated to projects, and
corporate waste in the form of unnecessary meetings and beaurocracy is rampant. Management’s
failure to track and remove these corporate impediments causes productivity loss. A recent CIO
Magazine survey showed that 96% of CIOs in the United States have no information on project
status during the middle of a project. Flying blind leads to regular project crashes. Lack of
visibility inside projects, particularly waterfall projects, allows management to blame the
developers instead of taking the consequences and learning from their actions.
3
First published in Cutter Agile Project Management Advisory Service: Executive Update, 2006.
132
Project success is only a preliminary objective of Scrum. Our goal is to create an environment
where developers exceed management expectations and create new products faster than
marketing, sales, and customers can absorb them. This enables a world-class Scrum development
team that can focus on exceptionally high quality and usability while positioning their company
to dominate a market.
After evolving concepts at four companies, the first Scrum was launched at Easel Corporation in
1993 with the help of Jeff McKenna and John Scumniotales, the first ScrumMaster. Results
exceeded all expectations and the experience was life-changing for the development team. In
1995, Ken Schwaber observed Scrum in action and presented the first paper on Scrum at
OOPSLA. Ken and I co-developed Scrum for industry-wide deployment using many companies
as laboratories to enhance and scale Scrum. In 1999, Mike Beedle standardized Scrum as an
organizational pattern at the Pattern Languages of Programming Design (PLoP) Conference and
in 2001, the three of us co-authored the Manifesto for Agile Software Development along with
14 other software experts. Two Scrum books were written by Ken Schwaber and Mike Beedle
and the Scrum Alliance was formed .
A recent innovation has been automating real-time data collection for Scrum activities to reduce
administrative time required from developers and product managers, and to allow better
reporting to senior management. A side effect of this automation has been (1) collapsing bug
reporting and status of development tasks into one system, (2) real time reporting on thousands
of tasks on dozens of projects, (3) enhanced ability for the Product Owner to steer the
development team in the right direction by better prioritization of the Product Backlog, (4)
providing developers automated tasks lists where they can more easily self-manage work, (5)
radically increasing throughput for multiple development teams, and (6) reducing project
tracking overhead to 60 seconds a day for developers and 10 minutes a day for a project
manager.
A side effect of reducing data collection costs to near zero has been increased awareness and
management insight through total transparency of project status. This allows development
leaders and senior management to evaluate process bottlenecks and remove wasted effort. The
relationship between project failures and the lack of enforcement of lean manufacturing and lean
product development principles can be precisely demonstrated through Scrum project tracking
data. Lean principles used at Toyota and other companies drive continuous quality improvement
and can be used as a method to analyse project management failures. Rothfuss [90] reviewed
PatientKeeper’s system and reported that it achieved:
• Unprecedented transparency
• Companywide visibility
• Metrics driven decision making
• Peer pressure and pride in workmanship driving productivity
These qualities make Scrum useful for implementation of Lean practices. These practices are
useful for open source development as well as ISV product development or internal IT
133
initiatives. Scrum is moving towards more broad application, while becoming faster, better,
cheaper, and more rewarding for customers, developers, and managers.
Improving Scrum Implementations

One of the key influences that led to the creation of the Agile development processes,
particularly Scrum, was a paper on Japanese new product development by Takeuchi and Nonaka
[1]. The authors presented a chart showing product development activities (requirements,
analysis, design, implementation, testing, deployment) separated into siloed phases of work
(Type A or waterfall), phases slightly overlapped (Type B), and all phases of development
overlapping (Type C). The Japanese viewed Type A product development at NASA as an
outmoded relay race type of process. Type B at Fuji-Xerox and Type C at Honda they
envisioned as similar to a Scrum formation in Rugby where all team members joined arms and
phases of product development were overlapping.
Type A – Isolated cycles of work
Type B – Overlapping iterations
Type C – All at once
Figure 12: Type A, B, and C strategies for delivering product [11].
One can reinterpret this diagram at a higher level and apply it to different types of Scrum
execution. Early Scrum implementations focus on helping the software development team to
become more productive. Teams often do work within a timebox called a Sprint. If no work is
done within the current Sprint to prepare for the next Sprint, time may be lost between iterations
while reorganizing for the next Sprint (reset time in Lean manufacturing). If the Product Owner
team in Scrum works on reprioritizing, specifying, and estimating (with developers) prior to the
next Sprint Planning meeting, this time delay can be eliminated. By reducing reset time to zero, a
Continuous Flow Scrum allows work to flow smoothly from Sprint to Sprint and can often
double throughput. An All-at-Once Scrum can be envisioned as pipelining Sprints, i.e. running
multiple overlapping Sprints through the same Scrum team. This is the type of process seen at
PatientKeeper and at a few other software development companies.
134
Many companies, large and small, are experienced with Scrum and have the challenge of
delivering a portfolio of products developed by multiple synchronized teams. While few are
ready for a All-At-Once Scrum, most want to implement a Continous Flow Scrum. This involves
introducing lean product development principles to guide the self-organization of a Scrum
implementation in a company setting. There is also value in understanding and implementing
some principles from lean manufacturing, lean logistics, and lean supplier management as lean is
a system of management of an entire organization, more than a collection of techniques and
practices. Many of these practices are directly useful for software development [8].
Enhancing Scrum Implementations

The differences in a Team Scrum, Continuous Flow Scrum, and an All-At-Once Scrum have
been well summarized by Ryan Martell [91] below. A major thrust of Scrum is examining,
prioritizing, and removing impediments to shorten cycle time and deliver production software
faster. Shorter delivery cycles means quicker time to market, faster evolution of the product, and
increased knowledge gained by the product teams. Toyota has demonstrated that introducing
lean manufacturing to shorten cycle time improves product quality and overwhelms competition.
New product is introduced so fast and functionality increases and improves so quickly, the
competition cannot keep up.

135
Table 1: Scrum characteristics [91].
Team Scrum Continuous Flow Scrum All-At-Once Scrum
Iteration/Sprint Down-time between Slight overlap Complete and
Overlap Sprints/No overlap (Planning/Prep) Multiple
(insufficient Product
Owner engagement)
Level of Development Team Product Management Entire Organization
Involvement Team
Release Every 4 - 6 Every 2 – 4 Iterations Every

Iterations Iteration/Sprint
Iteration Lengths Fixed Fixed Multiple
Overlapping w/
varying lengths
Cycle Time for 4-6 Months 2-3 Months Monthly
New Requests
delivery
Release per year Total – 2 Total – 4 Total - 12

2 Major year Quarterly Major (4) Weekly patch, (8)
patch as needed patch as need Monthly update (8)
Quarterly Major (4)
Here, we focus on Continuous Flow Scrum which is the next step for companies that have
mastered the basic Scrum process. Following the examples reviewed by Takeuchi and Nonaka
[11] we set an audacious goal – cut release time in half, one of the prime directives of the Toyota
Prius project. Average development time for design and implementation of a new automobile at
Toyota was four years. The Prius team delivered in 15 months.
Aggressive goals put pressure on the system and processes must change. Engineering practices
must improve. Software build processes must be automated. Testing must be started at the
beginning as it cannot wait until development is done. Product reporting must be automated and
visible to all in the company. A long list of impediments to progress emerge when cycle time is
compressed. In Scrum, impediments are prioritized by the ScrumMaster and systematically
eliminated with the assistance of the management team.
In Team Scrum, the focus is on team performance. Continuous Flow Scrum focuses on Program
Management performance and product release (a combination of management and Product
Owner activities). Continuous Flow Scrum does not exist if the company as a whole ignores (or
is incapable of resolving) prioritized impediments raised by the ScrumMasters. The Program

136
Management team must take responsibility for dealing with tough issues embedded in company
cultures that sabotage performance.
Management Challenges for a Continuous Flow Scrum

The Japanese use the terms Muda, Muri, and Mura when applying lean manufacturing principles.
Muda is waste. All waste, including every step that does not add customer value, must be
eliminated. Muri is overburdening people, a process, or a system. Slack time must be introduced
into a system to eliminate bottlenecks. Mura is unevenness or undue variation in process or
product. Process flow must be smoothed out. These practices follow directly from queuing
theory.
Many companies find it difficult to eliminate nonproductive assets, outmoded processes, or

unnecessary hierarchy. Bloated artifacts, heavyweight project plans, and people who are not
adding direct customer value need to be moved out of the way. The root of these problems [92]
derives from the accounting system introduced by Sloan in the 1920’s to measure success. Cost
Accounting drives device optimization at the local level which suboptimizes global performance.
Higher utilization of resources (people and machines) becomes an end in itself leading to higher
production of inventory, rather than optimizing for sold products. Management forces people to
work harder and harder to produce less and less
The Toyota process minimizes inventory and assumes process change occurs on a daily basis
and these changes may affect the entire company. The Director of the Toyota Training Institute
points out that the practice of “Kaizen mind” assumes that workers will become better at what
they do every day [93]. Failure to do so is a corporate crisis and corporate cultures that are
impervious to change cannot implement lean manufacturing practices.
Muri and Mura are the most insidious and difficult process burdens to eliminate. Old style
management tries to load up a development team 110%. This makes them run slower, just as
trying to run a computer process over 80% starts to slow it down, eventually causing the system
to hang up. This Muri overload causes Mura (disruptions in flow) by generating process
bottlenecks.
The Japanese have moved away from push type processes to pull type processes. When a worker
needs to start the next step in a process, requirements and resources are provided just in time.
The worker “pulls” these requirements and resources only when needed, allowing workflow to
move at a steady pace. Scrum is designed to take advantage of lean manufacturing processes and
a Continuous Flow Scrum can more than double productivity of Team Scrum.

137
Summary
During 2002-2007, the Scrum community of practice produced over 12000 Certified
ScrumMasters in an introductory course. While these are only ScrumMasters in training, some of
them have been able to transform entire companies. For example, Yahoo! has gone from one
Certified ScrumMaster in 2005 to over 100 Scrum teams in 2007. For each Certified
ScrumMaster, there appear to over 10 Scrum teams active on projects. These tens of thousands
of projects are providing feedback to the Scrum Community through
[email protected], biannual Scrum Gatherings, and an increasing number
of research and experience publications.
Most of the leading software development companies worldwide have adopted Scrum as a
development processes [94]. Many of those who have successfully executed Team Scrums for a
year or more are ready to move to a Continuous Flow Scrum. This requires a corporate culture
change to eliminate all processes that do not add immediate customer value. Management has to
move to a facilitative leadership style and break down traditional hierarchies.
The Toyota consulting group recently selected the most Agile company it could find in the
United States. In six months they improved productivity by 80%. Scrum has repeatedly
demonstrated it can do the same or better. In 2006, a CMMI Level 5 company introduced Scrum
as a process improvement intiative. Defects were reduced 40%, planning costs were reduced
80%, and overall productivity was up 100% [95]. They now bid Scrum projects at half the cost
of waterfall projects. Winners may put their competition out of business just as Toyota is taking
the lead away from General Motors.
Postscript
This paper provided a brief overview of a Continuous Flow Scrum when it was originally
published. A more profound insight into these issues can be found through study of set-based
concurrent engineering [57] and the theory of constraints [96].
When this paper was originally written for the Cutter Agile Project Management Advisory
Service, the editors asked that it be watered down. They did not want me to state that Toyota
would overtake General Motors. It had forcefully made the statement that Continous Flow Scrum
is designed to create the Toyota effect of four times productivity and 12 times better quality than
competitors. This level of performance will put the competition out of business. It is inevitable.
The only question is one of timing and one only has to look at Toyota’s impact on General
Motors. Little did I know how quickly that problem would become of immediate importance.
General Motors in recent years has improved the performance of employees such that the time
spent on adding real value to the customer has moved from a traditional 20% up close to
Toyota’s 80%. Nevertheless, as Forbes columnist Marilyn Cohen observes:

138
“The domestic automakers are in a stew, with General Motors in the deepest. Nevertheless, GM,
the largest company in what seems to be a dying U.S. industry, may get a new life. At least that’s
the possibility held out ... (for) an alliance between GM’s Chief Richard Wagoner, and
Renault/Nissan Motor’s, Carlos Ghosn… Then long-suffering GM bondholders won’t be
exposed to a Chapter 11 filing, which is the fate many investors mentally assigned the company
not long ago [97].”
Such is the Toyota Way side-effect. Of even more interest, the strategic target for Toyota and
some other Japanese car manufacturers is not General Motors, it is to eliminate the internal
combustion engine. The lack of Agility and the ability to innovate in an environment where
leaders are totally changing the playing field will ultimately force the creative destruction and
acquisition of assets of the lame and the weak.

139
Type C All-At-Once Scrum: Creating a Scrum Company
Patientkeeper, Inc., Newton, MA, US
“This is a very important paper! It lays out a masterful series of business process innovations
that desperately need to be emulated by most organizations.“ Tom Poppendieck
Abstract
Scrum was invented to rapidly drive innovative new product to market. Six month releases used
to be a reasonable time from for an enterprise system. Now it is three months for a major new
release, one month for upgrades, and one week for maintenance releases. The Scrum
development process was designed to enhance productivity and reduce time to market for new
product. In this paper, one of the inventors of Scrum goes back to Scrum basics and designs All-
At-Once Scrum using multiple overlapping Sprints within the same Scrum teams. This
methodology delivers increasing application functionality to market at a pace that overwhelms
competitors. To capture dominant market share requires senior management participation in a
MetaScrum for release planning, variable length Sprints, overlapping Sprints for a single team,
pre-staging Product Backlog, daily Scrum of Scrums meetings, and automation and integration
of Product Backlog and Sprint Backlog with real-time reporting. A practical example of All-At-
Once Scrum describes how mobile/wireless product teams implemented Scrum process
automation beginning in 2000 and achieved hyperproductive revenue growth by displacing
dominant vendors in 2007. Administrative overhead for over 45 enterprise product releases a
year is less than 60 seconds a day per developer and less than 10 minutes a day for a Project
Manager. While All-At-Once Scrum is not for beginners, this professional implementation of
Scrum companywide is faster, better, and cooler.
1. Scrum Evolution
Evolution occurs in dynamic response to environmental demands. Now that the Scrum
community has 15000 Certified ScrumMasters and hundreds of thousands of projects under their
belt, retrospection can help guide future activities. In particular, what did you do yesterday that
worked (Scrum theory), what makes sense to do tomorrow (Scrum evolution), and what is
blocking the way (Scrum dogma) is worthy of analysis.
One of the key influences sparking the creation of the Scrum Agile development process was a
Harvard Business Review paper on Japanese new product development by Takeuchi and Nonaka
[1]. A key component of their presentation was a chart showing product development separated
into silo’s (Type A), phases slightly overlapped (Type B), and all phases of development
overlapping (Type C). The Japanese viewed Type A product development as an outmoded relay
race type of process. Type B they thought was similar to Sashimi because slices of work
140
overlapped requiring collaboration between phases. Type C they envisioned as Scrum where all
phases of product development were done simultaneously. Scrum is a Rugby formation and they
viewed an “all-at-once” process as similar to a Rugby team moving down the field passing the
ball back and forth to one another.
Type A – Isolated cycles of work
Type B – Overlapping iterations
Type C – All at once
Figure 13: Type A, B, and C strategies for delivering product [11].
After discussing the notion of various types of Scrum with Certified ScrumMasters in Scrum
Alliance workshops and with development teams at Google, Microsoft, Yahoo, Arriba, Adobe,
GE Healthcare, and other companies, it appeared that the chart above can be applied to a higher
level of thinking about three styles of Scrum implementation.
In early Scrum implementations (called here a Team Scrum), all development occurs within the
timebox of a Scrum iteration called a Sprint. A side effect of this approach is downtime between
iterations when reorganizing for the next Sprint. Well executed Sprints can double productivity
and repeatedly deliver projects on time, within budget, with functionality precisely targeted to
end-user demands.
By adding product definition tasks for the next Sprint into the current Sprint, a Continuous Flow
Scrum allows work to flow smoothly from Sprint to Sprint. Product backlog requirements for the
next Sprint are developed in the current Sprint. While this requires that a development team
allocate a small amount of time from the current Sprint to help the Product Owner estimate the
Product Backlog for subsequent Sprints, it can enable high performance development
organizations to deliver more working product than sales, marketing, or customers can absorb.
By eliminated the development bottleneck the company can adopt new strategies and create new
products that were previously impossible to deliver. Most companies today have to implement a
Continuous Flow Scrum out of necessity to deliver a continous set of product portfolio releases
141
to the market. However, high performance of a Continuous Flow Scrum requires rigorous
implementation of lean practices.
All-At-Once Scrum can be envisioned as pipelining Sprints by running multiple overlapping

