ARTIFICIAL INTELLIGENCE

POWERED BANKING
 Preface
Artificial Intelligence has been around since 1956. Over the following
decades, it continued to be a topic of fascination for corporates,
but it never really found its way into real world applications.
However, recent developments and maturity of certain underlying
technologies meant that AI powered applications became
commercially viable. A new generation of emerging businesses
and the progressive conventional ones across industries saw this
as an opportunity to integrate AI as part of their value proposition.
This led to consumers being exposed to a slew of smart virtual
assistants, be it Alexa, Siri, Google Home or Amelia redefining
how they found answers to some of their everyday questions.
Amazon recommended a product that would best match their
taste or interest. Even businesses in the industries like healthcare,
manufacturing, aerospace and agriculture found that AI offered a
better way to operate their business.

The banking domain has not been different here. AI driven start-
up ventures are looking to redefine banking and progressive banks
have launched AI based pilots, be it in the space of customer services,
fraud management, or credit scoring, among others. These ventures
and pilots have sprung up because AI powered banking is viable
now and it is also seeing acceptance among end consumers. All this
has led to banking business and technology leaders agreeing that
artificial intelligence is among the hottest banking trends that will
reshape banking in 2017.

While there are multiple point of views published every day

reiterating this belief, there is a lack of documentation giving a clear
direction to a bank on how they should go about their AI journey.
This point of view, put together by the Banking Visionaries Council
instituted by Infosys Finacle, is an attempt to bridge this gap. This
council brings together a select group of senior business and
technology leaders from global banking community with a singular
purpose - Solve most pertinent problems with the research and
collective thought leadership efforts. The objective of the paper
is to serve banks as a practical guide in their AI journey. We hope
banks find this research useful in crafting their organization’s AI
adoption strategy.

2 I External Document © 2017 EdgeVerve Systems Limited

The Meaning of Artificial
Intelligence
In concept, Artificial Intelligence (AI) has been around for decades,
ever since John McCarthy defined it as “the science and engineering
of making intelligent machines”. But it is only lately that AI
technology has undergone rapid evolution and consequently
sparked significant interest among enterprises in virtually every
industry. Today, there is widespread agreement that AI is one of the
hottest trends for 2017.

However, there is less agreement on what AI actually means. This is

because AI is not one, but a group of related technologies, which
includes among others, big data analytics, machine learning, deep
learning, predictive/prescriptive analytics, virtual agents, and
avatars (which understand natural language). The fact that
everything from robotic process automation to actual robotics falls
under that umbrella only complicates the understanding of AI even
further.

Actually, Artificial Intelligence is all of these things. When a

computer system simulates a process, such as thinking or sensing,
which is one of the building blocks of human intelligence, it needs
AI to do so.

This Point of View focusing on the commercial application of AI in

the banking industry assumes the following definition for AI:

Artificial Intelligence is an area of computer science that emphasizes

the creation of intelligent machines that sense, comprehend,
reason and act to emulate human behavior. Some of the activities
that computers with AI are designed for include image and speech
recognition, learning, planning and problem-solving. Examples of
applied AI technologies include (but are not limited to): machine
learning, deep learning, predictive/prescriptive analytics, virtual
agent and natural language understanding technologies (Siri, Alexa,
Google Home, Amelia etc.).

External Document © 2017 EdgeVerve Systems Limited I 3

 The Drivers of Adoption
AI is poised at a point of inflection, where it is mature enough to such as big data, process automation and cloud solutions, which is
step out of the lab and enter the real world. In parallel, banks have a precondition for a successful foray into AI.
also attained a degree of maturity in building digital components,

The graphic below summarizes the business and technology factors driving AI adoption today.

Business factors driving AI adoption

A generation of consumers exposed to Siri &
Amazon recommendations

Need for extensive automation with intelligent

processes to stay competitive on profitability

Progressive banking incumbents, challenger

banks & payment providers, leading the way
in AI adoption

AI’s ability to process big data, recognize speech, images,

text, and patterns is the ticket to personalization at scale

Technology factors driving AI adoption

AI is taking up where big data left off to give

enterprises a real chance to extract value from
their idle data resources

Open banking has led to more comprehensive

customer data, banks can build better models,
and more intelligent apps and services

Better access to computing power & cloud

helps build and run intelligent applications

Maturity from descriptive to prescriptive and

predictive analytics to contribute to evolution of AI

Readily available, open-source – hence affordable –

AI platforms are playing a major role in AI adoption

4 I External Document © 2017 EdgeVerve Systems Limited

Forces Driving AI Deployment
44 percent of executives participating in an Economist survey said This places a great deal of responsibility on senior leadership to
delaying AI implementation would make their business vulnerable drive AI adoption within their organizations. The below graph
to disruption at the hands of startup companies. When Infosys depicts how respondents to the Infosys survey rate the drivers of AI
reached out to 1,600 IT and business decision makers, three out of deployment: clearly considerations such as gaining a competitive
four said that AI was fundamental to the success of organizational advantage, drive by executives and solving business problem are
strategy. Those currently using or planning to use AI technology rated as the key drivers for adoption of AI. This goes to say that the
anticipated revenues to go up 39% on average by 2020. drivers are mostly top down currently.

