Omega 36 (2008) 535 548

www.elsevier.com/locate/omega

Network design for reverse logistics

Samir K. Srivastava
Indian Institute of Management, Lucknow 226 013, India
Received 31 March 2006; accepted 24 November 2006
Available online 24 May 2007

Abstract
Collection and recycling of product returns is gaining interest in business and research worldwide. Growing green concerns
and advancement of green supply chain management (GrSCM) concepts and practices make it all the more relevant. Inputs from
literature and informal interviews with 84 stakeholders are used to develop a conceptual model for simultaneous locationallocation
of facilities for a cost effective and efcient reverse logistics (RL) network. We cover costs and operations across a wide domain
and our proposed RL network consists of collection centers and two types of rework facilities set up by original equipment
manufacturers (OEMs) or their consortia for a few categories of product returns under various strategic, operational and customer
service constraints in the Indian context.
In this paper, we provide an integrated holistic conceptual framework that combines descriptive modeling with optimization
techniques at the methodological level. We also provide detailed solutions for network conguration and design at the topological
level, by carrying out experimentation with our conceptual model. Our ndings provide useful insights to various stakeholders
and suggest avenues for further research.
2007 Elsevier Ltd. All rights reserved.
Keywords: Reverse logistics; Product returns; Locationallocation of facilities; Value recovery; Green supply chain management; India

1. Introduction
Green supply chain management (GrSCM) is gaining increasing interest among researchers and practitioners of operations and supply chain management.
Three drivers (economic, regulatory and consumer
pressure) drive GrSCM worldwide. It integrates sound
environmental management choices with the decisionmaking process for the conversion of resources into
usable products. GrSCM has its roots in environmental

This paper was processed by Guest Editors Angappa Gunasekaran and T.C. Edwin Cheng.
Tel.: +91 522 2736667; fax: +91 522 2734025.
E-mail addresses: [email protected],
[email protected] (S.K. Srivastava).

0305-0483/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.omega.2006.11.012

management orientation of supply chains. Producing

environmentally friendly products has become an
important marketing element that has stimulated a
number of companies to explore options for product
take-back and value recovery [1].
Managers have been giving increasing importance to
the environmental issues, their impact on operations and
potential synergies [2,3] since the early 1990s. Earlier
literature is generally restricted to the plant or rm level
focusing on green purchasing, industrial ecology, industrial ecosystems and corporate environment strategies [3]. Gradually, environmental management aroused
increased interest in the eld of supply chain management resulting in a growing literature on green supply
chains [46].
For the purpose of this paper, we consider GrSCM
as dened by Srivastava [4]. He denes GrSCM as

536

S.K. Srivastava / Omega 36 (2008) 535 548

Integrating environmental thinking into supply chain

management including product design, material sourcing and selection, manufacturing processes, delivery of
the nal product to the consumers as well as end-of-life
management of the product after its useful life. An interesting and signicant trend in GrSCM has been the
recognition of the strategic importance of reverse logistics (RL) as evident from classication and categorization of the existing GrSCM literature by Srivastava [4]
shown in Fig. 1.
RL shall become vital as service management activities and take-back for products such as automobiles,
refrigerators and other white goods, cellular handsets,
lead-acid batteries, televisions, personal computers
(PCs), etc. increase in future. A well-managed RL
network cannot only provide important cost savings
in procurement, recovery, disposal, inventory holding
and transportation but also help in customer retention.
Since RL operations and the supply chains they support are signicantly more complex than traditional
manufacturing supply chains, an organization that succeeds in meeting the challenges presents a formidable
advantage not easily replicable by its competitors [7].
Today, India is the fourth largest country in terms of
purchasing power parity (PPP) and constitutes one of
the fastest growing markets in the world [5]. However,
RL is yet to receive the desired attention and is generally carried out by the unorganized sector for some recyclable materials such as paper and aluminum. Some
companies in consumer durables and automobile sectors have introduced exchange offers to tap customers
who already own such products. The returned products
are sold either as it is or after refurbishment by third
parties.
Successful exchange offers have been marketing focused and no original equipment manufacturer (OEM)
has come up with repair and refurbishing or remanufacturing facilities for the returned products and their
sale. A summary of productmarket characteristics for
the wide category of products covered in our study is
presented in Table 1. The cumulative annual growth rate
(CAGR) shown is for the sales in the past decade and
the expected demand in the next decade.
We cover the literature on GrSCM, primarily focusing on RL. We do not consider literature and practices
related to green logistics as the issues are more of operational rather than strategic nature and may not be signicant in the RL network design per se. We also do not
focus in detail on literature on corporate environmental
behavior, green purchasing, industrial ecology and industrial ecosystems as it is generally either regulatorydriven or rm-specic. We rather focus more on RL