Sprints through the same set of Scrum teams. This requires experienced Scrum teams, well
designed product architecture, and automation of Product and Sprint Backlogs. Throughput can
be enhanced to deliver dozens of new releases of enterprise software annually. A MetaScrum is
implemented to create Agile product release strategies. On a weekly basis, companies can alter
their direction to deal with competitive threats and market changes. Competitors can be
overwhelmed and market dominance achieved.
Takeuchi and Nonaka observed collapsing phases of product development improved innovation,
throughput, time to market, and product acceptance. As market pressures have evolved and
changed, it is possible to collapse Scrum Sprints to create more business opportunity. Market
domination is the goal of the All-At-Once Scrum.
2. Scrum Evolution in Practice

The evolution of Scrum in five companies from 1993-2001 has been described previously [17,
40]. Here we focus on continued evolution of Scrum theory using PatientKeeper, Inc., as a test
bed. During 2001-2005 we automated a solution for a Continuous Flow Scrum. This eliminated
lost time and productivity between Sprints and, as observed previously at Easel Corporation in
1994, significantly increased throughput compared to completing work only within the Sprint
time box for which it is defined.
In addition, we solved the problem of multiple projects pipelined through the same team (or set
of teams) and have been running an All-At-Once Scrum for over six years. This required careful
automation of the sprint backlog with improved tools and metrics in order to maintain team
focus. Daily build processes and automated regression testing was significantly enhanced. Our
approach to Quality Assurance (QA) was modified to provide a small QA team for each of four
to six overlapping production software releases. Pair programming was used selectively and
team programming was common where many programmers worked together around a table for
the entire day.
The result has been delivery of production code to a new set of enterprise customers for every
Sprint with maintenance Sprints weekly, customer enhancement Sprints monthly, and new
application releases quarterly. By 2004 more than 45 enterprise level releases of PatientKeeper
production software were completed, installed, and brought live at customer sites. Many of
PatientKeeper’s customers are large multi-hospital systems like Partners (Massachusetts General
and Brigham and Womens Hospitals) in Boston, Johns Hopkins in Baltimore, and Duke
University Health System in North Carolina.
Recently, PatientKeeper broke a record in deployment time. For HCA, the largest hospital
system in the U.S., PatientKeeper brought 12 hospital deployments into full production in seven
142
months with new physician and administrative desktop and PDA applications. This was actually
slow by PatientKeeper standards, yet was a historical record for HCA.
These large clients provide an excellent test bed for scalability of an All-At-Once Scrum. They
require a high level of product adoption by difficult and discriminating physician users, support
for large multi-institution wireless networks, integration with many clinical and financial systems
at sites with diverse IT infrastructures, and thorough testing and certification by the customer.
2. The First Scrums – Team Scrum and Continuous Flow Scrum

Early Team Scrums were useful for education and training on the pace of Scrum and particularly
suited to new Scrum teams. However, it creates a loss of time between Sprints where the team is
reorganizing for the next Sprint and did not adequately address enterprise scaling issues.
At Easel Corporation in 1993 we initially applied Team Scrum to software development teams
when we built the first object-oriented design and analysis (OOAD) tool that incorporated round-
trip engineering from design to code and back again In a Smalltalk development environment
[21]. Code was generated from a graphic design tool and any changes to the code from the
Smalltalk integrated development environment (IDE) were immediately reflected back into
design. There were six Sprints for the first product release and the gap between Sprints took at
least a week and sometimes two weeks. As a result, we could only do 9 Sprints a year, losing
25% of our productivity as compared to potentially running 12 Sprints per year. This was viewed
as a serious impediment by the Product Owner and management.
This loss of time needed to be removed because survival of the company depended on delivery
of an innovative product as early to market as possible. Each month of delay cost millions of
dollars of lost revenue and gave the competition the opportunity to overtake us. For example, the
Product Manager at Rational Rose was regularly showing us demos of support for roundtrip
engineering as we were struggling to be first to market with the first roundtrip engineering
object-oriented design tool.
In addition to loss of productivity between Sprints in a Team Scrum, it took time during the
Sprint for the developers to get enough clarity about the user requirements to start coding. It may
be halfway through a Sprint before the developers understand the user experience well enough to
implement a solution. This creates tension between the Product Owner and the Scrum Team
concerning lack of understanding of what to do next, substantial slippage of features into
subsequent Sprints, and dissatisfaction on the part of the Product Owner with delays in feature
delivery. This churning phenomena can cut Sprint productivity in half, a huge impediment that
needs to be remedied.
The need to start development with adequate functional specifications was observed by
MacCormack [98] when he gathered extensive data on 29 Hewlett Packard software projects to
assess development practices. One of the strongest productivity enhancers noted in his
correlation analysis was completeness of the functional specification. While Agile specifications
143
are designed to be just enough and no more, a product specification needs to be “ready” before it
can be allowed into a Sprint. MacCormack showed the definition of “ready” has major
performance implications.
Regarding the use of specifications, there was a significant relationship between the
completeness of the functional specification and productivity. There was a weak relationship
between the completeness of the detailed design specification and defect rate (p = 0.078). The
former result suggests that developers are more productive to the degree that a complete
functional specification exists prior to coding. This is intuitive, given that the functional
specification outlines the features that developers must complete. To the degree that these are
stated up front, developers can focus solely on “executing” these features in code.
Agile developers use a minimum amount of documentation and do not require completeness of
the specification to start a Scrum. McCormack found that completeness of the design
specification was not correlated with enhanced productivity and only slightly reduced the defect
rate which is consistent with Agile thinking. However, he found a strong correlation between
adequate product specifications and productivity. This suggests that minimal functional
specifications should be clear at the beginning of a Sprint and that design and technical
specifications are best done within a Sprint.
MacCormack’s multivariate analysis showed three primary factors that lowered defect rate (early
prototype, design reviews, and integration or regression testing at code checkin) and two primary
factors that increased productivity (early prototype and daily builds). Releasing a prototype to
customers that is only 40% functionally complete increases productivity by 36% and adopting
the practice of daily builds increases productivity by 93%. These were clearly the most effective
Agile practices in the Hewlett Packard projects.
Incremental and early delivery of working software is at the core of the effectiveness of Agile
processes. In addition, a functional specification that is complete enough for the next interation
to allow developers to begin work without false starts will enhance feature delivery within
Sprints and improve throughput. Despite the fact that the implementation phase is a small part of
the overall cost of a software project, the biggest resource bottleneck on a software project
typically occurs with a shortage of expert developers whose skills are not easily transferable.
Constraint analysis shows mathematically that the biggest bottlenecks should be eliminated first
[96] (just as in tuning of a computer system) and early delivery of a functional specification for a
single increment helps eliminate the critical development resource bottleneck.
While a Continuous Flow Scrum can improve productivity with experienced Scrum teams, a
Team Scrum with intervals between Sprints to assure the Product Backlog is “ready” may be the
best way for a company to pilot Scrum, even though it may not be most efficient. It allows
systematic application of the Scrum process with enough time to refine operations and regroup
between Sprints. It also forces all-at-once type thinking when everything has to happen for a
specific Sprint within the time box of that Sprint. Initially, the benefits in training may

144
overwhelm the lost productivity. Without the ability to execute a Team Scrum well, it is not
possible to effectively implement a more sophisticated process.
The benefits of Team Scrum are:

• Total focus on iteration in process
• Ease of implementation
• Developing and understanding the pace of Scrum
• Clearly defined iterations
The problems with Team Scrum were:

• Loss of time to market
• Disruption of pace of Scrum because of delays in understanding of the user experience
• Loss of productivity (and market share) due to resulting delays
Scrum has a primary focus on an impediment list managed by the ScrumMaster. By prioritizing
this list which includes personal, team, and company issues, the ScrumMaster puts an intense
focus on improving the Scrum implementation. When removing impediments most companies
will find they need to go to a Continous Flow Scrum to maximize throughput.
3. Continuous Flow Scrum

The way to overcome loss of time to market with a Team Scrum is to insert tasks in a current
Sprint that prestage work for a subsequent Sprint. A minimal specification of the user experience
for a feature is defined and estimated prior to the Sprint where it is implemented. This allows
Sprints to be executed continuously with the Sprint Backlog always full at the beginning of each
Sprint. At the same time, it requires that a Scrum Team allocate resources to help the Product
Owner estimate features for subsequent Sprints during the current Sprint.
A caveat is that Continuous Flow Scrum will not work in a company that has not implemented a
sustainable development policy and process. That means that Scrum teams decide on what tasks
can be implemented in a Sprint and who will implement them using a normal work week as the
standard way to do business. Many companies using Scrum still have management trying to jam
more work into increments than Scrum teams can deliver in an allotted time. This results in lack
of team autonomy, excessive overtime, high defect rates, personnel burnout, and high employee
turnover. This violates a fundamental principle of lean product develop and makes it impossible
for a team to enter the high performance state for which Scrum was designed.
The key indicators that Scrum is working must be visible in a Team Scrum before moving to
Continous Flow Scrum:
• Team autonomy – the Scrum team is (and feels) totally responsible for their product and
no outside agency impacts the workplan of the team inside a Sprint.

145
• The Product Owner is part of the Scrum and affects product design and implementation
within a Sprint without disrupting self-organization.
• Self-transcendence – individuals move beyond self-gratification to focus on team
performance.
• Cross-fertilization – expertise is regularly shared across team members and no single
person is a bottleneck.
Fully loading the development queue in a Type B Scrum at all times without building a
sustainable pace of development will negatively impact morale. On complex development
projects, it typically takes a new engineer six months to come up to full productivity. If turnover
is 20%, you lose one quarter in hiring a new development and two quarters training them. Your
development team productivity is down 15% from this alone. This personnel churn can cause
development tasks to stop and start as specialized resources must be shifted to complete them,
reducing productivity by another 15%. If morale drives the pace of development down further,
you may cut productivity in half with Type B Sprints that are implemented too early.
Conversely, if Type A Sprints are running well, pre-staging functional specifications in the right
way in a Type B Scrum will eliminate churn within a Sprint and downtime between Sprints. This
has doubled productivity for experienced Scrum teams. In companies seeking to expand market
share and dominant a market segment, this advantage is absolutely compelling.
In several companies using Scrum, management and staff stated that maximizing development
throughput was not a priority. These companies invariably were having delivery problems and
were outsourcing major pieces of their development. Outsourcing was not solving their delivery
problems and in many cases was aggravating it. In the long run, this last to market approach is a
losing strategy. If companies are not continually getting better, failure is just a matter of time as
the competition is always improving.
3.1 Staging Functional Specifications for a Type B Sprint
A Type B Scrum accelerates throughput by keeping the Sprint backlog full and at times. This
requires prestaging functional specifications of the product prior to the start of a Sprint.
Maintaining the agility of the Scrum process requires a minimalist approach to functional
specifications which is just enough, and no less than just enough.
A minimal amount of documentation for a product feature is typically a few pages and definitely
not hundreds of pages. Just enough documentation so that engineers understand the user
experience will suffice. This typically means screen shots, data requirements, workflow from
screen to screen, and business logic that must be executed. The minimum documentation
required to achieve Jacobsen’s overview of an object-oriented analysis of the problem [99] is an
excellent guideline even though it may be beyond the capability of some product management
organizations. Fortunately, PatientKeeper had well educated physicians that served as Product
Owners. While some of them had no formal training in software development, they learned
quickly how to elaborate use cases in a way that defined the user experience for the physician
146
when using PatientKeeper products. In addition, these physicians where action oriented and
strongly resistant to analysis paralysis. They avoided time spent on excess documentation,
making them excellent Agile developers by inclination.
Moving to a Type B Scrum requires analysis and design resources from the development team in
order to help the Product Owner create functional specifications and pre-stage the Sprint backlog
for the next sprint. In the worst case, this might require 25% of the Scrum resources during a
sprint. However, it avoids the 25% lag time between sprints. So in the worst case you may break
even on resource allocation.
The real gain from a Type B Scrum is having the Sprint backlog fully loaded at all times. A
developer never wonders what to do next because the queue is always full. If the Sprint backlog
is automated, team members simply logon at the beginning of the day and self manage the queue
of work in front of them on a web page. The Product Owner and ScrumMaster are continuously
working to assign items to a developer’s queue and the developer decides how to order the work
or, in some cases, will reassign it to a more appropriate team member. This radically increases
throughput during a Sprint, often doubling it.
A relevant analogy is water flow through a garden hose. If the faucet is turned on and off, you
disrupt flow and generate pressure surges in the line. These cause side effects and sometimes
structural breakdown in water lines feeding the hose. Any structural breakdown will reduce
productivity to zero for an extended period. Keeping the faucet turned on, even with reduced
flow, may generate more throughput without negatively impacting upstream systems.
At the time of the PatientKeeper first major round of venture funding in 2000, I asked Ken
Schwaber to help get a Type B Scrum launched immediately. Product Management owned the
products and they were required to define the user experience of a feature before it could enter a
Sprint backlog. More specifically, because of the demanding requirements of a physician user,
the screen shots, the logic, the workflow between screens, and the data items required had to be
defined. In addition, a prototype had to be tested with physician users and validated so that we
knew conclusively that (1) the functionality was needed, and (2) the physicians would use it
when it was implemented.
The requirement that a Product Owner provide a functional specification sufficient to clarify the
user experience creates a positive dynamic tension between the Product Owners and the Scrum
teams. The Product Owner cannot get a feature into the queue for development unless it is
defined enough for the Scrum team to begin work immediately, either by building a technical
design document or coding directly from the Product Management specification when possible.
At the same time, we gave the Product Owner resources in any Sprint to create a prototype of a
new feature using very rapid mockup tools that would create screen interactions on a mobile
device. In addition, the Product Owner had complete control over moving items in and out of the
Sprint Backlog before and during the Sprint. This puts the Product Owner in the driver’s seat.

147
This is effective only if the Product Owner knows how to get to the destination, i.e. the right
product specification is ready for the Scrum team to implement at the right time.
By holding Project Owners responsible for defining the user experience for a feature prior to the
start of the Sprint, a rigorous process had to be introduced into Product Marketing at
PatientKeeper. The process made it crystal clear that building new product began only when the
Product Manager’s job was done. At the same time, it was important to always have the
development queue loaded. Management insisted that the development team not have downtime.
The developers self-managed their work queue and the work queue was always full or a
company emergency was declared, similar to the build process breaking. The only way Product
Marketing could get something on the queue is was to complete their minimalist functional
specification, just enough information in just the right format at just the right time.
A positive dynamic tension was created because the Product Owner (in this case, Product
Marketing) always wants more product in a release and tries to jam features into the
development queue. Developers always have more work than they have available time. In a Type
B Scrum, it does not matter whether Product Marketing introduces new features into the queue in
time or not, development productivity is not impeded. They just work on what is in the queue.
Management is happy because they are always seeing features roll out. If the mix of features is
off, they hold the Product Owner responsible when the required functional specifications were
not ready.
3.2 Product Owner as Part of the Scrum Team
The original Japanese view of a product development Scrum created a team that was totally
responsible for the product [1]. In some companies, such as Individual in 1996, the Product
Owner was at every Scrum meeting. In others, like the original Scrums at Easel Corporation in
1993-94, the Product Owner was on the road much of the week and was always at the Friday
Scrum meetings [17, 40].
The Product Owner owns the business plan for the product, the functional specification for the
product, the product backlog for the product, and prioritization of the product backlog. As a
member of the Scrum s/he works side by side with the ScrumMaster to introduce product
backlog items into a Sprint where they are broken down into tasks by the team for execution as
Sprint backlog. At PatientKeeper, the Product Owner manages the movement of tasks in and out
of the Sprint backlog in consultation with the ScrumMaster.
The linkage can be very tight between Product Owner and ScrumMaster with highly skilled
people. For example, at PatientKeeper, the mobile device development team leader is the lead
designer and coder on the team, the ScrumMaster, and one of three Product Owners in the
company, reporting to both the VP of Marketing and the Director of Engineering. The two other
Product Owners are responsible for clinical applications and financial applications on handheld
and web devices. For clinical applications, the clinical Product Owner and the mobile device
Product Owner are joined at the hip with the ScrumMasters. Together, they are totally
148
responsible for the business plan, the product specification, and working day to day with
engineers on product design.
After working as head of engineering in nine companies, I have found that the best way to think
about Scrum responsibilities is to think of Scrum team as analogous to a high performance car in
a rally race. The Product Owner is the navigator and the ScrumMaster is the driver. The team is
the engine, the chassis, the drive train, and the wheels. The ScrumMaster follows the
navigational directions of the Product Owner precisely and drives the car adroitly. The car and its
occupants are totally responsible for winning the race. At the end of every Sprint, other players
move in and can make modifications to improve the timing of the next Sprint.
New ScrumMasters tend to view this analogy as too prescriptive to a team that assigns its own
responsibilities and self-organizes. Consider a football team. It self-organizes under the Coach
where those best suited to roles, assume positions of quarterback, center, tight ends, and so forth.
In the huddle, they may quickly make minor refinements to individual roles and responsibilities.
However, when the ball is hiked, there is no discussion of what anyone is supposed to do. To be
successful, they must know what they are to do and execute it quickly and professionally.
In some companies new to Scrum, engineers have claimed no one is responsible because there is
not a “project manager.” If you look at project managers in technically driven companies, they
are usually at the mercy of the technical team. As a result, product management, and therefore
product ownership, is weak. This compromises the effectiveness of Scrum and prevents a Scrum
team from entering a hyperproductive state [25].
In a well run Scrum, particularly a Type B or C Scrum, the ScrumMaster must be able to drive
the race car and the Product Owner must be able to direct the ScrumMaster to the destination in
the timing necessary to win market share and attain profitability. Failure at either of these tasks
leads to replacement of the appropriate person. Success means one or both go on to greater
responsibilities. For those who lead hyperproductive Scrums, career advancement is rapid and
they usually wind up as CTOs, CEOs of their own companies, or VPs of Engineering of larger
companies within a few years. The ScrumMaster of the IDX Web Team left the Scrum team to
lead the U.K. division of IDX and closed over $2B of new business within 3 years of leaving the
Scrum. This is a example of a great ScrumMaster who learned how to own the business as well
as the technology, side by side with the Product Owner.
3.3 Type B Scrum Can Enable a Hyperproductive Team
Giving the Product Owner control of the Sprint backlog along with strong accountability builds
strong product managers. It also conditions the development team to move rapidly towards the
goal without analysis paralysis. A combination of dynamic forces wins a race when a forceful
driver is coupled to a high performance sports car or a high spirited stallion. The same
phenomenon happens on sports teams when everyone understands the plays and can execute
them immediately on demand. It allows the team to move up to a higher level of play where the
basic moves are on autopilot and superlative moves are possible. The first Scrum began
149
executing Type B Scrum as they mastered the process. They were able to enter the “zone” using
this technique, where they could deliver functionality faster than the customers, the marketing
team, or sales could absorb product. The feeling of power in a development team that can deliver
more product than anyone can absorb is exhilarating.
Scrum was designed for this hyperproductive state, to get ordinary developers to function as a
champion team. It only happens to about 10% of Scrums and it only starts to happen when the
organization moves to a Type B Scrum. The doubling of throughput from a team that is already
very productive results in an organizational breakthrough.
4. Type C Scrum
Scrum is an organizational pattern that is designed for an activity that is difficult to control
because its predictability is limited [25]. It is useful in any context where the activity requires
constant change in direction, unforeseen interaction with many participants, and the need to add
new tasks as work unfolds. These factors were amplified at PatientKeeper when it received a
$50M round of venture funding in 2000.
A decision was made to become a platform as well as application company with a software
framework and open application programming interfaces (APIs) that would allow integration
with development partners on both back end servers and the front-end devices. A web services
platform architecture was selected.
PDA – Web Portal - External Web Services - Telemedicine
Web Services
Bus Web Server Process Workflow Rules Logistics
Manager Engine Engine Engine
Architecture Application
Server Cache Cache Cache Cache
Workflow/
Logistics Middle tier
Management Database
MER
Internal Web Services
Adapter Adapter Adapter