Competitive advantage

Executive-led decision

Solution to particular business/operational/technical problem

Indicates
Internal experiment a top down
approach

Customer demands

Unexpected solution to problem

Offshoot of another project

Source: Infosys research on AI maturity “Amplifying Human Potential”

External Document © 2017 EdgeVerve Systems Limited I 5

 Although the banking and financial
services sector is showing interest
in AI, our research found that it is
clearly not very mature in its journey
to adoption, coming in at the 8th
position. This is surprising considering
that financial services is a data
intensive business. Our research,
which covered respondents from 10
vertical groups, tried to assess their
respective progress in the AI journey
with the help of a maturity index, that
is depicted in the graph below. On an
average, most banks are explorers,
with AI related skills on the increase
and more initiatives planned in the
coming 12 months.

Comparison with other industries

58
52 54
47 50 50 50 51
41
Fast moving consumer goods

32 Pharmaceuticals/life sciences
Automotive and aerospace
Energy, oil gas and utilities
Financial services

Manufacturing
Public sector

Healthcare

Telecoms
Retail

Percentages indicate average maturity score by industry

6 I External Document © 2017 EdgeVerve Systems Limited

The AI Stack
AI applications have proliferated over the past few years, and today just consumer-facing businesses that are taking an active interest in
there are more than a hundred well-known applications for different AI; from aerospace to manufacturing, and healthcare to public sector,
industries. These have subtly permeated human life in recent years every industry is in the fray. It is important to understand that all these
without our realizing it. Leading this is the virtual agent, better known applications and business use cases have a combination of technologies
as Alexa, Siri, Google Home or Amelia, which answers simple queries under the artificial intelligence umbrella. The schematic below shows a
and executes basic tasks. The recommendation engine probably marks broad AI stack consisting of AI building blocks and applied AI solutions,
consumers’ first brush with AI, as they consulted Amazon on what to which go into making business use cases. The next few pages defines
read next. And Pittsburgh residents got their first taste of a driverless each of these technologies, their applications in banking along with case
car service last December when Uber launched its trial. But it’s not examples of banks leveraging these technologies for the AI initiatives.

AI infused
business use Customer Sales and Fraud Financial
...
Service Marketing Management Advisory
cases

Applies AI Virtual Experts

RPA Robots ...
Solutions Assistants Advisors

Natural Natural
AI Building Machine Deep
Language Language Visual
Blocks Learning Learning
Processing Generation Perception

AI Foundation Data and Analytics

Broadly speaking, an organization will use AI to do one or all of

the following
1. Observe and Sense: View what’s happening (Emulate the sensing 3. Interact and Act-Interact: with the human / machine and take
aspect of human behavior) action (Emulate the action aspect of human behavior)
2. Interpret and evaluate: Create hypothesis and evaluate whether The schematic below lists these functions and the AI
the hypothesis is right or wrong. Accordingly decide and choose technologies enabling them.
the best response (Emulate the thinking aspect of human behavior)

Observe and sense Interpret and evaluate Interact and act

• Natural Language • Machine learning Natural Language
Processing • Deep learning Generation (NLG)
• Speech Recognition
• Visual Perception

External Document © 2017 EdgeVerve Systems Limited I 7

 A Closer Look at the AI Stack
1. AI Foundation: Data and Analytics
Digitization has been a huge factor in the creation of big data – data • Sales and marketing is another popular application. Traditionally,
that is both structured and unstructured, appears in formats ranging banks have had access to all the financial information about their
from text to speech to video to gesture, and originates in a variety of customers. Augmenting this with a consumer’s behavior, banks
sources. Modern Artificial Intelligence platforms owe their genesis are well placed to personalize customer communication and offers.
to the evolution of data/ process automation technologies, and to • For credit scoring, banks need to expand the number and quality
the quest for a viable way to manage huge amounts of data and of data sources required to assess customers’ creditworthiness
massive numbers of processes. Compared to humans, machines are based on varied data, including past behavior.
much more competent at processing and deducing patterns from
big data, including text, images and speech, from various sources. Case example: Fidor Bank
That makes them a natural fit for banking, which is both data and Fidor Bank leverages data analytics to create a community rating,
process-heavy. Fidor Karma, based on a customer’s activities, connections and
interactions. This enables Fidor to offer products that are linked
Popular applications of big data & analytics in banking:
to customer behavior. Fidor Karma creates a banking profile by
• Fraud detection and prevention is one of the popular use integrating community contribution, raised questions, answered
cases for big data in banking. Banks can now access millions of
questions, social media profiles and connections with other
transactions and non-traditional data sources to identify suspicious
community users.
activity and fraud.