from resource-based viewpoint as establishment of efcient and effective RL and value recovery networks is
a pre-requisite for efcient and protable recycling and
remanufacturing. This has received less attention in the
GrSCM literature so far.
This paper is further organized as follows. In
Section 2, we describe briey our methodology in
light of our objective. This is followed by contextual literature review in Section 3. To address some
of the research issues and gaps related to designing
RL networks for product returns, we develop a conceptual model in Section 4. The development of the
corresponding mathematical model formulation for optimizing the decision-making is described in Section 5.
Data collection in the Indian context is described in
Section 6. Experimentation results for a few scenarios
for decision-making using our model are discussed in
Section 7. In Section 8, we conclude by describing the
contributions as well as the limitations of our work and
also suggest directions for further research.
2. Methodology
Our methodology consists of a theoretical part (literature review and conceptual model development) and
an applied part (maximizing prots for various scenarios in practical settings using a hierarchical optimization model and drawing useful managerial insights and
implications). A focused literature review seems to be a
valid approach, as it is a necessary step in structuring a
research eld and forms an integral part of any research
conducted. We focus mainly on RL literature deriving
from related areas like natural resource based view of
the rm, GrSCM, supply chain risk, etc. to maintain the
theoretical roots and linkages. We identify a few issues
and gaps as well as challenges and opportunities. Our
endeavor is to highlight the importance of RL and develop a more formal framework for analyzing the RL
and value recovery network design. To achieve this, we
develop a bi-level optimization model; use secondary
data for product returns for a period of 10 years; conduct
informal interviews with 84 stakeholders to gauge and
estimate various costs and other parameters; and nally
derive important strategic and operational implications
for various stakeholders.
3. Literature review
The resource-based-view of the rm draws primarily
from Hart [8] who proposes a theory of competitive
advantage based upon the rms relationship to the natural environment. He provides a conceptual framework

S.K. Srivastava / Omega 36 (2008) 535 548

537

Green Supply Chain Management

ECD

Disposal

Waste
Management

Location & Distribution

(Network Design)

Inspection/
Sorting

Pre-processing

Collecting

Production
Planning &
Scheduling

Remanufacturing

Inventory
Management

Reverse Logistics
& Network Design

Reuse

Recycling

Reducing

Green Manufacturing
& Remanufacturing

Pollution
Prevention

LCA

Product /
Material
Recovery

Green Operations

Green Design

Source
Reduction

Importance of GrSCM

Disassembly

Repair/Refurbish

LCA: Life Cycle Analysis

ECD: Environmentally Conscious Design

Disassembly Leveling

Disassembly Process Planning

Fig. 1. Classication and categorization of existing GrSCM literature. Source: Srivastava [4].
Table 1
ProductMarket characteristics of products covered in the study
Product
category

Product
variety

Product
ownership

Sales
CAGR (%)

Demand
CAGR (%)

Television sets

Large

High

6.5

4.5

Passenger cars

Medium

Low

11.7

11.8

Refrigerators

Large

Medium

7.8

9.7

Washing machines

Medium

Low

20.9

6.4

Cellular handsets

Large

Very low

69.5

18.1

Personal computers

Medium

Very low

35.0

17.4

Market characteristics

Large and still growing. Highly segmented.

Features being added. Stiff competition.
Many players.
Medium and growing. Clear segmentation.
Enhancements. Established and new players.
Large and growing. Segmented. New features. Stiff competition.
Medium and growing. Technological enhancements. Stiff competition.
Small and growing rapidly. Highly emergent
technology. Many entrants.
Small and growing rapidly. Highly segmented. Emergent technology. Stiff competition.

Source: Srivastava and Srivastava [32].

comprising three interconnected strategies: pollution

prevention, product stewardship and sustainable development along with their corresponding driving forces,
key resource requirements and their contributions to
sustained competitive advantage. Bloemhof-Ruwaard
et al. [2] elaborate on the possibilities of incorporating
green issues when analyzing industrial supply chains
and more generally of the value of using Operations
Research (OR) models and techniques in GrSCM
research.

Coming to RL, it is the collective noun for logistic

environments related to reuse of products and materials. Possible cost reductions, more rigid environmental
legislations and increasing environmental concerns of
consumers have led to increasing attention to RL in the
recent past [4]. The existence, effectiveness, and efciency of service management activities such as repair
services and value recovery depend heavily on effective RL operations [7]. Rogers and Tibben-Lembke [9]
dene RL as. . . the process of planning, implementing