Execution Interface
Tier Engine
Custom CDR Cerner Departmental

Meditech Systems
…
We Make Healthcare MobileTM

150
Figure 14: Patientkeeper platform architecture [100].
In addition to the server/network architecture based on Java and XML, a cross platform software
framework on Palm and Pocket PC handheld devices was implemented in C/C++. This
framework provided open APIs and a software development kit that allowed third party vendors
and end users to tightly integrate their mobile applications with other applications already
available on a handheld device.
The tight integration between software components required tight integration of software
development teams internally at PatientKeeper and externally with partners and offshore
developers. This, combined with time to market pressure and rapid growth of new client
deployments in large enterprises, each demanding new increments of functionality along with
30-90 day installations, forced a new type of Scrum to be implemented at PatientKeeper.
4.1 Case Study Context
PatientKeeper builds a software platform that takes information from many clinical systems
across multiple hospitals and clinics and presents it on an intuitive user interface to physicians
using handheld devices and the web. The application software has a four tier architecture with
four levels of data cache:
• Primary data source is a clinical data repository

• Data is forward cached in a mini-clinical data repository
• In-memory cache exists on a middle-ware server to improve performance
• On a handheld device, the data is stored locally
Software and data must be consistent across four tiers at all times. This forced PatientKeeper to
go to totally integrated builds at all times to assure that software in all four tiers of the
architecture worked consistently. Software testing has to validate that version of all architectural
layers work together to provide consistent data to the end user.
The developer team working on this product was split into a backend integration team, a clinical
repository team, a middleware server team, two PDA teams (Palm and Pocket PC) and a Web
team. It was necessary to tightly couple these teams together to assure that all software was
interoperable all the time.
4.2 Case Study Market Requirements
As an early-stage, venture funded company, PatientKeeper had to establish a new product

offering in the rapidly growing mobile/wireless market. Early customers had to be implemented
as quickly as possible with limited functionality. Subsequent customers needed to be installed as
soon as possible with upgraded functionality. The imperative was to gain market share and
151
achieve market dominance in a highly competitive environment. Speed to market needed to be
used as a strategic weapon. Company viability and success demanded it.
The customer base rapidly evolved to 5-10 hospital systems to be installed each quarter. Each
group of hospitals needed more extensive functionality in a rapidly growing portfolio of
applications that included business partners with integrated back end clinical systems, portal
vendors, and handheld device application vendors. A major release with new applications was
required on a quarterly basis.
Customers consistent of large hospital systems with multiple installations, academic research
institutions with integration into home-grown healthcare research and treatment systems, and
medium size community hospitals with different requirements. The quarterly major release had
to be supplemented with monthly minor releases to allow multiple new installs of similar clients
to take place on a monthly basis. Finally, bugs and unanticipated implementation issues that had
to be resolved to go live at new sites required maintenance releases every week or two.
4.3 Case Study Problem
The challenge for PatientKeeper quickly became how to simultaneously do weekly, monthly, and
quarterly releases of a system that was tightly integrated across four architectural layers with six
software development teams that needed to be tightly coupled to assure data and application
consistency across multiple back ends, diverse wireless networks, and multiple front end devices.
Furthermore, each release had to be tested and certified across all levels of architecture and
applications for deployment.
PatientKeeper started up as a Scrum company doing daily meetings. Type B Scrum had been
implemented from the beginning with strong product owners required to produce functional
specifications before any product backlog item could be transformed into tasks for a sprint
backlog to be implemented in the next iteration. The challenge was to a automate project
management to monitor thousands of tasks across dozens of releases a year without disrupting
the Scrum process.
4.4 Case Study Forces
Resource constraints forced every developer to be focused 100% on building the system.
ScrumMasters and team leaders spent the majority of their time designing and coding the system.
Separate project leaders were not an option.
High caliber developers, many with doctoral degrees, did not want excessive administrative
overhead. They felt that project management could be automated and taken to a new level of
efficiency. The CTO of PatientKeeper was asked by the Scrum teams to organize a project
management system that required less than 60 seconds per day of administrative time per
developer and less than 10 minutes per day for a ScrumMaster to provide comprehensive
reporting to management, the development team, and other areas of the company.
152
• Estimation was important. How were developers going to provide valid estimates and
update them in less than sixty seconds a day?
• Planning and prioritizing takes time. How was this going to be accomplished without
impeding development throughput?
• Architecture was critical for a platform company. How was it going to evolve using the
Scrum process to provide flexibility, scalability, performance, reliability, and
maintainability?
• Customer requirements in the form of use cases that could be rapidly transformed into
deliverable code were essential. Who was going to do them and how would they be
delivered?
4.5 Type C Scrum Solution
The Type C Scrum solution required several innovations that affected all parts of the company.
In effect, the company had to become a Scrum company with all activities driven by an
automated data system that reflected release planning and Sprint execution, as well as installation
and support team and customer feedback.
• Team organization was changed

• Build process became more highly automated
• Regression testing automation improved significantly
• All data collection had to be automated
• New tools for data collection and reported had to be developed
• A MetaScrum had to be created to allow company leadership to manage multiple
simultaneous product releases
• New reports had to be developed
• The company had to become totally transparent. All data was available to everyone in
real time all the time.
4.5.1 Team Organization

153
Figure 15: Open space for Type C Scrum of Scrums daily meeting
The daily Scrum meeting quickly evolved into a daily Scrum of Scrum meetings. All members of
the development team are present for 15 minutes meetings. Team leaders do most of the
reporting:
• What did each of the six integrated teams complete in the last 24 hours? The Scrum of
Scrums leader logs what tasks were completed and sends out an email to the company
immediately following the Scrum of Scrums.
• What blocks were found in performing tasks in the last 24 hours. These are logged,
reported, and followed-up after the meeting.
• What will teams work on in the next day. Team members volunteer for tasks. The
ScrumMaster and the Lead Architect may help bring focus to appropriate tasks.

154
Typical Day in a Type C Scrum
Scrum Master email at close of Scrum daily meeting
Friday Releases 19 Nov 2004
245g5
– getting feedback from Cerner,
– they're trying to get micro susceptibilities data into the test system
– added MAR suppression to address issue at SOM
245m
– upgrade testing this morning, should release by noon
246
– 246g1 palm released with timeout issue fixed
– 246i - post t-giving
251b2
– SUNY patched released last night / installed into SUNY test system
251d
– Mt Sinai release, should release by noon
251e
– Monaco clinicals, targeting Alverno
3.0.1 102 open PRs, 57 verification (down from 110 on Monday!)
– beta release today
© Jeff Sutherland and ADM 2004
Figure 16: Summary of daily email after Scrum of Scrums meeting shows seven releases in progress
simultaneously. All teams work on all releases and all releases result in customer deployment.
The Scrum of Scrums meeting takes place at the same time and place every day. An open space
was secured by the development team for this purpose. Pair programming was done primarily on
tasks with difficult design and coding requirements. Many of the developers stayed in the open
meeting space for the entire day working together as a group. Innovative and open cube space
and a few offices and conference rooms are provided for those who need quiet, focused time.
The rapid pace of delivery of production code releases initially created a Quality Assurance (QA)
bottleneck. The solution was to assign an small QA team to every release. QA was expanded to
four small teams of 2-4 people. This enabled them to work continuously on four of the top
priority releases. In the Figure above, where six releases are being simultaneously developed,
QA is doing final release testing and packaging on four of them. QA is part of the Scrum of
Scrums and reports on daily status of ongoing releases.
4.5.2 Data Collection
A user group study and focus group analysis was performed for data collection for tasks,
estimates, and updates that would be used to automate the standard Scrum burndown charts [17].
A wide variety of Scrum tracking tools had been used by members of the team in various
companies over a 15 year period, none of them considered adequate. The 60 second requirement
for data entry implied that a new application would not be possible, because simply starting up a
new application might require 60 seconds.
155
The best application to use was one that developers had to use every day, the bug tracking
system. In addition, the speed at which developers could do data entry was dependent on the
questions they were asked, and the order in which they were asked. It was determined that only
three questions would be asked as developers could answer them without thinking, they could
give a gut level response:
• What is the initial estimate for this task if it is a new task?
• At this moment, how much time have you spent on this task?
• At this moment, what percent complete is this task?
These were the only additional data items to be collected daily from developers for tasks. All
other data analysis and reporting was to be automated.
4.5.3 Tools for Data Collection and Reporting
PatientKeeper used the open source GNATS bug tracking system [101]. Since developers needed
to use the bug tracking system daily, there was no additional time overhead for opening the
application to enter task data.
GNU GNATS is a set of tools for tracking bugs reported by users to a central site. It allows
problem report management and communication with users via various means. GNATS stores all
the information about problem reports in its databases and provides tools for querying, editing,
and maintenance of the databases.
Thanks to its architecture, GNATS is not bound to a single user interface – it can be used via
command line, e-mail, Emacs, or a network daemon, but is usually used with a Web interface.
Together with the fact that all GNATS databases and configuration can be stored in plain text
files, it allows easy use and provides good flexibility. Basically, if the GNATS tools do not
provide everything you need, you can add your own additional utilities using standard GNU
tools. http://www.gnu.org/software/gnats/
A PERL expert on the development team was assigned to build utilities around GNATS to
support Scrum. These were addition of required data items, new queries, minor changes to the
user interface, and automated file dumps for management reporting via Excel. Sample data items
maintained by GNATS are shown in Figure 3 below.

156
Figure 17: Typical data items in GNATS for problem reporting by email or the web.
It was decided that sprint tasks would be treated like problem reports. This minimized new data
entry requirements and allow tasks and bugs to be packaged together seamlessly for a release.
Only three data items were added to GNATS for developer entry:
• Initial estimate
• Days invested
• % complete
The initial estimate was fixed at initial entry and could never be changed in order to allow for
accurate historical reporting of estimates versus actual time to complete tasks. Two additional
data items were added for reporting purposes. These are automatically calculated from the three
items above.
• Days remaining
• Actual time to complete
If the initial estimate is 2 days, for example, and no work has been accomplished, the days
remaining are 2 days. If a developer has invested 1 day and states that it is 25% complete,
157
GNATS calculated the days remaining as 3 days. Initial estimates are automatically expanded
based on real time data.
Figure 18: Developer workstation task listing for assigned tasks for a Sprint. Right click on the mouse
generates drop down list that allows any data item to be updated almost instantaneously.
The cumulative time remaining for a release can be obtained in real time by anyone in the
company with access to GNATS. At PatientKeeper, that is every person in the company. The
days remaining for all tasks assigned to a release are totaled to calculate cumulative backlog, the
number charted on a Scrum burndown chart. Because there are thousands of tasks in the system
and any tasked that is touched is updated every day it is touched, the phenomenon of statistical
regression towards the mean [102] makes the summary data on cumulative time to release very
accurate. It achieves the holy grail of accounting software, microcosting of every activity in a
company [103].
4.5.4 Product Development Process Refinements
Product Management serves as the Product Owner at PatientKeeper and must provide functional
specifications for features that sufficiently describe the user experience so that developers can
begin design and coding. This typical means screens shots, workflow between screens, business
logic, and data items required. A working demo is normally prototyped and approved by a
physician user group before a feature can enter a sprint backlog.

158
The Product Owner controls the GNATS entries for a release and the bug flow into the system.
Bugs initially go into a triage category and the Product Owner assigns them to a release based on
priority, customer requests, implementation issues, and so forth. Features are initially placed into
GNATS as placeholders assigned to a release. Developers can pick them up and translate them
into tasks specific to a sprint.
Figure 7 shows a burndown chart generated by GNATS for PatientKeeper Release 6. It shows
product backlog accumulating as placeholders until the Sprint started on 12 June 2002. At that
time, developers began breaking down product features into estimated tasks for the next sprint.
This drove the backlog up as new tasks are discovered that were unanticipated in original
product feature estimates.
On 26 June 2002, the PatientKeeper CEO decided to commit the company to a major user
interface rework during Release 6. When the Product Owner entered over 80 new tasks into the
release the burndown chart shot up quickly in a two day period. This was visible immediately to
the entire company because burndown charts were mailed out to the company on a daily basis.
This caused a lot of consternation because there was no way that this iteration would be a
standard 30 day Sprint. PatientKeeper has a weekly MetaScrum meeting which includes leaders
from all departments in the company where release priorities and features are reexamined on a
weekly basis. It was determined that the value to the company of refining the user interface of
the product was very high in a competitive environment and the sprint would be extended to 24
August based on the Scrum burndown chart. This would require the development team to have
perfect development days from the beginning of July through the end of August.
An ideal developer day is the amount of work that can be accomplished if a developer works
uninterruptedly for a normal work day. Most teams preplan to get 40-60% of a development day
due to meetings and extraneous interruptions for team members. Getting 180 developer days
with an 8 person team in 42 calendar days without overtime was not going to be easy. The policy
at PatientKeeper was to sustainable development using a normal work week with night and
weekend activity required only when rare emergencies occurred, i.e. production customers hard
down in the field.
PatientKeeper had the advantage of smooth running Scrums with few interruptions other than
release priorities. As a result, their normal velocity or number of days work accomplished per
day per developer ran better than 60%. Years of historical data also showed they finished their
tasks in an average of 85% of the original estimate. This often did not show up in calendar days
due to release priority conflicts. However the leadership team realized that slippage was
normally due to management prioritization problems, not developer slippage, and GNATS has
mountains of data to prove it.
The solution was to focus the entire team on one release and over a 42 calendar day period, or 30
business days, the developers delivered 180 days of work for a velocity of 2 days of work for

159
every 3 business days invested or 66% of the ideal. This was planned in advance, the MetaScrum
leadership repositioned all customers at the beginning of July, and overtime was not required.
Figure 19: Scrum burndown chart autogenerated by GNATS. Product management was entering
placeholders from product backlog until 6/12/2002 when the sprint started. It extended beyond 30 days for
reasons described elsewhere. The pain of this Scrum reconvinced everyone that 30 days is the maximum
Scrum length.
The capability of total transparency where all data is available to everyone extends the concept
of global visibility during a Scrum to the entire company. This allows all parts of the company to
replan activities routinely. In the case of PatientKeeper, new Sprints can be started, changed, or
destroyed in the weekly MetaScrum without disrupting the focus of the Scrum teams.
4.5.5 Project Reporting
The GNATS reporting system was refined to allow sophisticated management reporting. Two of
the most useful examples are (1) tracking the quality of a product portfolio and (2) automated
generation of Gantt charts for MetaScrum planning.
4.5.5.1 Tracking Quality of a Product Portfolio
A useful measure of product quality, code stability, and forward progress is a chart that shows
arrival of new development tasks, completion of development tasks that change status to
verification (where they become the responsibility of the QA team), and closing of tasks when
testing by the QA team is complete. The cumulative number of outstanding defects has been
divided by 10 in Figure 8 to allow charting of the cumulative total in the same range and daily
defect arrival.