8 I External Document © 2017 EdgeVerve Systems Limited

2. AI Building Blocks
In this section, we will briefly talk about the six key technology 2.2 Deep Learning
building blocks for AI powered solutions. Some of these Deep Learning is a subset of machine learning, accomplished
technologies are inter-connected, as you will discover while going through a hierarchy of artificial neural networks, which resemble
through this section. human brain architecture, complete with a web of neuron nodes.
Where traditional programs take a linear approach to building
2.1 Machine Learning
analyses, deep learning systems mimic the human brain and its
Machine Learning is simply, the ability of computers and other
non-linear style of working.
smart machines to learn without being “taught” or programmed.
It is at its heart, computer programs which change with data. The
Popular applications of deep learning in banking:
machine learning process has some similarities to data mining.
Like machine learning, deep learning again has multitude of ways
Both search and identify patterns from data, but where data mining
in which it can be applied in banking processes. For instance, deep
presents the findings to human beings for their attention, machine
learning can be used to check fraud. The traditional approach bases
learning adjusts its program and actions on its own.
its judgment largely on the amount of the transaction, while deep
learning also considers typical behavioral patterns of the users too.
Popular applications of machine learning in banking:
PayPal is a good example of an organization that claims to use deep
In banking, machine learning finds application in pretty
learning to combat fraud.
much everything such as customer service, personal
financial/ wealth management, and fraud/risk management.
Deep learning is also useful for finding new business opportunities.
With an increasing amount of business promotion happening
For instance, machine learning can be used to identify fraud
in social media, deep learning can help banks access relevant
or proactively assess its payment systems’ vulnerability. A
customer information and behavior on social networks to identify
computational algorithm will process the payment transactions
opportunities from their likes and preferences. Several internet
under assessment, identify patterns from the data, and flag any
majors like Amazon and Alibaba are trying to leverage deep learning
inconsistencies or anomalies. Basis feedback from human on the
to make relevant offers to the users in real-time.
anomalies identified, the program readjusts its logic dynamically.
Machine learning’s impact can be felt everywhere – front, middle,
The below schematic offers a useful comparison of these different
or back office – by way of fewer errors, higher efficiency, better
approaches:
decisions, and great customer experience.

Rule based approach

For instance, transaction amount
Traditional Methods triggers an alert.

Model based approach

Fraud Machine Learning
Computational algorithm that builds patterns
Management based on users spending and flags off anomalies

Deep Learning Non linear approach-Multi layered evaluation

Check for transaction amount
Check for user IP
Check for location
Check for linkages with bad actors
Check for KYC ratings

External Document © 2017 EdgeVerve Systems Limited I 9

 A major benefit of course is better decision making, closer to the When it was launched a year ago, the Virtual Assistant could
human variety. Deep learning can be implemented for recollecting anticipate and answer about 10,000 typical queries, and was
previous interactions that have a bearing on the current decision learning more each day. Another example of AI in use at digibank
that assist in drawing conclusions independently while rapidly is the intelligent budget optimizer, which helps customers budget,
processing large quantities of data from highly diverse sources. track and analyze expenses. It is smart enough to understand
customer preferences and provide suitable recommendations, for
2.3 Natural Language Processing (NLP) instance, based on spending habits, suggest the right marketing
Natural Language Processing (NLP) is a technique computers use offers, or guide overspending customers on how to manage their
to analyze, understand, and make sense of text and human language. resources better.
Developers can leverage NLP to organize and structure knowledge
for automatic summarization and answering, translation, speech 2.3.1 Speech Recognition
or entity recognition, extracting relationship, text mining, and
Speech Recognition technology endows machines and programs
sentiment analysis. Important use cases include consumer sentiment
with the ability to identify words and phrases used in spoken
analysis and in virtual agents and intelligent bots.
language and change them into a machine-readable format.
Early stage speech recognition software had limited vocabulary,
Popular applications of NLP in banking:
and could only identify words that were spoken very clearly.
Sentiment Analysis: Banks can use NLP to discover and parse
Current solutions are much more sophisticated and are able to
customer sentiment about their offerings and brands from social
understand natural speech.
media conversations.