160
Defects Open/Closed by Day:
Managing Quality of Product Portfolio
Figure 20: Daily arrival of defects along with cumulative defect count. Company objective is to drive total
bug count across a deployed portfolio of releases below 100. This is challenging as there are about 100
hospitals deployed on the 2.4.* releases.
4.5.5.2 Gantt Chart for MetaScrum Planning
Gantt charts are useful for planning purposes. However, they are poor for tracking software
projects because dependencies change on a daily basis. A full time developer can be absorbed
keeping Microsoft Project up to date for a single Scrum team and Scrum was designed to
eliminate this wasteful activity. For the last six years, PatientKeeper has evaluated whether to
have any project management other than Product Owners and ScrumMasters. The decision has
always been that they are unnecessary waste.
A MetaScrum can find a Gantt chart useful, but only if it is machine generated. A human
generated Gantt chart is inaccurate in the beginning and completely outdated within a week in a
fast-paced company. An effective Gantt chart can be calculated in real time based on data
capture in the reporting system.

161
Dynamic GANTT Chart:
Managing Multiple Releases
Release 2.5.1c & 3.0
svakil
sanj
sally
rkozak
legli
R2.5.1c
Days
Ver 2.5.1c
jcarp
R3.0
Ver 3.0
gaspeslagh
dkoulomzin
dcohan
ctashjian
abonventre
0 5 10 15 20 25 30 35 40
Figure 21: Dynamically generated Gantt chart. End points reflect anticipated release date in days from day of
generation.
Dynamic Gantt charts can be generated by staff members for a small team, or by teams for a
company. The chart above shows two releases broken into development tasks and QA tasks. The
X axis is days from generation day. The Y axis is staff members by name showing assignments
for the team.
The result of an automatically generated Gantt chart is a surprise to traditional managers. It will
show most individuals are loaded only 1-3 days out. They will choose their next task when they
complete their current task. Key people like the lead architect or QA manager will have tasks
queued up to be passed off to whoever is ready to take them “just in time.”
When PatientKeeper managers were asked if they wanted to manage resources to allow an
autogenerated Gantt charge to show release dates they were counting on, they immediately
declined, noting that disruption of a self-organizing system would radically cut the velocity of
the team and create unnecessary work for managers. They gave up the notion of trying to use a
Gantt chart to plan releases and went back to the Product Owner’s roadmap for release planning.
This is a milestone based timeline that shows the Product Owner’s best estimate of release dates
with specified highest values features based on known team velocities and project dependencies.
4.6 Type C Scrum Rationale

162
As noted in our Pattern Languages of Program Design paper [25], “it is very easy to over- or
under-estimate, which leads either to idle developer time or to delays in the completion of an
assignment. Therefore, it is better to frequently sample the status of small assignments.
Processes with a high degree of unpredictability cannot use traditional project planning
techniques such as Gantt or PERT charts only, because the rate of change of what is being
analyzed, accomplished, or created is too high. Instead, constant reprioritization of tasks offers
an adaptive mechanism that provides sampling of systemic knowledge over short periods of
time. Scrum meetings help also in the creation of an anticipating culture [104] because they
encourage productive values:
• They increase the overall sense of urgency.

• They promote the sharing of knowledge.
• They encourage dense communications.
• They facilitate honesty among developers since everyone has to give a daily status.
In a Type C Scrum, the urgency, sharing, communications, and honesty behaviors are extended
company wide. “From the Complexity Theory perspective [105, 106], Scrum allows flocking by
forcing a faster agent interaction, therefore accelerating the process of self-organization because
it shifts resources opportunistically through the daily Scrum meetings.[25]” When extending
company wide, the entire company can self-organize on a weekly basis. The following behaviors
become commonplace:
• There is never an unexpected late release as problems are seen long before the release
date. The company self-organizes around the issues raised in the MetaScrum.
• Changes in customer requirements are reflected immediately in product backlog and
relevant Sprint backlog. Decisions are made to reorganize on a weekly basis in the
MetaScrum.
• Company imperatives and management changes that affect product backlog are made
only in the MetaScrum. This eliminates most politics, lobbying, and closed door
meetings.
• Customer impact and schedule impacts are deal with immediately in the MetaScrum at
the time of decision. The CEO, sales staff, and account management walk out of the
meeting with assigned tasks to deal with customers affected by decisions.
4.7 Type C Scrum Resulting Context
The move to a Type C Scrum to improve development productivity had far reaching effects on
the company making it more flexible, more decisive, more adaptable, and a better place to work.
The same effects commonly seen on Scrum teams were reflected throughout the company.
Project management was totally automated. The result is paperless project management and
reporting, largely without human intervention. Scrum execution has become exceptionally
efficient and the automated tracking system has become mission critical. Burndown charts have
evolved to frame the entire status of a project on one chart. The chart below instantaneously
reflects project state for Release 3.20 at a glance to those familiar with the data. With all tasks
163
entered at 16 hours or less and bug fixes typically less than a day, the aggregate number of tasks
can be monitored and downward velocity is highly predictive of delivery date. Information is
presented as follows:
• Dark Blue Diamond – Release 3.20 current open – cumulative work remaining
• Yellow Triangle – Release 3.20 daily closed - items closed by QA each day
• Purple Star – Release 3.20 total closed - cumulative closed (on scale at right)
• Pink Square – Release 3.20 current verification - current total in verification (items QA
needs to test and close)
• Light Blue X – Release 3.20 daily open – new tasks opened per day
Com prehensive Burndown Chart

320 PR B urndow n
180 450
160 400
140 350
total closed
120 300
PR count
100 250
80 200
60 150
40 100
20 50
0 0
5
5
05
05
00
00
00
00
20
20
/2
/2
/2
/2
9/
6/
16
23
30
13
5/
6/
5/
5/
5/
6/
da te
320 current open 320 current verification

320 daily 'closed' 320 daily open
320 total 'closed'
© Jeff Sutherland 1993-2007
Figure 22: Comprehensive Burndown Chart showing daily task inflow/outflow and cumulative project churn
[15].
The cumulative closed (right scale) is much higher than the starting number of about 150 tasks
(left scale). The reason for this is that the Sprint Backlog minor changes are constantly coming
into the Sprint Backlog for the following reasons:
• QA is finding bugs, often generating multiple tasks that can be closed with one developer
fix.
164
• Product development is adding tasks primarily because of customers moving in and out
of the mix for go-live at end of Sprint (this is not allowed in Type A and B Sprints).
• Development is discovering new tasks as they flesh out technical design.
The cumulative closed tasks is an indicator of the churn on a project and the reason why Brooks
[5] notes that development always take three times as long as initial estimates. Automated
reporting and rapid turnaround can radically reduce time to complete new tasks. Note the strong
downward velocity on the Burndown Chart despite project churn. PatientKeeper was able to
move quickly into the marketplace and achieve leadership in the healthcare mobile/wireless
market [11] through delivering over 45 production releases of the PatientKeeper Platform in
2005 for large enterprises such as Partners Healthcare in Boston, Johns Hopkins in Baltimore,
and Duke University Health System in Durham. Gartner Group put PatientKeeper as the leader
in their “magic quadrant” for the industry segment. Type C Scrum was a key contributor to this
success.
Challengers Leaders
Medical Information Epic Allscripts

Technology Siemens Systems Healthcare
(MEDITECH) Solutions
Eclipsys McKesson PatientKeeper
Ability QuadraMed
to ePocrates MDanywhere
Execute MDeverywhere Technologies
MercuryMD MedAptus
ePhysicia
Niche Players Visionaries

Completeness of Vision
Figure 23: Gartner Group “magic quadrant” for healthcare mobile applications [107].

165
5. Conclusions
Moving to a Type C Scrum is not for the faint of heart. It requires Scrum teams that can execute
a standard sprint flawlessly, an automated data collection and reporting system that is easy to
implement and update, and a corporate culture that embraces change. Going to a Type C Scrum
will transform a company in an organization where Scrum becomes mission critical for the entire
organization, not just software development.
Gregor Rothfuss provided an excellent summary when seeing the reporting mechanisms for a
Type C Scrum for the first time [90]:
i was a guest at the Agile roundtable near Boston last night. The event drew a crowd of veteran
software engineers, i was the youngest in attendance by about 20 years.
ken schwaber outlined his and jeff sutherland's Scrum approach, which struck me as interesting
and worthwhile to follow up on.
jeff sutherland, CTO of patientkeeper, demonstrated how he manages his teams of developers
with GNATS. jeff figured that developers loathe red tape, and had the goal to limit the effort
required to 1 minute per day for developers, and 10 minutes per day for project managers.
and he was not using gantt charts to achieve this either. calling gantt charts totally useless for
project management beyond giving warm fuzzies to the client, he explained how he leveraged
their bug tracker to double as a means to keep track of effort.
each morning, developers review their tasks and update the work remaining estimates which
have a granularity of one day. the project managers, in turn, analyze the reports that GNATS
automatically creates. reports such as number of new tasks vs. closed tasks, total work
remaining and other metrics that can be derived from the task data.
tasks are the cornerstone here. jeff was able to demonstrate to the business side that the high
level business goals were off by 100% with their effort estimates, while the low-level tasks
achieved an accuracy of 10% on average. this led to enthusiasm from all parties to drill down on
any project and get to the task level ASAP to get meaningful estimates. and, like psychohistory,
project management is inherently stochastic.
‘nowhere to run, nowhere to hide’

the level of transparency of this system is unprecedented. with everyone in the company able to
see on a daily basis how much work was remaining and what the roadblocks were, the initial
fears that developers would be pounded on by management turned out to be unfounded. instead,
the transparency enables everyone to do real-time adjustments and to detect problems early,
which has taken a lot of politics and second-guessing out of the equation.
when analyzing a project, jeff focuses on burn down, the part of a release where open tasks are
relentlessly driven down to 0 by a joint effort of developers and business people. the
corresponding graphic (roughly a bell curve) illustrates the importance of the burn down nicely,
166
adding weight to jeff's assertion that burn down is the only thing that matters to get a release
done in time.
“which prompted me to ask for advice on how to drive an open source release as a release
manager. people are not exactly required to do your bidding, but metrics may help there too.
collect these useful data points, as the bugzilla-bitkeeper integration is doing, and let them speak
for themselves. peer pressure and pride in workmanship will take over from there. that's the idea
anyway…”
Key features mentioned in the Rothfuss report are:

• Unprecedented transparency
• Companywide visibility
• Metrics driven decision making
• Peer pressure and pride in workmanship driving productivity
Type C Scrum increases speed of development, aligns individual and corporate objectives,
creates a culture driven by performance, supports shareholder value creation, achieves stable and
consistent communication of performance at all levels, and enhances individual development and
quality of life.

167
Chapter 7: Case Studies
One of the most interesting things about Scrum is the unique case studies that have been
published at IEEE conferences. Scrum is used by some of the most productive, high maturity,
and most profitable software development teams in the world. It powers:
• The most productive large development project (over a million lines of code) ever
documented.
• The most unique CMMI Level 5 implementation on the planet.
• The most profitable software development project in the history of software development.
The SirsiDynix project reimplemented a large system that supported over 12,500 libraries in 42
countries with 120 million users. In the middle of the project, the development team was doubled
in size with teams from StarSoft Development Labs in St. Petersburg, Russia. Their velocity
more than doubled the day the Russian teams came online. The lines of code delivered per
developer was as high as the best small team collocated Scrum projects.
Systematic Software Engineering in Aarhus, Denmark, spent seven years and over 100,000
person hours of process engineers to achieve CMMI Level 5 certification, reduce rework by
80%, and improve productivity by 31%. Within six months after a Scrum Certification course
they had reduced planning time by 80%, defects by 40%, total cost of a project by 50% while
simultaneously enhancing customer and employee satisfaction. They now bid Scrum projects at
50% of the cost of waterfall projects.
One of the most interesting Scrum implementations is Google’s AdWords implementations. This
application drives the majority of Google revenue growth and helps create market capitalization
that is higher than Intel and just below that of Chevron, the most profitable oil company in the
world. The AdWords project, powered by Scrum, has distributed teams in five locations and
interfaces with virtually all Google products on every release. As a result, the Google project
manager needed to insert more structure than is usually associated with Google teams. His
seamless introduction of Scrum based on resolving the highest priority impediments observed by
the teams resulted in an implementation that no longer needed a ScrumMaster to function. The
teams ran by themselves.

168
Ssh! We are adding a process… (at Google)
Mark Striebeck, Google Inc.
[email protected]
Abstract
Google is very successful in maintaining its startup culture which is very open and engineering-
centric. Project teams don’t have a project manager, but organize themselves and communicate
directly with all stakeholders. Most feature decisions are made by the engineering teams
themselves. As well as this works for products like search, gmail … it creates issues for the
AdWords frontend (AWFE) application. AWFE is much more product management and release
date driven then other Google applications. This presentation discusses how we carefully
introduced agile practices to coordinate the AWFE development teams and made the process
more efficient and predictable.
1. Introduction
Google is known for its startup culture and its efforts to maintain it. In terms of the Agile
Manifesto Google is almost entirely on the “left hand side” (with the exception of “Working
Software”). Traditionally, project teams do not have a project manager, but organize themselves
and communicate directly with all stakeholders. Even now, where Google has more then 6000
employees in numerous offices around the world, Google is still very engineering driven. Many
new product ideas come from the 20% projects4 of its employees.
The overall mindset at Google is to have as little as possible standard processes as possible. The
reason is that the individual engineering teams will know best what is right for them. Upper
management on the other side has trust in its engineers that they would not abuse this autonomy
but do what is best for their project and the company.
AdWords is different. Being a B2B application, means that it needs much more business focus,
sales material has to be updated, support has to be trained, and external communication about
major features has to be prepared (forums, blogs and emails).
Therefore AdWords had a few standards:
• From the initial product idea, the product manager together with a UI designer and
usability specialists creates almost final UI mockups. These mockups are used for a final
project review by senior management and then given to engineering for implementation.
4
Every Google employee is encouraged to spend 20% of his/her time on a personal project. This project should not be too closely related to the
employees’ actual work.
169
• During the whole project lifecycle, the product manager holds weekly meetings with all
stakeholders (engineering, QA, UI, support, marketing). These core team meetings are
the main communication channel. All major product decisions are made or at least
discussed here. Lots of change requests come from these core team meetings during the
project lifetime.
• Although, the core team sets initial release dates (with input from engineering), the final
release date is determined by engineering progress and quality of the code. Given the
scale of AdWords (number of users, business relevance, load, infrastructure); a small bug
can have very severe consequences. Therefore features are rather delayed then released
with insufficient or even unknown quality.
• This level of process worked well in the beginnings of AdWords. But the AdWords
product development outgrew this ultra lightweight process
• The application code consists of more then 500KLOC
The engineering team is distributed in 5 offices worldwide – there are constantly 15-20 major
projects ongoing plus maintenance and small improvements.
And the AdWords application and development team is still growing…
The unpredictability of launch dates caused more and more concern. Nobody wanted to lower
the high quality standards. But the initial release dates needed to be more reliable and delays
should at least be known and communicated much earlier.
Because of its size and complexity AdWords has fairly large management team (for Google
standards). In order to be effective the management team needed much more visibility into the
projects and their status.
Finally, the rate of change is very high in AdWords. Teams who work on a project for a few
months might find that they have a lot of cleanup to do before they can finally launch. Not so
much because of code integration issues (the AdWords team runs a fairly comprehensive
continuous integration suite) but because of feature changes. Often, projects that run for a long
time have to play catch-up with all the feature changes before they release. In a few cases this
lead to significant delays.
2. First agile attempts