Acoustic and language modeling algorithms are what make speech

Virtual Assistants and Intelligent Bots: The largest bank in the
recognition possible. The former represents how units of spoken
United States offers an NLP solution with customized content to
language and audio signals are related, while the latter matches
clients. There is also a virtual agent supporting the loans, banking
sounds with word sequences to tell similar sounding words apart.
and credit cards sections of the website. Capital One has a chatbot
called Eno, which uses NLP to render consistent, personalized
Popular applications of speech recognition in banking:
service to customers.
Speech recognition technologies can help banks in providing
a frictionless customer service and enable more efficient
NLP can lead to faster and more efficient customer service
authentication. Barclays offers a voice banking facility to customers,
rendered through AI-based digital assistants. Eventually, the
which, by recognizing customers’ unique voice patterns obviates
system would learn enough from customer and agent behavior the need for other security measures. HSBC and First Direct are
to resolve certain issues automatically. among those offering a voice and touch ID option to customers.

Speech recognition technology, with its ease of use, is a natural

Case example: DBS Bank
fit for securing mobile payments, wearable devices and devices
Singapore’s DBS Bank is using an Artificial Intelligence (AI)-powered connected to the IoT. More than 26 million users of a South Korean
Virtual Assistant called KAI to enhance the experience at digibank, financial service provider’s mobile payments app can simply talk
its mobile-only bank in India. KAI – which can understand language into a phone to authenticate transactions.
the way humans speak it, and is endowed with learning ability –
will help digibank to anticipate and reply to thousands of customer 2.4 Natural LanguageGeneration (NLG)
queries, and customers to fulfil banking transactions in real-time, at Natural language generation (NLG) is a set of technologies
any time, anywhere. generating natural language from a machine to converse and interact
intelligently with humans, and provide information, insights, and
advice in the same natural language. NLG is currently implemented
in customer-facing and business-user facing applications.

10 I External Document © 2017 EdgeVerve Systems Limited

Popular applications of NLG in banking: 3. Applied AI Solutions
NLG is mostly used in places where data from a lot of sources has
In this section, we will cover some of the most popular AI solutions
to be combined to generate insights in an understandable format.
which are designed using the six technology building blocks
Cognitive agents, such as Amelia, implement NLG to converse
discussed in the previous section. It’s important to note that these
intelligently with customers and provide them with insights to
are just few examples of variety of AI solutions which can be created
transform customer experience. Raw data can be tied up in a story
through the combination of multiple underlying AI technologies.
by using NLG – for example Quill software by Narrative Science is
used by financial institutions such as Credit Suisse. Quill is mostly
3.1 Smart Virtual Assistants (SVAs) and Bots
used to generate portfolio reviews.
Smart Virtual Assistants and Bots are software that can interact,
receive and deliver information, and act on human commands. SVAs
2.5 Visual Recognition
help perform day to day tasks, such as making an appointment,
Visual Recognition, as the name suggests, is the recognition
finding information or taking actions on a customer’s behalf. A
of images and their content. Employing deep learning, visual
common application is the chatbot, which can converse with
recognition (VR) technology is capable of finding faces, tagging
people. One can find chatbots in many places, and especially in
images, identifying the contents of a picture, and spotting similar
messaging apps. SVAs and bots offer banks a number of benefits,
images from a large set.
including 24/7 customer service, cost savings, and personalized &
targeted content delivery.
Popular applications of visual recognition in banking:
Very similar to speech recognition, visual recognition technologies
also enable frictionless customer experience. Westpac was the first Case example: RBS Luvo
Australian bank to allow customers to activate a new card through Royal Bank of Scotland has launched “Luvo”, a chatbot that assists
their smartphone cameras.
customers online. Luvo appears as a web chat tool that pops up
to ask customers if they need help. It frees bank staff from wasting
Visual recognition enables bank customers to pay bills by simply
time on addressing simple queries, so they may devote themselves
taking a picture on their smartphone camera, and also restaurant
to more complex issues. Luvo understands natural language, which
users to quickly pay up by scanning their bills.
means customers can actually write to it in their own words rather
UK’s Santander and South Africa’s ABSA Bank are among those to
than choosing from a menu. Eventually, the bank hopes to deploy
use ID scan to authenticate documents.
Luvo to improve personalization and detect problematic issues
before they surface.
Several banks, including Bank of America, Citibank, Wells Fargo,
TD Bank etc. offer remote check deposit via mobile, which is an
application of VR technology
3.2 Robotic Process Automation (RPA)
Robotic Process Automation (RPA) is the use of software and
2.5.1 Optical Character Recognition
machine learning to automate highly repeatable, high volume
Optical Character Recognition is a field of research spanning pattern tasks, thereby enabling the human workforce to focus on high-
recognition, artificial intelligence and visual perception. It involves
value tasks. The difference between IT automation and RPA is that
conversion of handwritten, typed or printed text into machine
the RPA system learns continuously by observing human actions
encoded text. Leveraging machine learning can help banks to
and uses machine learning to adjust its responses according to
improve accuracy and automate the process of converting physical
changing circumstances.
documents into fields in systems, leading to an intelligent OCR. OCR
with machine learning will go a long way in enhancing customer
Robotic Process Automation (RPA) can be used across a multitude
experience, automation and compliance for banks.
of processes such as customer onboarding, workflow acceleration,
data entry, validation and reconciliation.