Trying to introduce a process in a start-up environment such as Google often meets resistance.
Because of the googley way of developing software, many engineers simply do not believe that
any formal process can have a benefit but will only slow them down.
When I took on my first projects at Google I was just a few months with the company. The
engineers did not know me at all. But it was interesting to see how the Google culture helped me
here: A big part of Google culture is trust. This goes through the whole organization. And
although I was new to Google and AdWords, the engineers and PMs trusted me that I would do
the right things. Or better: they trusted the people who hired me that I am someone who would
do a good job.
170
So, my strategy was to get as little involved as possible in the actual coding part and to start with
a few practices that would just help us to track progress and show issues. Then we would
introduce individual agile practices to fix such issues during development. I decided to start with
the following practices:
A release backlog and burndown charts [i]. These two tools provide high visibility into the
development progress for the project team, but also outsiders. Using simple wiki pages to store
the backlogs allowed the engineers to update their progress in very little time. I decided to
measure the burndown rate by tasks complete, not feature complete. Measuring progress in
feature complete has many advantages but also forces a team to change their development
process a lot. It was one of the areas where I decided to rather introduce this practice later in
order not to overwhelm the team.
In past projects I made very good experience with estimating features/tasks in points [ii].
Especially in an environment like AdWords, where engineers are often interrupted by meetings
or tech talks, real time estimates are a problem. If the burndown graph tells us that we are
implementing 3 days of work per week then it often leads to discussions what the team is doing
the other 2 days. Or people try to match their updates to real days. Points are a good abstraction
layer that avoids any such discussion.
Scope changes are included in a controlled way by first estimating them, adding them to the
backlog. Here, the burndown charts helped tremendously to get a quick assessment of the
impact.
A weekly development checkpoint meeting to plan the next week and work on scope changes.
These checkpoint meetings were attended by the engineers, QA, PM and UI. At this point I did
not introduce real iterations. My personal experience was that changing to iteration-based
development is a significant change for developers and QA. It sounded too heavy to introduce at
this point.
For the adoption of these practices, I tried very hard not to implement anything top-down but to
get buy-in from engineers and the product managers. The initial changes sounded reasonable to
the engineers. Because I was managing several projects, I could not be too closely involved in
the development activities itself. This probably worked to my advantage – the engineers realized
quickly that I would not try to tell them how to do their job, but that I only structure the project
in a certain way which was not too intrusive. Also, one of the goals was to keep the self-
organizing character of teams intact. After all, this is a big part of Google culture and our agile
adoption approach would have failed if we had severely impacted it – no matter how successful
the projects would have been.
This approach also helped me to work with several projects at the same time. Many meetings
regarding UI, features, design… took place without me. Only when we discussed scope,
scheduling or planned the next steps, I was there and was usually leading the meeting.
2.1. The guinea pig projects

171
Changes at Google are often done in some kind of guerilla approach: one project team adopts
something new. If it works, other project teams get interested and will try it as well. Therefore,
we started only with two projects:
Project A: This was a very new piece of functionality which did not overlap with existing
features. The UI was fairly complex; the engineering team consisted of new recent college
graduates working in a remote office.
Project B: This project was a simplified version of AdWords. It was heavily integrated into
existing features (we basically had to think about every other feature and had to integrate or
disable it). The team consisted of experienced engineers. Some of which had already work for
some time at Google, others were new to Google).
2.2. The first process steps

In both projects, we used the UI mockups to generate the release backlog by dissecting the
screens into individual features. This pre-development process is very well established at Google
and it seemed too complicated to make this part more agile.
The release backlogs were stored in wiki pages which made it very easy for engineers to update
them. From these wiki pages we automatically generated burndown graphs to visualize the
project progress. The concept of giving status updates in work left and not in work completed
was initially strange to both teams. But the engineers quickly realized the advantage.
As stated earlier I did not introduce iterations at this time. Instead I installed weekly checkpoints
with the development team (PM, UI, Engineering and QA). In these checkpoint meetings, we
discussed progress, additional feature requests and other issues. Additional features were
estimated and added to the release backlog. I extended the burndown graphs and used a variable
floor to indicate the scope changes. The graphs gave us quick feedback what the estimated
impact of these additional features was.
Table 1: Burndown graph with variable floor

172
Although I did not try to implement an immediate testing of implemented features, I wanted to
get away from the purely phased approach where testing starts after development is finished. To
push for this, we setup staging servers that were rebuilt on a nightly base with the latest code.
These staging servers were used for testing the application but also for UI walkthroughs5.
Usually, they are performed towards the end of a project when the system is nearly complete.
But because we staged the application early on and implemented end-user features (from the UI
mockups) we could start with these UI walkthroughs much earlier and gather important feedback
for the further development.
2.3. Issues to overcome

In both projects we faced similar issues:
Customer / Product Owner concept

Most agile processes have this role which is responsible for features, prioritization and ultimately
scope vs. release date decisions. It is usually an individual or team outside of the development
team. But at Google, many of these responsibilities rest with the team leads. The product
managers usually have more then 10 projects at the same time. This does not give them the
bandwidth that the product owner role requires. Also, they trust the tech leads and UI designers
enough that they will make good decisions (often, when I asked a product manager for
prioritization of a feature, he turned to his tech lead and simply asked “what do you want to
do?”).
This gives the planning and prioritization meetings a different dynamic. Often, the tech leads do
not see the need to make such decisions during the planning meetings as they know that they will
be involved enough during development itself that they can make such decisions at a later point.
I usually drove the team to make at least those decisions which are necessary to create good
effort estimates and priorities for the backlog. I always wanted to leave the weekly checkpoint
meetings with good updates to the release backlog.
Retrospectives
For me, frequent retrospectives [iii] became such an important part of software development that
I tried to install them in the weekly checkpoints from the beginning. It would have helped a lot
with improving our process through constant feedback.
But both teams were not (yet) used to having a formal development process. The weekly
retrospectives usually turned into a status report from the last week but very little about the
process itself. This was aggravated by the engineering centric culture at Google. When an issue
comes up, most engineers at Google only consider technology to fix it.
After a few weeks, I silently dropped retrospectives from the weekly checkpoints. I decided to
wait until the teams embraced the concept of a development process and that they own it and
could change it to fix problems.
Constant scope increase
5
UI walkthroughs are live demonstrations of the system with the whole core team to gather feedback and uncover usability issues early enough.

173
In both projects, the scope increased significantly (more then 30%) during development.
Interestingly, these scope changes were not the result of additional feature requests by the
product managers. Most additional tasks were the results of oversights during the release
planning:
The engineering team missed features in the UI mockups when we created the release backlog
Integrations into other AdWords features were overlooked. Also, the rate of change in AdWords
is very high. During development others areas of the application changed and we had to change
our integration as well.
Most of these additional tasks could not be down prioritized for a later release but had to be
added to the release.
In both projects, this lead to several postponements of the release date as no other feature could
be dropped from the first release.
Although, there was considerable frustration about these delays, both project teams and
management appreciated that we at least knew about these postponements early enough and not
just the week before release. The burndown graphs gave a good visualization and the release
backlogs made it easy for everyone to understand what was left to be implemented.
The backlog was handy as things came up over time and as we dived deeper. One function was
to not loose the line items but more important it was useful for the team to see how many un
anticipated issues cropped up and have a good snap shot in time.
Product Manager
2.4. Working with the remote team

As stated earlier, project A was implemented in a remote location. The rest of the core team was
in our headquarters. Initially, I was concerned how that team would react to my leadership – if
they would appreciate it as much as the other team or if the would regard it as a heavy-handed
approach from headquarters.
To my surprise I did not encounter many issues with this project. Only providing tools to get
more visibility into development progress and facilitating planning meetings seemed to be the
right level to give the remote team enough room to work mostly autonomously. Also, I could
make myself very useful in facilitating lots of communication with other engineers in our
headquarters. The team realized quickly that I indeed tried to help the project progress and not to
control them remotely.
3. Adding agility – one practice at a time
3.1. Daily standup meetings
Both project teams initially rejected the idea of daily standup meetings [iv]. They were seen as an
unnecessary overhead.
174
But during development we discovered issues in the weekly checkpoints from the past weeks:
QA tested unfinished features or was not sure how to test new features
Engineers who worked on related features worked on the same refactors. The AdWords
engineering team has a very healthy culture of constantly refactoring the code. The downside is
that two engineers who work on related features often start to improve the same code.
Engineers could not continue with their implementation because they depended on a task from
another engineer. Often enough, the other engineer was not aware of this dependency.
It was clear to everybody that these issues could have been avoided had the team communicated
earlier. At this point it became easy to convince both teams to try out daily standup meetings and
to include QA in these meetings.
The first standup meetings were quite lengthy. Everybody had a lot to talk about and had
problems to focus just on a quick status update (“done”, “to-do”, and “issues”). But after a few
days nobody had a big baggage anymore and everybody realized that there is not much to talk if
you restrict yourself to the past 12 hours and next 12 hours. Several issues were resolved or at
least uncovered during these meetings. After a couple of weeks, both projects did not need a
reminder anymore but made the standup meeting part of their daily routine.
3.2. Small steps – Completely finishing a feature/task
In project A, the progress looked very good. Initially, we estimated 3 weeks for a set of screens.
When we did low-level estimates, we came to 40 points. After the first week, the team did 8
points – in the second week 7.5 points. I looked as if the initial estimate was too low and the
team would need 5 instead of 3 weeks.
Interestingly, the tech lead of the team was convinced that the screens could still be implemented
in 3 weeks (i.e. all remaining 24.5 points in 1 week!) quote: “It just does not feel that much
anymore”.
After week 3, the team was not done. The team implemented another 9 points. The velocity
looked very stable: ~8 points per week.
To my big surprise, the tech lead announced in the core team meeting once again that his team
will be done in one week…
It took me some time to learn to trust the burndown graph and to question my gut feeling when a
feature would be finished.
Tech Lead
The fourth and fifth week showed a significant drop in velocity: 4 points and 2.5 points! It turned
out that the team did not completely finish the tasks: tests were not written, code was not
reviewed (which is mandatory at Google), features were not completely integrated. This caused

175
the burndown graph to go down because we did not measure progress in finished features, but in
tasks.
This caused a further delay and the screens were finally implemented after 7 weeks. This
additional delay caused some concern with the core team. To avoid this situation I added a
green/yellow/red color coding to the burndown charts to indicate how many tasks are
new/started/finished. This made it very clear if velocity was high because many features are
partially finished or if the team completely finished a feature before moving to the next one.
Figure 2: Indicating started and finished tasks
The team responded very positively. It was quite a shock for the engineers to see that up to 80%
of all tasks were in a ‘started’ state. They started to keep the corridor of started tasks as small as
possible.
Overall, this was a very healthy learning experience for the team. It showed them the difficulty
that we tend to have when trying to estimate a release date instead of deriving the release date
from effort estimates and progress. It also showed them that we can only measure progress well,
if we completely finish tasks and not leave small bits and pieces around which sometimes turn
out to be much larger then we thought.
3.3. Spikes
In the weekly checkpoint meetings we often discovered that tasks took much longer then initially
estimated. Or the team had problems with estimating a new feature.
Initially, the engineers just wanted to pad estimates for such unknown tasks. Often enough, these
padded estimates were much too high or still too low. And everybody could see that they
lowered the usability of our burndown graphs significantly. So, we added in a spike (an
investigative task) to help determine what the effort for the implementation would be. Especially
when the scope continued to grow, everybody realized the value of getting a better estimate of
implementing a feature before actually starting to work on it.

176
4. Release experience
The two projects had somewhat different releases:
Project A)
The team had fixed many bugs already during development, only few bugs were discovered in
the final integration test phase. It was a very smooth launch.
Project B)
Because of the integration into all other AdWords features, QA found many issues during
development – most of them through exploratory testing [v] (i.e. not really tied to a particular
product feature). The team tried to keep the bug backlog under control but did not want to fix all
bugs. When we came close to launch, we had to review the bug backlog several times and down
prioritize many bugs. Until a few days before launch it was not clear if we could fix enough bugs
to release it.
At least the team did not encounter any issues that required a complete redesign of some area –
which could have easily happened for such a far reaching feature.
Still, the overall release experience was very positive. Both projects were very successful in
production and had very few issues.
5. Feedback and next steps

I held post-mortem meetings with both projects. In these meetings I focused the teams on listing
positives and negatives and not jumping to discuss solutions immediately. From the overall list,
the teams selected the worst issues and best practices to keep:
Positive
Project Management and tools (burndown charts and backlogs)
Early QA and availability of a staging server
Teamwork and collaboration
Negative
Unclear or non existent prioritization
Felt as if team missed release date several times
Too risky at end because of bug backlog (Project B)
It was very encouraging that both teams found the overhead of maintaining and updating the
release backlogs worth doing.
Burndown charts made it easy to see when were making progress, and gave us a nice sense of
satisfaction and completeness.
Engineer
And, furthermore that the process did not impact the great teamwork and collaboration that
Google teams have. Also, the effort of maintaining a dedicated staging server was appreciated.
177
The engineers from both teams were very positive about the early testing and feedback by QA
that the staging server afforded.
I think it took some time getting used to the approach of testing so early in development, and also
making sure that QA and dev were on the same page. I think that our daily standups and also
having QA co-located with dev has helped greatly here.
Engineer
6. The second version

From the feedback of the post-mortem meeting I tried to modify the development process further
to address the worst issues.
In both teams I gave at this point a presentation about a full Scrum process [vi]. During the first
projects there were many tech talks at Google about agile development (by internal and external
speakers). Both teams got very interested in it. They could see that their practices fit into agile
development but heard a lot about other practices too. Also, the very positive feedback of my
project management style and tools showed me that the engineers trusted me and my guidance.
In both teams we discussed which additional practices to adopt:
Product/Release Backlog
To address the prioritization issue, I worked with the product managers of both projects to
organize their requirements in prioritized lists. It took a little bit of time for them to get used to
it, but was not a major effort. The core team members liked the backlogs a lot. It gave them
much more visibility and input into development. Initially, there was still the desire to make each
feature high priority. But soon everybody realized that even if a feature is not included in the
current iteration, it will still get done fairly soon.
Iteration based development

This was the hardest practice to introduce. Without practical experience it is hard to explain why
iterations are better than scheduling the whole release at once and adding to it when necessary.
But with the feedback about missing deadlines and too many bugs, I could explain how an
iteration based approach would address these. The concept of not only implementing but also
testing and completely fixing features within the same iteration sounded very appealing to the
engineers. Although, they were somewhat skeptical of this high-quality approach, both teams
wanted to give it a try.
The teams soon realized the advantages. The planning meetings became much more focused than
the weekly checkpoint meetings from the previous projects. No time was wasted with discussing
the same feature for 5 weeks but never implementing it. Or to discuss and design a feature that
finally gets dropped.
We agreed to start with 2 week iterations. This synchronizes well with the 2 week release cycle
of AdWords. We are finishing the iterations with the code freeze for the next push. This means
that a high-priority feature that gets put on the product backlog can be implemented and release
within 4 weeks without any interruption.

178
Development Pre-Iteration 2 Iteration 2
iterations Pre-Iteration 1 Iteration 1

week 1 week 2 week 3 week 4 week 5 week 6
Translation Code Push 1

Release
Deadline – Freeze –
cycles Push 1 Push 1 Translation Code Push 2
Deadline – Freeze –
Push 2 Push 2
Figure 3: Synchronized development iterations and release cycles
Retrospectives
After the previous projects, both teams had some experience with a defined development process
and that they can influence/change it. I started the iteration planning meetings again with a
retrospective and this time it was much more fruitful. Most contributions were about how we
develop our application and how we can improve that.
Review of iteration features with core team

In the first projects, we reviewed the application by clicking through it during the core team
meeting and collected some feedback. Now, with the iteration based development we do these
reviews at the end of each iteration and only on newly implemented features. This made the
reviews more focused and gives us feedback early enough so that we can integrate it in the next
iteration.
Testing tasks for features in same iteration

In order to test features in the same iteration as they are developed in, we added testing tasks to
the iteration backlog. The QA engineers were asked to provide effort estimates for these tasks so
that they can be included in the burndown chart.
Overall, the teams could see how these process changes would address the negative feedback
from the post-mortem meetings. Both teams did not fully understand how these practices would
work together but agreed to give it a try.
At this point I took on a third project where I implemented the new process from the beginning.
The product manager of this team was from Project A, the QA engineer from Project B. This
made the adoption much easier. Also, many people in AdWords had heard about how I ran my
projects and the barrier to try it out was considerably lower.
6.1. The world is better, but …
Overall, the more agile processes worked really well. Everybody noticed that the additional
structure comes with very little overhead and fixes many of the issues that we had before.
We're still getting up to speed on the iteration-based development. It's been nice for
development, now that our iterations are in sync w. AdWords code freeze cycle. It was hard at
first for UI/PM, but has gotten easier as PM has assembled farther projecting roadmap, to give
UI a clue what will be needed for a coming iteration.