External Document © 2017 EdgeVerve Systems Limited I 11

 Popular applications of RPA in banking: Popular use cases of Robotics in banking:
Robotic Process Automation (RPA) can be used across multitude Robots can be used to provide customer service in brick-and-mortar
of processes such as customer onboarding, workflow acceleration, branches for banks; this will allow human customer service agents to
data entry and validation, reconciliations, data enrichment – pretty focus on more complicated and priority tasks. For example, Pepper,
much every banking process which contains highly repeatable tasks. in Mizuho Financial Group Inc Bank, greets customers and is able to
recognize facial expressions. Mitsubishi UFJ Financial Group has also

Case example: ICICI Bank trialed a humanoid robot, “Nao”, to provide customer service.

India’s ICICI Bank has deployed software robotics in more than 500
3.4 Expert Systems
business processes, covering a million banking transactions every
day. It is the first bank in India to do so. Expert systems are similar to SVAs, except that they do not act on
behalf of users. Instead, expert systems collect and assimilate all the

The use of software robots has cut the time taken to respond to relevant content in a chosen domain area, and then provide users

customers by 60% and increased accuracy to 100%. The robots, with recommendations and answers. This is done both proactively

which are working in a variety of retail banking operations, as and reactively, based on the circumstances and the systems’

well as in treasury and human resources management among understanding.

others, capture and interpret information, recognize patterns and

run processes to perform functions like data entry and validation, Expert systems are used extensively in the financial services

automated formatting, text mining, reconciliation and exchange industry, especially for providing investment advice. Wealthfront

rate processing etc. and Betterment are examples of two fintechs that have deployed
software that work as expert systems.

3.3 Robots Popular use cases of expert systems in banking:

A robot is an electro-mechanical or bio-mechanical device that Expert advisors are used extensively in the financial services
can perform autonomous or pre-programmed tasks. Until a few industry, especially for providing investment advice. Wealthfront
years ago, robots were mainly used to perform tasks that were too and Betterment are examples of two fintechs that have deployed
dangerous or difficult for humans, such as cleaning radioactive software that work as expert advisors. These platforms take into
waste, or to automate repetitive tasks, such as automobile account a user’s demographic and savings goals, and then analyze
production. With AI technologies maturing, we are likely to see the current environment. Following this, they design an investment
robots entering the mainstream. portfolio that is tailored specifically for the user’s financial goals

Robots take on many different forms, ranging from humanoid, Clearly, this is not limited to wealth management space. One can
which mimic the human form and way of moving, to industrial, train advisory on other domains such as compliance, internal
whose appearance is dictated by the function they are to perform. policies, tax management etc.
Robots can be used to provide customer service in brick-and-
mortar bank branches. Mizuho Financial Group and Mitsubishi UFJ
Financial Group are using humanoid robots for this purpose.

12 I External Document © 2017 EdgeVerve Systems Limited

Key Areas of AI Application
From the above discussion it is clear that AI technologies can to completely reimagine banking processes and gain
be infused into several banking processes to great advantage. unprecedented efficacies in the front, middle and back office.
In fact, advancement in AI technologies offers an opportunity The figure below gives a snapshot of some of these processes.

i i

Settlements
Customer service Risk assessment
processing

Cash & liquidity

Sales and marketing Fraud detection/AML
management 

Wealth advisory &

Credit  scoring Reconciliation
financial assistance

Front office Mid office Back office

Following are three examples of how banks are deploying AI in the front, middle and back office.

Front Office: Sales and Marketing

The schematic given below captures how the process of sales and marketing can be re-imagined with AI technologies.

SALES AND MARKETING

Marketing campaign Selecting the most likely adopters using the data available

Delivering offers At the device and time most likely to get positive attention

Reviewing offer Log in through a frictionless experience (biometric authentication)

details in an app to review details of an offer

Immediate interaction Answers questions, qualifies lead based on the discussion, if needed,
with virtual advisors it can pass the context to human experts

Empowering Virtual advisor - send details on conversation and potential areas for
relationship manager readiness of the meeting, including answers to questions expected

External Document © 2017 EdgeVerve Systems Limited I 13

 Front Office Case Example: Swedbank Nina
At Swedbank Group, an intelligent virtual assistant named Nina The Bank’s service agents are also leveraging Nina to find information
is rendering conversational customer service to help customers for customers quickly. This has improved the experience for both
and agents help themselves. Customers type their queries via the parties, and within the first three months of deployment, taken first
website and Nina helps them find answers and also products and contact resolution to 78 percent. Customers have been positive
services best suited to their needs. about Nina, using the virtual assistant for 30,000 conversations per
month within the first 3 months of introduction. Nina is already
answering 80% of questions from Swedbank customers.