179
Tech Lead
After a month or two, both product managers realized that they need to establish a requirement
process that ensures that we not only implement little bits and pieces at a time but keep the
overall release. This is an issue that I had with previous agile teams. I could persuade the product
managers to dissect their releases into smaller chunks and prioritize them.
For these I created release burndown charts to track when they will be finished. At this point I
started to measure progress on the release level in features complete. At this point it was very
easy to convince the teams that this is the right measurement as it would give us a much better
guidance where the release is.
The teams first thought that it was strange to have one iteration burndown chart and one release
burndown chart. But after a few iterations they saw the benefit of both. The iteration burndown
to guide actual development efforts. And the release burndown to guide the overall release
planning.
An ongoing issue is the QA involvement. I constantly have to push the QA engineers to test
features immediately after they are implemented. The reason is that the QA engineers support
several projects. And the other projects are not agile, i.e. don’t require much attention during
development, but a lot at the end. This made it hard for the QA engineers to constantly spend a
little bit of time each day on or project to give the engineers the immediate feedback. Right now,
both teams question if it is worth the effort to include QA tasks and effort estimates in our
planning as it does not seem to have any benefit.
For me, it seems like an extra task of updating a table with data (QA estimates) that’s not of
significance for me. But I’d really like to know if it’s helpful to others. So far, most of the
estimates have been 0.1 points.
QA Engineer
Finally, the teams do not try to create a releasable product at the end of the iteration (which is
even harder because of the QA issue mentioned above). There are always tasks half
implemented, not tested, need review… For now, I am not pushing too hard on this. The teams
completely implement enough features per iteration that we can release those with the next
AdWords update.
6.2. The project manager is dispensable
Recently, I went on a 3 week vacation. I was concerned how the teams would continue with the
agile process during my absence and reminders and reinforcements of our agile practices.
But it turns out that the teams embraced the process enough to continue it even without any
reinforcement. Iteration planning meetings happened, backlogs were created according to
previous velocity, and daily standup meetings took place …
7. Where are we going from here

180
With the success of three project teams we are now prepared to make much bolder steps.
Everybody in AdWords had at least heard about the advantages of the agile approach. Resistance
at this point will be much less.
• Establish backlogs and burndown charts as status reporting standards in AdWords. Even if
teams do not adopt other agile practices, these practices are easy to implement and provide
a very good visibility for the teams themselves but also management and other outsiders.
• Other managers voiced interest. With a shadowing approach I will guide them through the
agile process and try to give them enough experience to implement agile practices in their
projects by themselves
• A few projects involve teams from several AdWords departments (frontend, backend,
NetAPI…). Such projects always required much more management attention. As great as
Google engineers and tech leads are, coordinating and synchronizing a teams efforts with
other teams often distracts tech leads too much. We will either try to coordinate these
teams as one big team (one backlog, one burndown chart) or use the “Scrum-of-Scrum”
[vii] approach.
• During the first months at Google I heard from other departments who are using some agile
practices or full-on Scrum/XP processes. To support this effort we started a grouplet6 that
focuses on agile development. We just recently started this grouplet and the initial response
/ interest was overwhelming – not only from engineering, but also other groups (QA,
Product Management, UI, Usability)
• The UI development and usability part of our development projects is still very
frontloaded. Almost all of this work is done before development starts. A few usability
experts and UI designers showed interest in making this also part of the iteration-based
development.
8. Summary
With the help of an experienced agile leader (ScrumMaster, XP coach…) it was possible to
carefully introduce agile practices into Google - an environment that does not have an affinity to
processes in general. Instead of introducing a grand new process, individual practices could be
introduced either to fix observed issues or just to “try them out” – the development teams
realized the advantages very soon.
Along with these practices came a visibility into the development status that gave the approach
great management support.
All this could be done without destroying the great bottom-up culture that Google prides itself of.
The practices only affect how the projects are structured. Design and implementation remains
fully an engineering responsibility. With some modifications, we could even keep the very
strong role of tech leads and UI designers.
6
Google grouplets are cross-department groups which focus on a specific area of the software development process (there is a
tech documentation grouplet, a build tools grouplet…) The members of the grouplet use their 20% time for their participation.

181
In keeping the great culture and self-organization of the teams, I could easily manage several
projects in parallel. I could continue to rely on all core team members to communicate
effectively without introducing any heavy processes.

182
5. References
[1] H. Takeuchi and I. Nonaka, "The New New Product Development Game," Harvard
Business Review, 1986.
[2] J. O. Coplien, "Borland Software Craftsmanship: A New Look at Process, Quality and
Productivity," in 5th Annual Borland International Conference, Orlando, FL, 1994.
[3] A. C. Kay, "The early history of Smalltalk," in The second ACM SIGPLAN conference on
History of programming languages (HOPL-II), 1993.
[4] Collins, Good to Great: Why Some Companies Make the Leap... and Others Don't
Collins, 2001.
[5] R. A. Brooks, "Intelligence without representation," Artificial Intelligence, vol. 47, pp.
139-159, 1991.
[6] C. G. Langton, "Life at the Edge of Chaos," in Artificial Life II, SFI Studies in the
Sciences of Complexity, Held Feb 1990 in Sante Fe, NM, 1992, pp. 41-91.
[7] C. Jones, Software assessments, benchmarks, and best practices / Capers Jones. Boston,
Mass.: Addison Wesley, 2000.
[8] M. Poppendieck and T. Poppendieck, Lean Software Development: An Implementation
Guide: Addison-Wesley, 2006.
[9] M. Poppendieck, "PatientKeeper, the only company in the world that is not thrashing.," J.
Sutherland, Ed. Boston, MA, 2007.
[10] C. Alexander, The Timeless Way of Building, Volume 1, 1979.
[11] H. Takeuchi and I. Nonaka, Hitotsubashi on Knowledge Management. Singapore: John
Wiley & Sons (Asia), 2004.
[12] M. Poppendieck, "A History of Lean: From Manufacturing to Software Development," in
JAOO Conference, Aarhus, Denmark, 2005.
[13] J. K. Liker, The Toyota way : 14 Management Principles from the World's Greatest
Manufacturer. New York: McGraw-Hill, 2004.
[14] W. D. Holford and M. Ebrahimi, "Honda: Approach to Innovation in Aerospace and
Automotive/Pick-Up Truck Development: A Dialectical Yet Coherent Firm," in 40th
Annual Hawaii International Conference on System Sciences (HICSS-40), Big Island,
Hawaii, 2007.
[15] P. M. Senge, The Fifth Discipline: the Art and Practice of the Learning Organization.
New York: Currency, 1990.
[16] K. Schwaber, Agile project management with Scrum. Redmond, Wash.: Microsoft Press,
2004.
[17] K. Schwaber and M. Beedle, Agile software development with scrum: Prentice Hall,
2002.
[18] J. Sutherland, A. Viktorov, J. Blount, and N. Puntikov, "Distributed Scrum: Agile Project
Management with Outsourced Development Teams," in HICSS'40, Hawaii International
Conference on Software Systems, Big Island, Hawaii, 2007.
[19] J. Sutherland, C. Jacobson, and K. Johnson, "Scrum and CMMI Level 5: A Magic Potion
for Code Warriors!," in Agile 2007, Washington, D.C., 2007.
[20] M. Striebeck, "Ssh! We're Adding a Process..." in Agile 2006 Minneapolis: IEEE, 2006.
[21] J. Sutherland, "Agile Development: Lessons Learned from the First Scrum," Cutter Agile
Project Management Advisory Service: Executive Update, vol. 5, pp. 1-4, 2004.

183
[22] K. Schwaber, "Scrum Development Process," in OOPSLA Business Object Design and
Implementation Workshop, J. Sutherland, D. Patel, C. Casanave, J. Miller, and G.
Hollowell, Eds. London: Springer, 1997.
[23] K. Beck, Extreme Programming Explained: Embrace Change. Boston: Addison-Wesley,
1999.
[24] K. Beck, "Request for SCRUM Information," J. Sutherland, Ed. Boston: Compuserve,
1995.
[25] M. Beedle, M. Devos, Y. Sharon, K. Schwaber, and J. Sutherland, "Scrum: A Pattern
Language for Hyperproductive Software Development," in Pattern Languages of
Program Design. vol. 4, N. Harrison, Ed. Boston: Addison-Wesley, 1999, pp. 637-651.
[26] A. Scotland, "Scrum @ the BBC," in JAOO, Aarhus, Denmark, 2005.
[27] G. Benefield, "Rolling Out Agile at a Large Enterprise," in HICSS'41, Hawaii
International Conference on Software Systems, Big Island, Hawaii, 2008.
[28] K. Schwaber, The Enterprise and Scrum. Redmond, WA: Microsoft Press, 2007.
[29] J. Sutherland, A. Viktorov, and J. Blount, "Adaptive Engineering of Large Software
Projects with Distributed/Outsourced Teams," in International Conference on Complex
Systems, Boston, MA, USA, 2006.
[30] C. Jones, Applied software measurement : assuring productivity and quality, 2nd ed. New
York: McGraw-Hill, 1997.
[31] M. Cohn, Agile Estimation and Planning: Addison-Wesley, 2005.
[32] P. DeGrace and L. H. Stahl, Wicked problems, righteous solutions : a catalogue of
modern software engineering paradigms. Englewood Cliffs, N.J.: Yourdon Press, 1990.
[33] F. P. Brooks, The Mythical Man Month: Essays on Software Engineering: Addison-
Wesley, 1995.
[34] W. A. Wood and W. L. Kleb, "Exploring XP for Scientific Research," IEEE Software,
vol. 20, pp. 30-36, May/June 2003.
[35] C. Larman, Agile & Iterative Development: A Manager's Guide. Boston: Addison-
Wesley, 2004.
[36] S. J. Gould, The structure of evolutionary theory. Cambridge, Mass.: Belknap Press of
Harvard University Press, 2002.
[37] M. Fowler, "Is Design Dead?," Software Development, vol. 9, April 2001.
[38] J. Sutherland, D. Patel, C. Casanave, J. Miller, and G. Hollowell, Business Object Design
and Implementation: Springer, 1997.
[39] K. Schwaber, "Scrum Development Process," in OOPSLA Business Object Design and
Implementation Workshop, J. Sutherland, D. Patel, and J. Miller, Eds. Austin, TX, 1995.
[40] J. Sutherland, "Agile Can Scale: Inventing and Reinventing SCRUM in Five
Companies," Cutter IT Journal, vol. 14, pp. 5-11, Dec 2001 2001.
[41] J. Sutherland, "Future of Scrum: Parallel Pipelining of Sprints in Complex Projects," in
AGILE 2005 Conference, Denver, CO, 2005.
[42] F. P. Brooks, "No silver bullet - essence and accidents of software engineering,"
Computer, vol. 20, pp. 10-19, April 1987.
[43] G. Booch, Object-Oriented Analysis and Design with Applications: Benjamin Cummings
Publishing Company, 1994.
[44] B. A. Ogunnaike and W. H. Ray, Process Dynamics, Modeling, and Control: Oxford
University Press, 1994.

184
[45] J. Sutherland, "ScrumWeb Home Page: A Guide to the Scrum Development Process," in
Jeff Sutherland's Object Technology Web Page Boston, 1996.
[46] K. Schwaber, "Controlled Chaos: Living on the Edge," American Programmer, April
1996.
[47] AberdeenGroup, "Upgrading to ISV Methodology for Enterprise Application
Development," Product Viewpoint, vol. 8, December 7 1995.
[48] M. Pittman, "Lessons Learned in Managing Object-Oriented Development," IEEE
Software, vol. 10, pp. 43-53, Jan/Feb 1993.
[49] G. Booch, Object Solutions: Managing the Object-Oriented Project: Addison-Wesley,
1995.
[50] I. Graham, Migrating to Object Technology: Addison-Wesley, 1994.
[51] B. W. Boehm, "A Spiral Model of Software Development and Enhancement," in
International Workshop on Software Process and Software Environments, Coto de Caza,
Trabuco Canyon, California, 1985.
[52] J. Rumbaugh, "What Is a Method?," Object Journal, vol. Oct, 1995.
[53] W. S. Humphrey, Introduction to the Personal Software Process: Addison Wesley, 1996.
[54] H. Ziv and D. Richardson, "The Uncertainty Principle in Software Engineering," in
submitted to Proceedings of the 19th International Conference on Software Engineering
(ICSE'97), 1997.
[55] P. Wegner, "Why Interaction Is More Powerful Than Algorithms," Communications of
the ACM, vol. 40, pp. 80-91, May 1997.
[56] B. Boehm, "Project Termination Doesn't Mean Project Failure," IEEE Computer, vol. 33,
pp. 94-96, 2000.
[57] D. K. I. Sobek, A. C. Ward, and J. K. Liker, "Toyota's Principles of Set-Based
Concurrent Engineering," Sloan Management Review, vol. 40, pp. 67-83, 1999.
[58] M. Fowler and J. Highsmith, "The Agile Manifesto," Dr. Dobbs, July 13 2001.
[59] W. S. Humphrey, A Discipline for Software Engineering: Addison-Wesley, 1995.
[60] Matisse Software, "The Emergence of the Object-SQL Database," Mountain View, CA
2003.
[61] W. D. Hillis, The Connection Machine. Cambridge, MA: MIT Press, 1985.
[62] S. Levy, Artificial Life : A Report from the Frontier Where Computers Meet Biology, 1st
ed. New York: Vintage, Reprint edition, 1993.
[63] C. Jakobson, "The Magic Potion for Code Warriors! Maintaining CMMI Level 5
Certification With Scrum.," J. Sutherland, Ed. Aarhus, Denmark: Agile 2007 paper in
preparation, 2006.
[64] K. E. Nidiffer and D. Dolan, "Evolving Distributed Project Management," IEEE
Software, vol. 22, pp. 63-72, Sep/Oct 2005.
[65] R. Zanoni and J. L. N. Audy, "Projected Management Model for Physically Distributed
Software Development Environment," in HICSS'36, Hawaii, 2003, p. 294.
[66] J. Barthelemy, "The Hidden Costs of Outsourcing," MITSloan Management Review, vol.
42, pp. 60-69, Spring 2001.
[67] B. Gorzig and A. Stephan, "Outsourcing and Firm-level Performance," German Institute
of Economic Research October 2002.
[68] StandishGroup, "2003 Chaos Chronicles," The Standish Group International, 2003.

185
[69] J. Sutherland, "Future of Scrum: Parallel Pipelining of Sprints in Complex Projects with
Details on Scrum Type C Tools and Techniques," PatientKeeper, Inc., Brighton, MA
May 30 2005.
[70] M. Cohn, User Stories Applied : For Agile Software Development: Addison-Wesley,
2004.
[71] C. Jones, "Programming Languages Table, Release 8.2," Software Productivity Research,
Burlington, MA 1996.
[72] M. C. Paulk, C. V. Weber, B. Curtis, and M. B. Chrissis, The Capability Maturity Model
®: Guidelines for Improving the Software Process. Boston: Addison-Wesley, 1995.
[73] M. B. Chrissis, Konrad, and Shrum, CMMI – guideline for process integration and
product improvement, 2002.
[74] D. R. Goldenson and D. L. Gibson, "Demonstrating the impacts and benefits of CMMI,"
CrossTalk, October 2003.
[75] D. D. o. Science, "Maturity of customer and suppliers in the public sector," 2006.
[76] M. C. Paulk, "Extreme Programming From a CMM Perspective," IEEE Software, vol. 34,
2001.
[77] M. C. Paulk, "Agile Methodologies and Process Discipline," CrossTalk, 2002.
[78] Surdu, "Army Simulation Program Balances Agile and Traditional Methods with
Success," CrossTalk, 2006.
[79] P. E. McMahon, "Lessons Learned Using Agile Methods on Large Defense Contracts,"
CrossTalk, 2006.
[80] D. Kane and S. Ornburg, "Agile Development: Weed or Wildflower," CrossTalk, 2002.
[81] Krasner and Houston, "Using the Cost of Quality Approach for Software," CrossTalk,
November 1998.
[82] M. Diaz and J. King, "How CMM Impacts Quality, Productivity, Rework, and the
Bottom Line," CrossTalk, March 2002.
[83] J. V. Sutherland, "ScrumMaster Certification Syllabus," Boston: Scrum Alliance, 2007.
[84] M. K. Kulpa and K. A. Johnson, Interpreting the CMMI: A Process Improvement
Approach. Boca Raton: Auerbach Publications, 2003.
[85] J. Rose, I. Aaen, and P. Axel Nielsen, "The Industrial Age, the Knowledge Age, the
Internet Age: Improving Software Management," Aalborg University 2004.
[86] J. O. Coplien, "Personal issues caused over 50% of productivity losses in the ATT Bell
Labs Pasteur Project analysis of over 200 case studies," Personal communication ed, J.
Sutherland, Ed. Lynby, Denmark, 2006.
[87] T. Sulaiman, B. Barton, and T. Blackburn, "AgileEVM - Earned Value Management in
Scrum Projects," in Agile 2006, Minneapolis, 2006.
[88] NetObjectives, "Managing the Work." vol. 2006: NetObjectives, 2005.
[89] P. M. Senge, The Fifth Discipline: The Art and Practice of the Learning Organization,
Revised Edition. New York: Doubleday, 2006.
[90] G. Rothfuss, "Metrics for Agile Methods." vol. 2005, 2003.
[91] R. Martens, "On Be(coming) Agile: SCRUM Gathering – Advanced SCRUM and Rally
as an Agile Enterprise." vol. 2005 Boulder, CO: Rally Development Corporation, 2005.
[92] T. Poppendieck, "Roots of poor management practices," J. sutherland, Ed. Boston, MA,
2006.
[93] M. Fackler, "The "Toyota Way" is Translated for a New Generation of Foreign
Managers," in New York Times New York, 2007.