Mid Office: Fraud Management

The figure on the right
Real time transaction
gives a snapshot of how AI based engine offers capabilities to sift through
data analysis instead
transactions and recognize suspicious activity in real-time
AI can be leveraged to of analysis on past data
redefine the process
Deeper KYC and
of fraud management Algorithms to identify risk indicators, unusual behavior
understanding of
for transacting parties, auto linkages with bad actors
spanning Anti Money parties involved
Laundering (AML),
Know Your Client (KYC), Reduce false positives Machine learning capabilities help improve system capabilities
fraud detection and and negatives to continuously learn from false positives and negatives
regulatory compliance.

Proactive, instead of Augment risk analysts with recommended mitigation

reactive management strategies and actions

Proactive compliance Automatically generate compliance reports such as

reports Suspicious Activity Reports (SARs)

Middle Office Case Example: Finance

ZestFinance is a fast growing American fintech firm, which first is still unpaid. This strategy has paid off by bringing the default
leverages big data technology in credit underwriting. With the help rate down to 15%, half that of a payday loan.
of machine learning, ZAML™, its proprietary platform, crunches • Data assimilation - Rapidly discover, acquire, and onboard data
massive volumes of data traditionally ignored by credit underwriters sources at a massive scale.
to identify underbanked creditworthy prospects as well as mitigate • Modeling tools - Train, assemble, and productionalize machine
the risk in credit decisions. ZestFinance’s value proposition is to help learning models in one streamlined workflow
• Modeling expandability - Unpack the “black box” of machine
many borrowers, like millennials for instance, who had no access to
learning models to clearly communicate economic value and
credit earlier to now be able to avail loans.
support compliance
• Safely grow the lending business - Increase approval rates by
ZAML™ uses the same variables as FICO, along with non-traditional
leveraging machine learning to dredge through non-
“meta-variables” to understand borrower credentials. It collects as traditional credit data sources
many as 10,000 data points to work out the APR and the platform • Cut credit losses, without losing borrowers - Improve
takes just a few seconds to arrive at its decision. ZestFinance allows underwriting by accurately identifying genuine borrowers, and
neither rollovers, which inflate the APR, nor a second loan when the cutting out the high-risk ones

14 I External Document © 2017 EdgeVerve Systems Limited

Back Office: Cash and Liquidity Management
The schematic below shows how a bank can leverage AI technologies focus here is on the most popular liquidity management tools
to redefine back office processes like liquidity management. The including target balancing and notional pooling.

CASH AND LIQUIDITY MANAGEMENT

Designing a more With the help of advanced data analytics techniques, one
appropriate liquidity can project the account balances of the underlying accounts
management structure and perform analysis for designing the structure betters.

Comparative analysis across the combination of multiple structures

What if analysis
both notional and target balancing is possible through AI

Recommending
AI technologies can also help optimize structure dynamically basis
structure changes
changes in the FX rates for the underlying accounts
based on emerging

External Document © 2017 EdgeVerve Systems Limited I 15

 Preparing for AI
AI adoption in banking is a mixed story so far. Our research However, the decision should be based on the overall purpose of
shows that banks are the biggest investors in this technology, the AI strategy. For integrity and robustness, the platform is best.
yet financial services ranks third from the bottom on the Yet another view is that the option that best delivers end consumer
AI maturity index. expectations is the right one.

For the respondent organizations in our survey, the priority of Are you Ready for AI?
deployment is as follows: Big data automation heads the list, with Internal readiness
65 percent of organizations having already deployed or planning How successful a bank is in pursuing AI depends on its prior
to do so. About half the respondents are considering predictive experiences in integrating new technological innovations. AI-
analytics (54 percent) and machine learning (51 percent), 44 percent powered banking needs a group of technologies and a vast array
are investing in expert systems and 31 percent in neural networks. of digital components-from big data to cloud-based solutions. The
bank also needs to figure out if it is ready internally, but before that
Organizations are readying themselves to deploy these it needs to answer certain questions – questions, such as whether it
technologies by making investments in IT infrastructure (60 really needs AI, and if so, whether it should build or buy it. The bank
percent) and developing the required knowledge and skills (53 should then determine if there is a non-AI solution that presents a
percent). However, they are also seeking outside help from experts better business case as compared to an AI solution.
in areas such as planning (46 percent) and knowledge gathering
(40 percent). For developing artificial intelligence capability completely in-
house, a bank should consider the following:

65%
of respondents say that their organization has
deployed or plans to deploy big data automation Business
for collecting, processing and storing data In-house Tech
vision and      Timelines …
skills readiness
differentiation

54%
are looking at
51%
are looking at
What is your bank’s vision and strategy ?
Where does AI fit into that ?

predictive or machine learning Does your bank have a business case where AI can deliver
prescriptive analytics differentiated results?