186
[94] D. K. Taft, "Microsoft Lauds 'Scrum' Method for Software Projects," in eWeek, 2005.
[95] K. Johnson, C. Jacobson, and J. Sutherland, "Scrum and CMMI Level 5: A Magic Potion
for Code Warriors!," in SEPG, 2007.
[96] E. M. Goldratt and J. Cox, The goal : a process of ongoing improvement, 2nd rev. ed.
Great Barrington, MA: North River Press, 1994.
[97] M. Cohen, "Make a Bet on General Motors," in Forbes. vol. 178, 2006, p. 162.
[98] A. MacCormack, C. Kemerer, M. Cusumano, and B. Crandall, "Exploring Tradeoffs
Between Productivity and Quality in the Selection of Software Development Practices.,"
IEEE Software, pp. 78-85, 2003.
[99] I. Jacobson, Object-Oriented Software Engineering: A Use Case Driven Approach:
Addison-Wesley, 1992.
[100] J. Sutherland and W.-J. van den Heuvel, "Enterprise Application Integration and
Complex Adaptive Systems: Could System Integration and Cooperation be improved
with Agentified Enterprise components?," Communications of the ACM, vol. 45, pp. 59-
64, October 2002.
[101] J. M. Osier, B. Kehoe, and Y. Svendsen, "Keeping Track: Managing Messages with
GNATS, The GNU Problem Report Managment System Version 4.0," Free Software
Foundation 2002.
[102] J. M. Bland and D. G. Altman, "Statistics Notes: Regression towards the mean," BMJ,
vol. 308, pp. 1499-, June 4, 1994 1994.
[103] W. E. McCarthy, "The REA Accounting Model: A Generalized Framework for
Accounting Systems in a Shared Data Environment," The Accounting Review, vol. LVII,
pp. 554-578, July 1982.
[104] G. Weinberg, Quality Software Management Vol. 4: Anticipating Change: Dorset House,
1997.
[105] J. H. Holland, Hidden order : how adaptation builds complexity. Reading, Mass.:
Addison-Wesley, 1995.
[106] J. H. Holland, Emergence : from chaos to order. Reading, Mass.: Addison-Wesley, 1998.
[107] K. Kleinberg and T. Berg, "Mobile Healthcare: Applications, Vendors and Adoption,"
Gartner Group 26 November 2002.
[108] S. Group, "There's Less Development Chaos Today," in Software Development Times,
2007.
[109] H. Kniberg, Scrum and XP from the Trenches. Stockholm: Crisp, 2007.
[110] I. o. Nonaka and H. Takeuchi, The knowledge-creating company : how Japanese
companies create the dynamics of innovation. New York: Oxford University Press, 1995.
[111] E. M. Goldratt, Theory of Constraints. Burlington, MA: North River Press, 1990.
[112] W. Cunningham, "Wiki Wiki Web," http://c2.com/cgi/wiki, 1999.
[113] J. Sutherland, "Jeff Sutherland's Scrum Log," http://jeffsutherland.com/scrum/index.html,
1997.
[114] K. Schwaber, "Ken Schwabers's Scrum Web Page," http://www.controlchaos.com, 1997.

187
Appendix I:
Scrum: A Pattern Language for Hyperproductive Software
Development
Mike Beedle, Martine Devos, Yonat Sharon, Ken Schwaber, and Jeff Sutherland
Can a repeatable and defined process really exist for software development? Some think this is
not only possible but necessary, for example, those who favor the CMM (Capability Maturity
Model) approach to software development [72].
However, many of us doing work in the trenches have found over time that the repeatable or
defined process approach makes many incorrect assumptions, such as the following:
• Repeatable/defined problem. A repeatable/defined process assumes that there is a step or

steps to capture requirements , but in most cases, it is not possible to define the
requirements of an application like that because they are either not well defined or they
keep changing.
• Repeatable/defined solution. A repeatable/defined process assume that an architecture
can be fully specified, but in reality it is evolved, partly because of missing or changing
requirements (as described above), and partly because of the creative process involved in
designing new software structures.
• Repeatable/defined developers. The capabilities of a software developer vary widely, so
that a process that works for one developer may not work for another one.
• Repeatable/defined organizational Environment. Schedule pressure, priorities (e.g.
quality vs. price vs. manpower), client behavior, and so on are never repeatable, and
because they are highly subjective, they are very hard to define.
The problem with these assumptions is that these variables do have large variances. In real life
projects there are always large dynamic variations that can have a great deal of impact on the
overall project. For example, newly found changes in the requirements during an application
implementation—a typical occurrence—may affect drastically a project’s schedule that assumed
that all the requirements would be captured up front. However, removing this uncertainty is
nearly impossible because of the nearly universal sources of requirements change: business
requirements driven changes, usability driven changes, re-prioritization driven changes, testing
driven changes, and so forth. This issue cannot be solved through improved methods for
identifying the user requirements. Instead it calls for a more complex process of generating
fundamentally new operating alternatives.

188
In other words, once we accept that these dynamic variabilities do exist, we clearly need more
adaptive ways to build software. However, what we fear is that most current methods do not
allow us to build soft enough software: present methods and design paradigms seem to inhibit
adaptability. Therefore the majority of software practitioners nowadays tend to become experts
at what they can specify in advance, working with the unstated belief that there exists an optimal
solution that can be planned a priori. Once technology is adopted by an organization, it oten
becomes a constraining structure that in part shapes the action space of the user. Thus we build
software too much like we build hardware—as if it were difficult to change, as if it has to be
difficult to change. In many organizations, “The system requires it” or the “System does not
allow it” have become accepted (and often unavoidable) justifications for human behavior before
and after the system is released to production.
In contrast, Scrum allows us to build softer software, so there is no need to write full
requirements up fronts. Since the users do not know what is possible, they will ask for the pre-
tech-paper solution that they perceive to be possible (“looking at the rearview mirror”). But in
truth, not even the software developers know fully what can be built beforehand. Therefore, the
user has no concept of what is possible before he or she can feel it or touch it. As such, The
Scrum patterns presented here offer a collection of empirical techniques that assume up front the
existence of uncertainty but that provide practical and specific techniques to tame it. These
techniques are rooted in complexity management, that is, in self-organization, management of
empirical processes, and knowledge creation.
In that sense, Scrum is not only a “parallel iterative and incremental” development method, it is
also an “adaptive” software development method.
How does Scrum Work?

Scrum’s goal is to deliver as much quality software as possible within a series ( three to eight) of
short time boxes (fixed-time intervals) called Sprints that typical last about a month.
Each stage in the development cycle (Requirements, Analysis, Design, Evolution, and Delivery)
is now mapped to a Sprint or series of Sprints. The traditional software development stages are
retained primarily for convenience tracking milestones. So, for example, the Requirements stage
may use one Sprint, including the delivery of a prototype. The Analysis and Design stage may
take one Sprint each, while the Evolution stage may take anywhere from three to five Sprints.
Editors Note: In recent years, release cycles have shortened to three months or less for most
software products. Requirements are specified just enough and just in time to be ready at the
start of the Sprint cycle. Sprints produce working software for review at the end of every Sprint.
As a result, Analysis, Design, and Evolution occur in every Sprint. Sprint cycles in many
companies have been shortened to two weeks or less. In the best companies, Delivery is included
in every Sprint [41].
Unlike a repeatable and defined process approach, in Scrum there is no predefined process
within a Sprint. Instead, Scrum meetings drive the completion of the allocated activities.

189
Each sprint operates on a number of work items called a Backlog. As a rule, no more items are
externally added into the Backlog within a Sprint. Internal items resulting form the original pre-
allocated Backlog can be added to it. The goal of a sprint is to complete as much quality software
as possible, but typically less software is delivered in practice.
The end result is that there are non-perfect releases delivered every Sprint.
During a Sprint, Scrum Meetings are held daily to determine the following:
• Items completed since the last Scrum meeting..

Issues or blocks that need to be resolved. (The ScrumMaster is a team leader role
responsible for resolving the blocks.)
• New assignments the team should complete before the next Scrum meeting.
Scrum Meetings allow the development team to “socialize the team members’ knowledge” as
well as produce a deep cultural transcendence. This “knowledge socialization” promotes a self-
organized team structure within which the development process evolves on a daily basis.
At the end of each Sprint there is a demonstration to:
• Show the customer what’s going on.

• Give the developer a sense of accomplishment.
• Integrate and test a reasonable portion of the software being developed.
• Ensure real progress, that is, the reduction of Backlog, not just the production of more
paper/hours spent.
After gathering and reprioritizing leftover and new tasks, a new Backlog is formed and a new
Sprint starts. Potentially, many other organization and process patterns may be used in
combination with the Scrum patterns.
ScrumMaster Sprint Backlog Scrum Meetings Demo After Sprint
Figure: The Scrum pattern language
The Patterns
Sprint
Context
You are a software developer or a coach managing a software development team where there is a
high percentage of discovery, creativity, or testing involved.
You are building or expanding systems, which allow partitioning of work, with clean interfacing,
components, or objects.

190
Problem
You want to balance the needs of developers to work undisturbed and the need of management
and the customer to see real progress, as well as control the direction of that progress throughout
the project.
Forces
• Developers need time to work undisturbed, but they need support for logistics;
management and users need to be convinced that real progress is being made.
• Often, by the time systems are delivered, they are obsolete or require major changes. The
problem is that input from the environment is collected mostly at the start of the project,
while the user learns mostly by using the system or intermediate releases.
• It is often assumed that the development process is a well-understood approach that can
be planned and estimated. If a project fails, that is considered proof that the development
process needs more rigor. These step by step approaches, however, don’t work because
they do not cope with the unpredictabilties, both human and technical, in system
development. Therefore, at the beginning of a project it is impossible to make a complete,
detailed specification, plan, or schedule because of the many uncertainties involved.
• Overhead is often created to prove that a process is on track. Current process automation
adds administrative work for managers and developers and often results in marginally
used development processes that become disk-ware. (Misfit: Activity is not synonymous
with results. More often that not, a project plan shows activities but fails to ensure real
progress or results.
Solution
Divide the project in Sprints. A Sprint is a period of approximately 30 days in which an agreed
amount of work will be performed to create a deliverable. Each Sprint takes a pre-allocated
amount of work from the Backlog, and it is assigned to Sprints by priority and by approximation
of what can be accomplished during the Sprint’s length. In general, chunks of high cohesion and
low coupling are selected—either horizontal or vertical “packets,” that is, vertical or horizontal
components.
As a rule, nothing is added externally to the allocated Sprint Backlog during the Sprint. External
additions are only added to the global Backlog, but blocks (unresolved issues) resulting from the
Sprint can be added to the allocated Sprint Backlog. A Sprint end with a demonstration (Demo
After Sprint) of new functionality.
This gives the developers space to be creative, to learn by exploring the design space and by
doing actual work. Undisturbed by outside interruptions, they are free to adapt their ways of
working using opportunities and insights. At the same time, this keeps management and other
project stakeholders confident by showing real progress instead of documents and reports
produced as proof of progress.

191
The net result is that each sprint produces a visible and usable deliverable that is shown to the
users at the demo (Demo After Sprint). An increment can be either intermediate of shippable, but
it should stand on its own. The goal of a Sprint is to complete as much quality software as
possible and to ensure real progress, not paper milestones as alibis.
Editors Note: It is impossible to overestimate the positive impact this strategy has on software
development. Iterations that demonstrate early working software in order to incorporate real-
time user feedback have increase project success industry-wide from 16.2% in 1994 to 35% in
2006. This increased industry-wide return on dollar invested in software from 25 cents in 1998
to 59 cents in 2006 for a compound annual growth rate of 24% [108].
Rationale
• The fact that no items are added to the Backlog externally allows development to
progress “full speed ahead,” without needing to think about changes in direction.
• The fact that developers are not “tested” during the Sprint is empowering.
• The ability to choose a process per Sprint is empowering and enables adaptation to
changing circumstances (different developers, different project phases, more knowledge,
etc.)
• Sprints are short; therefore, the problem of completing a Sprint is much simpler that that
of completing a project. It is easier to take up this smaller challenge.
• Developers get feedback frequently (at the end of each Sprint). They can therefore feel
there successes (and failures) without compromising the whole project.
• Management has full control—it can completely change direction at the end of each
Sprint.
• The end users are deeply involved throughout the development of the application through
the Demos after the Sprints, but they are not allowed to interfere with the day-to-day
activities. Thus ownership and direction still belong to the users but without their
constant interference.
• Project status is visible since the Sprint produces working code.
Known Uses
At Argo, the Flemish department of education, we have been using Sprints since January 1997
on a large number of end-user projects and for the development of a framework for database,
document management, and workflow. The Backlog is divided in sprints that last about a month.
At the end of each Sprint, a working Smalltalk image is delivered with integration of all current
applications. The team meets daily in Scrum Meetings, and Backlog is re-prioritized after the
Demo in a monthly meeting with the steering committee.
Resulting Context
The result is a high degree of “effective ownership” by the participants, including users who stay
involved through the Demos and the prioritizing of the Backlog. “Effective ownership” in this
case means both empowerment and the involvement of all the participants.

192
At the end of a Sprint, we have the best approximation of what was planned at the start. In a
review session, the supervisors have the opportunity to change the planning for the future. The
project is totally flexible at this point. Target, product, delivery date, and cost can be redefined.
With Scrum we get a large amount of post-planning flexibility (for both customer and
developer).
It may become clear in the daily Scrum Meetings throughout the Sprint that some team members
are losing time at non- or less productive tasks. Alternatively, it may also become clear that
people need more time for the ir tasks than originally allocated by management. Developers may
turn out to be less competent or experienced at the allocated task than assumed, or they may be
involved in political or power struggles. The high visibility of scrum, however, allows us to deal
with these problems. This is the strength of the Scrum method manifested through the Scrum
Meetings and the Sprints.
Difficulties in grouping Backlog for a Sprint may indicate that priorities are not clear to
management or to the customer.
Backlog
Context (From: Sprint)
You are connected to a software project or any other project that is chaotic in nature that needs
information on what to do next.
Problem
What is the best way to organize the work to be done next and at any stage of the project?
Forces
Traditional planning methods like Pert and Gantt assume that you know in advance all the tasks,
all their dependencies, all task durations, and all available resources. These assumptions are
wrong if the project involves any learning, discovery, creativity, or adaptation.
Solution
Use a Backlog to organize the work of a Scrum team.
The Backlog is a prioritized list. The highest priority Backlog item will be worked on first, the
lowest priority Backlog item will be worked on last. No feature, addition, or enhancement to a
product is worth fighting over; it is simply either more important or less important at any time to
the success and relevance of the product.
Backlog is the work to be performed on a product. Completion of the work will transform the
product from its current form into its vision. But in Scrum, the Backlog evolves as the product

193
and the environment in which it will be used evolves. The Backlog is dynamic, constantly
changed by management to ensure that the product defined by completing the Backlog is the
most appropriate, competitive, useful product possible.
There are many sources for the Backlog list. Product marketing adds work that will fulfill their
vision of the product. Sales adds work that will generate new sales or extend the usefulness to the
installed base. Technology adds work that will ensure the product uses the most innovative and
productive technology. Development adds work to enhance product functions. Customer support
adds work to correct underlying product defects.
Only one person prioritizes work. This person is responsible for meeting the product vision. The
title usually is product manager or product marketing manager. If anyone wants the priority fo
work changed, they have to convince this person to change that priority. The highest priority
Backlog has the most definition. It is also prioritized with an eye toward dependencies.
Depending on how quickly products are needed in the marketplace and the finances of the
organization, one or more Scrum Teams work on a product’s Backlog. As a Scrum Team is
available (newly formed or just finished a Sprint) to work on the Backlog, the team meets with
the product manager. Focusing on the highest priority Backlog, the team selects a subset of the
Backlog the team believes it can complete within a Sprint iteration (30 days or less). In doing so,
the Scrum Team may alter the Backlog priority by selecting a Backlog that is mutually
supportive, that is, one that can be worked on at once more easily than by waiting. Examples are
multiple work items that require developing a common module or interface and that make sense
to include in one Sprint.
The team selects a cohesive group of top priority Backlog items that, once completed, will have
reached an objective, or milestone. This is stated as the Sprint’s objective. During the Sprint, the
team is free to not do work as long as this objective is reached.
The team now decomposes the selected Backlog into tasks. These tasks are discrete pieces of
work that various team members sign up to do. Tasks are performed to complete Backlog to
reach the Sprint objective.
Resulting Context
Projec work is identified dynamically and prioritized according to:
1. The customer needs

2. What the team can do
Scrum Meetings
Context (From: Backlog)
You are a software developer or a coach managing a software development team where there is a
high percentage of discovery, creativity, or testing involved. An example is a first time delivery

194
where the problem has to be specified, an object model has to be created, or new or changing
technologies are being used.
Activities such as scientific research, innovation, invention, architecture, engineering and a

myriad of other business situations may also exhibit this behavior.
You may also be a “knowledge worker,” an engineer, a writer, a research scientist, or an artist, or
a coach or manager who is overseeing the activities of a team in these environments.
Problem
What is the best way to control an empirical and unpredictable process such as software
development, scientific research, artistic projects, or innovative designs where it is hard to define
the artifacts to be produced and the processes to achieve them?
Forces
Estimation
• Accurate estimation for activities involving discovery, creativity, or testing is difficult

because it typically involves large variances, and because small differences in
circumstances may cause significant differences in results. These uncertainties come in at
least five flavors:
1. Requirements are not well understood.
2. Architectural dependencies are not easy to understand and are constantly changing.
3. There may be unforeseen challenges with the technology. Even if the challenges are
know in advance, their solutions and related effort are not known.
4. There may be bugs that are hard to resolve in the software; therefore, it is typical to
see project estimates that are several orders of magnitude off. You can’t “plan bugs,”
you can only plan bug handling and provide appropriate prevention schemes based on
the possibility of unexpected bugs.
Example: You Got theWrong Number. In projects with new or changing
requirements, a new architecture, new or changing technologies, and difficult
bugs to weed out, it is typical to see project estimates that are off by several
orders of magnitude.
5. On the other hand, estimation is important. One must be able to determine what are
the future tasks within some time horizon and prepare resources in advance.
Planning
• Planning and reprioritizing tasks takes time. Involving workers in time planning meetings
decreases productivity. Moreover, if the system is chaotic, no amount of planning can
produce uncertainties.
Example: Paralysis by Planning. Some projects that waste everyone’s time in
planning everything to an extreme detail but are never able to meet the plans.

195
• A plan that is too detailed become large and is hard to follow; the larger the plan is, the
more errors it will contain (or at the very least the cost of verifying its correctness grows).
Example: The Master Plan Is a Great Big Lie. Many projects that try to follow a
master plan fall into the trap of actually believing their inaccuracies and often face
disappointment when their expectations are not met.
• No planning at all increases uncertainty among team members and eventually damages
morale.
Example: Lost Vision. Projects that never schedule anything tend to lose control
over their expectations. Without some schedule pressure no one will do anything,
and worse, it will become difficult to integrate the different parts being worked on
independently.
Tracking
• Too much montoring wastes time and suffocates developers.

Example: Measured to Death. Projects that waste everybody’s time in tracking
everything to an extreme detail but are never able to meet the plans. (You
measure the tire pressure until all the air was out!)
• Tracking does not increase the certainty of indicators because of the chaotic nature of the
system. In fact, trying to control normal variations of a system will cause wide
oscillations of the system, rendering it more chaotic.
• Too much data is meaningless—the Needle in the Haystack Syndrome.
• Not enough monitoring leads to blocks and possible idle time between assignments.
Example: What Happened Here? Projects that never track anything tend to lose
control over what is being done. Eventually no one really knows what has been
done.
Solution
To provide for accurate estimates, plans, and appropriate tracking, meet with the team members
for a short time (~15 minutes) in a daily Scrum Meeting, where the only activity is asking each
participant the following three questions:
1. What have you worked on since the last Scrum Meeting? The ScrumMaster logs the
tasks that have been completed and those that remain undone.
2. What blocks, if any, have you found in performing your tasks within the last 24
hours? The ScrumMaster logs all blocks and later finds a way to resolve the blocks.
3. What will you be working on in the next 24 hours? The ScrumMaster helps the team
members choose the appropriate tasks to work on with the help of the Architect.
Because the tasks are scheduled on a 24-hour basis, the tasks are typically small
(Small Assignments).
This will provide you with more accurate estimates, short-term plans, appropriate tracking, an
correcting mechanisms to react to changes and adapt every 24 hours.

196
Scrum Meetings typically take place at the same time and place every day, so they also serve to
build a strong culture. As such, Scrum meetings are rituals that enhance the socialization of
status, issues, and plans for the team. The ScrumMaster leads the meetings and logs all the tasks
from every member of the team into a global project Backlog. He also logs every block and
resolves each block while the developers work on other assignments.
Editors note: The Scrum Board has emerged as a best practice for a team to manage their own
tasks. Teams meet in front of the Board which has multiple columns. The first column has User
Stories from the Product Backlog (features to be delivered) on large cards prioritized in order of
business value. At the start of the Sprint, the tasks to be accomplished for a User Story are in the
left column as small cards. Each day developers move tasks to an “In Progress” column, then to
a “Validation” column, then to a “Done” column. Estimates are updated on tasks daily and the
Burndown Chart can easily be calculated and posted to the board [109].
The blocks logged by the ScrumMaster are now known as the “Impediment List” which needs to
be prioritized. The block which is the most critical constraint to system throughput should be at
the top of the list and the ScrumMaster should work on that one first. Tuning a development
project is similar to tuning a computer system. It may not be obvious where the critical
constraint lies and careful analysis may be required. The main choke point must be found and
fixed first. The development system as a whole should then be allowed to stabilize and measured.
The next critical block after restabilization may be in an unexpected place. That should be fixed
next. Fixing too many things at once generates waste by fixing constraints that have minimal
impact on throughput. This uses critical resources to change things that do not dramatically
improve velocity. It makes it difficult to clarify system dynamics and tires out and demotivates
the team, management, and the company.
Scrum meeting not only schedule tasks for developers, but can and should schedule activities for
everyone involved in the project, such as integration personnel dedicated to configuration
management, architectes, ScrumMasters, or a QA team.
Scrum Meetings allow knowledge workers to accomplish mid-term goals typically allocated in
Sprints that last a month or less.
Scrum Meetings can also be held by self-directed teams. In that case, someone is designated as
the scribe and logs the completed activities of the Backlog and the existing blocks. All activities
from the Backlog and the blocks and then distributed among the team for resolution.
The format of the Backlog and the blocks can also vary, ranging from a list of items on a piece of
paper, to software representations of it over the Internet/Intranet [17]. The Scrum Meeting’s
frequency can be adjusted and typically ranges between 2 and 48 hours.
These meetings are often held standing up. This ensures that the meetings are kept short and to
the point.

197
Rationale
It is very easy to over- or under-estimate, which leads either to idle developer time or to delays in
completion of an assignment. Therefore, it is better to frequently sample the status of small
assignments. Projects with a high degree of unpredictability cannot use traditional project
planning techniques such as Gantt or PERT charts only, because the rate of change of what is
being analyzed, accomplished, or created is too high. Instead, constant reprioritization of tasks
offers an adaptive mechanism that provides sampling of systemic knowledge over short periods
of time.
Scrum Meetings help also in the creation of an “anticipating culture” [104] because they
encourage these productive values:
• They increase the overall sense of urgency.

• They promote the sharing of knowledge.
• They encourage dense communications.
• They facilitate honesty among developers since everyone has to give a daily status.
This same mechanism encourages team members to socialize, externalize, internalize, and
combine technical knowledge on an ongoing basis, thus allowing technical expertise to become
community property for the community of practice [110]. Scrum Meetings are therefore rituals
with deep cultural transcendence. Meeting at the same place at the same time, and with the same
people, enhances a feeling of belonging and creates the habit of sharing knowledge.
Seen from the System Dynamics point of view [89], software development has a scheduling
problem because the nature of programming assignments is rather probabilistic. Estimates are
hard to come by because:
• Inexperienced developers, managers, and architects are involved in making the estimates.
• There are typically interlocking architectural dependencies that are hard to manage.
• There are unknown or poorly documented requirements.
• There are unforeseen technical challenges.
As a consequence, the software development becomes a chaotic beer game, where it is hard to
estimate and control the inventory of available developer’s time, unless increased monitoring of
small assignments is implemented [89, 111]. In that sense the Scrum Meeting becomes the
equivalent of the thermometer that constantly samples the team’s temperature.
From the Complexity Theory perspective [105, 106], Scrum allows flocking by forcing a faster
agent interaction, therefore accelerating the process of self-organization because it shifts
resources opportunistically through the daily Scrum Meetings.
This is understandable, because the relaxation of the self-organized multi-agent system is

proportional to the average exchange among agents per unit of time. And in fact, the “interaction
rate” is one of the levers one can push to control “emergent” behavior—it is like adding an
enzyme or catalyst to a chemical reaction.
198
In Scrum this means increasings the frequency of the Scrum Meetings, and allowing more
hyperlinks as described below, but up to an optimal upper-frequency bound on the Scrum
Meetings (meetings/time), and up to an optimal upper bound on the hyperlinks or the Scrum
Team members. Otherwise the organization spends too much time socializing knowledge,
instead of performing tasks.
Known Uses
(Mike Beedle) At Nike Securities in Chicago we have been using Scrum Meetings since
February 1997 to run all of our projects including BPR and software development. Everyone
involved in these projects receives a week of training in Scrum techniques.
(Yonat Sharon) At Elementrix Technologies we had a project that was running way late after
about five months of development. Only a small part (about 20 percent) was completed, and
even this part had too many bugs. The project manager started running bi-daily short status
meetings (none of us was familiar with the term Scrum back then). In the following month, the
entire project was completed and the quality had risen sharply. Two weeks later, a beta version
was out. The meetings were discontinued, and the project hardly progressed since. I don’t think
the success of the project can be attributed to the Scrum Meetings alone, but they did have a big
part in this achievement.
One of my software team leaders at Rafael implemented a variation of Scrum Meetings. He

would visit each developer once a day, and ask him the three questions; he also managed a
Backlog. This does not have the team building effects, but it does provide the frequent sampling.
C3 and Vcaps projects (described on wiki [112]) also do this. (BTW, I adopted this name in
Hebrew, since in Hebrew “meeting” is “sitting,” and so we say “standup sitting”.)
Resulting Context
The application of this pattern leads to:
• Highly visible project status

• Highly visible individual productivity
• Less time wasted because of blocks
• Less time wasted because of waiting for someone else
• Increased team socialization
Conclusion
Scrum is a knowledge creating process with a high level of information sharing during the whole
cycle and work progress.
The key to Scrum is pinning down the date at which we want completion for production or
release, prioritizing functionality, identifying available resources, and making major decisions
199
about architecture. Compared to more traditional methodologies, the planning phase is kept short
since we know that events will require changes to initial plans and methods. Scrum uses an
empirical approach to development where interaction with the environment is not only allowed
but encouraged. Changing scope, technology, and functionality are expected; and continuous
information sharing and feedback keeps performance and trust high.
Its application also generates a strong culture with well-defined roles and relationships, with
meaningful and transcending rituals.
Acknowledgements
We would like to thank all of the Scrum users and reviewers from whom we have received
feedback over the years. Also, we thank all of the members of the Chicago Patterns Group that
attended an early review session of the Scrum Meeting pattern (especially Brad Appleton, Joe
Seda, and Bob Haugen). Finally we thank our PloP’98 shepherd, Linda rising, for providing us
comments and guidance to make our paper better.
(Personal acknowledgement from Mike Beedle.) I’d like to thank both Jeff Sutherland [113] and
Ken Schwaber [114] for adapting the Scrum techniques to software in the early 1990s, and for
sharing their findings with me. Scrum has made a significant contribution to the software
projects in which I used the technique.

200
Index
Agile Manifesto, 7, 9, 134, 149 111, 112, 113, 114, 115, 123, 124, 127,
Borland Quattro Pro, 6, 96 138
business plan, 14, 114 Product Specifications, 80
Capers Jones, 6, 85, 86, 149 Pull systems, 14
Certified ScrumMasters, 8, 104, 106, 107 Quality Assurance, 19, 108, 120
Chief Engineer, 7, 8 Rodney Brooks, 6
CMMI, 7, 9, 59, 74, 86, 104, 133, 147, 149, Scrum Alliance, 10, 11, 12, 70, 74, 81, 87,
150 88, 99, 107
Dupont, 9 Scrum and XP, 7, 81
Fuji-Xerox, 32, 39, 70, 100 Scrum Roles, 2, 22
Gabrielle Benefield, 19 ScrumMaster, 2, 6, 7, 8, 12, 14, 15, 16, 17,
Gantt charts, 19, 92, 125, 126, 127, 162 22, 24, 26, 28, 37, 38, 58, 61, 78, 79, 99,
Hitotsubashi Business School, 7 102, 104, 111, 113, 114, 115, 118, 120,
Honda, 7, 32, 34, 39, 63, 70, 72, 100, 147 133, 153, 157
Jeff McKenna, 6, 8, 62, 99 SirsiDynix, 7, 58, 70, 73, 74, 75, 76, 77, 78,
Jim Coplien, 6 79, 80, 82, 83, 84, 85, 96, 133
John Scumniotales, 6, 8, 37, 99 Sprint Backlog, 7, 8, 12, 14, 15, 16, 25, 26,
Kent Beck, 7, 38 35, 36, 60, 84, 88, 106, 111, 113, 129
Managers, 22, 26, 150 Sprint Cycle, 13
microenterprise development, 6 Sprint Planning, 2, 12, 13, 14, 15, 24, 25,
Microsoft Project., 19 28, 29, 82, 100
Object Studio, 7, 8 Sprint Retrospective, 2, 17, 28
OOPSLA’95, 7, 8, 38 Sprint Review, 2, 12, 13, 16, 17, 28, 29
Pasteur Project, 6 StarSoft, 58, 70, 75, 76, 77, 78, 83, 84, 133
PatientKeeper, 1, 3, 9, 33, 38, 60, 62, 68, 69, Starting Scrum, 2, 23
84, 86, 98, 99, 108, 112, 113, 114, 116, subsumption architecture, 6, 72
117, 118, 121, 123, 124, 125, 126, 127, Takeuchi and Nonaka, 6, 7, 12, 34, 39, 40,
130, 149 72, 100, 102, 106, 108
Pete Deemer, 19 Team Members, 22, 28, 31
Product Backlog, 7, 12, 14, 16, 17, 23, 24, Toyota, 7, 8, 17, 32, 36, 60, 66, 70, 72, 86,
25, 26, 28, 29, 35, 36, 58, 60, 88, 93, 94, 99, 100, 101, 102, 103, 104, 105, 147, 150
99, 106, 107, 110 Waterfall, 9, 19, 20, 42, 43, 44, 46, 48, 84,
Product Owner, 2, 4, 6, 7, 8, 12, 13, 14, 16, 96
22, 23, 24, 25, 26, 28, 29, 30, 37, 58, 73,
78, 79, 80, 81, 94, 99, 100, 102, 107, 109,
[i] Controlchaos website - http://www.controlchaos.com/

about/burndown.php
[ii] Mike Cohn, “Agile Estimating and Planning”, pp. 35-42.
[iii] http://www.retrospectives.com/
pages/whatIsARetrospective.html

201
[iv] XP.org website - http://www.extremeprogramming.org
/rules/standupmeeting.html
[v] http://www.satisfice.com/articles/what_is_et.htm
[vi] Controlchaos website - http://www.controlchaos.com
[vii] Mountain goat website - http://www.mountaingoat

software.com/Scrum/Scrumteam.php

202

The Scrum Papers:: Nuts, Bolts, and Origins of An Agile Process

Uploaded by

Copyright:

Available Formats

The Scrum Papers:: Nuts, Bolts, and Origins of An Agile Process

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Scrum Papers:: Nuts, Bolts, and Origins of An Agile Process

Uploaded by

Copyright:

Available Formats

The Scrum Papers:

Nuts, Bolts, and Origins of an Agile Process

Jeff Sutherland, Ph.D.

Jeff Sutherland, Ph.D.

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 1

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 2

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 3

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 4

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 5

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 6

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 7

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 8

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 9

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 10

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 11

• Increase speed of development

Jeff Sutherland - Boston, MA USA, 2007

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 12

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 13

• Roles: Product Owner, ScrumMaster, Team.

The Product Owner has the following responsibilities

• Define the features of the product;

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 14

The Team is cross functional

• Seven plus/minus two members

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 15

Sprint Planning Meeting

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 16

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 17

Sprint Burndown Chart

Sprint Review Meeting

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 18

Continuing the Scrum

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 19

Traditional Software Development

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 20

Agile Development and Scrum

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 21

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 22

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 23

Figure 2. The Product Backlog

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 24

Figure 3. Estimating Available Hours

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 25

Figure 4. Sprint Backlog

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 26

Daily Standup Meeting

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 27

Figure 5. Daily Estimates and Burndown Chart

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 28

Starting the Next Sprint

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 29

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 30

Results From Scrum

• Productivity: 68% of respondents reported Scrum is better or much better (4 or 5 on a 5-

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 31

© 2006 Pete Deemer and Gabrielle Benefield

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 32

Draft 10/14/2007 ©Jeff Sutherland 1993-2007 33