60%
are investing in
53% What are the timelines for delivering the AI solution?

supporting Does your bank have the necessary in-house skills and investments
IT infrastructure required to not only develop and deploy AI solutions internally,
but also maintain and scale the technology in future?

 
Deployment Options What has been the rate of success for deploying and integrating
new technologies in your organization?
Banks can choose between platforms, applications and cloud
services for deploying AI. It is generally accepted that a combination
of all three is both practical and desirable. Does your organization have a framework to measure the project's ROI?

16 I External Document © 2017 EdgeVerve Systems Limited

For external partners
There are many technology providers with fairly mature offerings partner, a bank should evaluate the following criteria:
and services in Artificial Intelligence. When choosing a technology

Breadth   Depth of   Expertise  in

  MVP
  solutions …
of offering offering banking

How vast is the breadth and depth of the vendor's AI solution stack?

How well is the vendor’s AI offering and vision aligned with your organization's requirements ?

What is the expertise and experience of the vendor in deploying AI powered applications in banking?

What is the delivery model (on-premise / cloud) for AI capabilities?

Is it acceptable for your bank and for the regulations in your country to send
data to an external third-party vendor or host it on the cloud?

What is the vendor's track record in deploying AI based solutions among your peer group?

Key Barriers to Adoption

54% 54%
47% 49%
47%

37%

Senior Cultural Concerns about Lack of knowledge Lack of in-house Employee fear of
management acceptance handing over about where AI skills to implement change
resistance control can assist and manage

External Document © 2017 EdgeVerve Systems Limited I 17

 Lack of adequate infrastructure and skills is the biggest barrier to for processing data for reporting and compliance, a function that
AI adoption. Other key barriers stem from cultural issues including requires a large workforce today.
concerns about ceding control, (lack of ) acceptance and resistance
from the top. But it is also clear that AI will never replace human beings; on the
contrary, it can actually augment human abilities. For instance,
Ethical Implications of AI Implementation people engaged in tasks that will be performed by AI in future
An obvious question is what AI will do to existing jobs. No doubt can devote their time to more valuable pursuits such as creative

AI will take over many of the routine, repetitive jobs performed thinking, problem solving and innovation, which only they are
by humans today. It will also fundamentally transform the finance capable of. Our research shows that most organizations know this –
function performed by banks over the next few years. Going of the 75 percent of respondents planning to replace the workforce
forward, banks could use AI to build or redesign their operating with AI, the majority (80 percent) was going to retrain and redeploy
models and processes. Robotic Process Automation is a natural fit the impacted employees.

Consideration of AI ethics

53%
believe ethical concerns stop 8%
6% Yes, completely

AI from being as effective as it

No, only
can be 36%

No, not at all but we should

36%
20%
do

No, not at all and we don’t

think it is relevant

believe that their organization I don’t know

has completely considered the
ethical issues relating to the 30%
use of AI

18 I External Document © 2017 EdgeVerve Systems Limited

Readiness of AI for Banking
Even though there is lot of excitement when it comes to For instance, many chat engines aren’t great in conversations
implementation of AI for banking, there is still a lot of ground to where multiple commands are given. E.g. we tried giving simple
cover when it comes the readiness of AI technologies and solutions commands as following –
for business. • Can you set an alarm for 7 PM
• Can you change to 6 PM

AI Foundational Readiness for Potential for While few assistants followed the thread of conversations, many
               
Technologies business use banking others did not understand that the second command was associated
Machine Learning  Medium  High  with the first one.

Deep Learning  Low  High 

NLP  Medium  High 

NLG  Medium  Medium 

Visual Perception  Medium  High 

Applied AI Readiness for Potential for

Solutions
      business use
    banking
    

Virtual Assistants  Medium  High 

Robotic Process 
  High  High 
Automation

Robots  Low  Medium 

Expert Systems  Medium  High 

As per our analysis, following is the relative readiness of various

technologies and applied AI solutions for the banking industry.
The excitement surrounding AI needs to be tinged with an air of
caution as the technologies are supporting AI solutions are probably
not as ready for real-time use as the hype suggests. It’s critical to set
the realistic expectations for AI programs within the organization.
While the AI technologies hold great promise, many of them are in
the formative stages.

External Document © 2017 EdgeVerve Systems Limited I 19

 Conclusion
While the concept of Artificial Intelligence has been around for of initial hiccups and gestation times for expected results, banks
decades, it is only recently that the AI fantasy has started to turn into should waste no time in executing their AI plans. Because AI is
reality. Many of the technology pieces are already in place, albeit in evolving so quickly, it does not allow banks the luxury of waiting
varying stages of maturity. What’s left to do is stitch together these till it matures, and those who do, risk never being able to catch up
components to re-think banking processes and experiences. Many with the leaders. Quick movers have another advantage in that their
banks have made a start by incorporating several AI components AI systems will start learning earlier than others, and will therefore
into their processes and have experienced early results. evolve faster as well.

While the technology’s evolution is both rapid and impressive, With AI, the industry will go through a long voyage of reimagining
banks, and indeed, all enterprises, should ground their adoption banking, spanning several years, many milestones and at least a few
strategies and expectations in reality. Nevertheless, regardless challenges.

20 I External Document © 2017 EdgeVerve Systems Limited

About Banking Visionaries
Council (BVC)
Banking Visionaries Council has been constituted by Infosys Finacle council is to solve most pertinent problems with research and
to collaborate with senior business and technology leaders from collective thought leadership efforts. Currently, the council has
banking community to develop actionable point-of-views around twenty members strong board with representation from eleven
contemporary themes within the industry. The purpose of this countries across six continents.

Share key market development and trends observed in

respective geos with rest of the group

Collaborate to develop actionable point-of-view on how

banks can leverage emerging trends

Openly discuss learning from innovation initiatives taken

by respective banks

This point of view paper is an abridged version of the collaborative

research work done by the council. For more information on the
council, please reach out to [email protected].

External Document © 2017 EdgeVerve Systems Limited I 21

 References
1. Pharmafile: What could artificial intelligence mean for pharma?
http://www.pharmafile.com/news/502337/what-could-artificial-intelligence-mean-pharma

2. Celent Report on Fidor Bank

https://www.fidorbank.uk/documents/presse/celent_report.pdf

3. About ZestFinance: https://www.zestfinance.com/

4. Digibank by DBS
https://www.dbs.com/digibank/in/index.html

5. ICICI Bank introduces software robotics

https://www.icicibank.com/aboutus/article.page?identifier=news-icici-bank-introduces-
software-robotics-to-power-banking-operations-20160809103646464

6. About MeetCleo: https://meetcleo.com/

7. Amplifying human potential (AI maturity report): https://www.infosys.com/aimaturity/

8. Article ‘Swedbank sweet on virtual Nina’ on Banking Technology

http://www.bankingtech.com/480262/swedbank-sweet-on-virtual-nina/

9. Article ‘Meet Pepper robot at Emirates NBD’ on Khaleej Times: http://www.khaleejtimes.

com/20160921/meet-pepper-robot-at-emirates-nbd

10. Article ‘Encompass and ComplyAdvantage use AI for AML service’ on Finextra: https://www.
finextra.com/pressarticle/68815/encompass-and-complyadvantage-use-ai-for-aml-service/
transaction

11. Accenture report ‘Artificial Intelligence: Healthcare’s New Nervous System’: https://www.
accenture.com/us-en/insight-artificial-intelligence-healthcare

22 I External Document © 2017 EdgeVerve Systems Limited

External Document © 2017 EdgeVerve Systems Limited I 23
About Infosys Finacle
Finacle is the industry-leading universal banking solution from EdgeVerve Systems, a wholly owned product
subsidiary of Infosys. The solution helps financial institutions develop deeper connections with stakeholders,
power continuous innovation, and accelerate growth in the digital world. Today, Finacle is the choice of banks
across 94 countries, and serves over 848 million consumers – estimated to be nearly 16.5 percent of the world’s
adult banked population. Over a billion bank accounts are powered by Finacle globally.

Finacle solutions address core banking, online banking, mobile banking, payments, treasury, origination, liquidity
management, Islamic banking, wealth management, and analytics needs of financial institutions worldwide.
Assessment of the top 1000 banks in the world reveals that institutions powered by Finacle enjoy 50 % higher
returns on assets, 30 % higher returns on capital, and 8.1 % points lesser costs to income than others.

For more information, contact [email protected] www.finacle.com

©2017 EdgeVerve Systems Limited, a wholly owned subsidiary of Infosys, Bangalore, India. All Rights Reserved. This documentation is the sole property of EdgeVerve Systems Limited (“EdgeVerve”).
EdgeVerve believes the information in this document or page is accurate as of its publication date; such information is subject to change without notice. EdgeVerve acknowledges the proprietary rights
of other companies to the trademarks, product names and such other intellectual property rights mentioned in this document. This document is not for general distribution and is meant for use solely
by the person or entity that it has been specifically issued to and can be used for the sole purpose it is intended to be used for as communicated by EdgeVerve in writing. Except as expressly permitted
by EdgeVerve in writing, neither this documentation nor any part of it may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, printing,
photocopying, recording or otherwise, without the prior written permission of EdgeVerve and/ or any named intellectual property rights holders under this document.