538

S.K. Srivastava / Omega 36 (2008) 535 548

and controlling the efcient, cost-effective ow of raw

materials, in-process inventory, nished goods and related information from the point of consumption to the
point of origin for the purpose of recapturing or creating value, or for proper disposal. Therefore, designing
effective and efcient RL networks is a pre-requisite
for repair and remanufacturing and a key driver for providing the economic benets necessary to initiate and
sustain GrSCM initiatives on a large scale [4].
RL has been used in many applications like photocopiers [1], cellular telephones [10], rellable containers [11], etc. In all these cases, one of the major
concerns is to assess whether or not the recovery of
used products is economically more attractive than disposal. The added value could be attributed to improved
customer service leading to increased customer retention and sales. The added value could also be through
managing product returns in a more cost effective
manner [12] or due to a new business model [13].
Until recently, RL was not given a great deal of attention in organizations. Many of them are presently
in the process of discovering that improving their logistics processes can be a value-adding proposition
that can be used to gain a competitive advantage. In
fact, implementing RL programs to reduce, reuse, and
recycle wastes from distribution and other processes
produces tangible and intangible value and can lead to
better corporate image [14]. RL is one of the ve key
activities for establishing a reverse supply chain [15]
and comprises network design with aspects of product acquisition and remanufacturing [16]. Literature
identies collection, inspection/sorting, pre-processing
and logistics & distribution network design as four
important functional aspects in RL. In many cases, RL
networks are not set up independently from scratch
but are intertwined with existing logistics structures.
Fig. 2 shows the basic ow diagram of RL activities where the complexity of operations and the value
recovered increase from bottom left to top right.
RL can have both a positive and a negative effect
on a rms cash ow and needs further research attention [13,17]. Organizations and supply chains need
to understand the nancial impact of RL strategies
which can generate periodic negative cash ows that
are difcult to predict and account for. Toktay et al.
[18] nd that end-of-use returns have the potential
of generating monetary benets. Horvath et al. [17]
use a markov chain approach to model the expectations, risks, and potential shocks associated with cash
ows stemming from retail RL activities and actions
for avoiding liquidity problems stemming from these
activities.

A number of risks and uncertainties are associated

with end-of-use recovery. These are related to timing,
quality, quantity and variety of returns; estimation of operation and cost related parameters for RL networks; decisions about resolution for product returns and costs of
co-ordination along the reverse supply chain. These also
depend on consumer behavior and preferences. Various
incentives/disincentives to consumers based on product model and product quality can inuence the quality and quantity of product returns [19]. Therefore, RL
operations and the supply chains they support are signicantly more complex than traditional manufacturing
supply chains [7,9,20]. So, just as companies develop
efcient logistics processes for new goods, they must
do the same for returned goods, understanding that the
processes may be quite different from those dened for
forward distribution [21].
Fleischmann et al. [22] suggest that buy-back may
lead to higher returns leading to economies of scale.
Jayaraman et al. [23] have used resolution to customers
for this. Offering differentiated take-back prices to consumers based on product model and product quality
or charging a return fee is likely to reduce both the
number of returns as well as its variance [19]. Mont
et al. [13] present a new business model based on leasing prams where the productservice system includes
the organization of a RL system with different levels of
refurbishment and remanufacturing of prams, partially
by retailers. They focus on reducing costs for reconditioning, reduction of time and effort for the same and
nally on environmentally superior solutions.
Quantitative methods to support return handling decisions barely exist [5]. Most quantitative models in
RL literature deal with production planning and inventory control in remanufacturing, facility location,
resource allocation and ows [4]. The dimensions
used to characterize the RL environments are returns
volume, returns timing, returns quality (grade), product complexity, testing and evaluation complexity and
remanufacturing complexity. Fleischmann et al. [16]
provide a good review of quantitative RL models. Fleischmann et al. [24] give a theoretical investigation of
the synergy between the forward and the reverse chain.
Various modeling aspects relevant for designing RL
networks such as types of problem formulations, various decision variables and parameters used, data collection and generation techniques and various solution
techniques [10,22,23,25,26] can be seen in literature.
These resemble multi-level warehouse location problems and present deterministic integer programming
models to determine the location and capacities of RL
facilities.

S.K. Srivastava / Omega 36 (2008) 535 548

539

Fig. 2. Basic ow diagram of reverse logistics activities.

Uncertainty is not included explicitly in literature, but

usually addressed via scenario and parametric analysis.
Stochastic approaches are not much developed. Models do not incorporate multiple time periods although
the same has been suggested as possible extension and
future work [10,22,27]. Decisions may be long-term
such as those about facility location, layout, capacity
and design; or medium term such as those related to
integrating operations or deciding about which information and communication technologies (ICT) systems
are to support the return handling, which information is
to be kept and for how long? Short-term decisions are
related to inventory handling, vehicle routing, remanufacturing scheduling, etc.
Fleischmann et al. [22] consider the robustness issue
of RL networks. They do so at two levels: at methodological level by examining the appropriateness of
standard network design tools in a product recovery
context and at topological level by analyzing the impact
of product recovery on the physical network structure.
As a basis for analysis, Fleischmann et al. [24] refer
to a survey comparing nine case studies on recovery networks in different industries including carpet

recycling [26], reusable packages [25] and electronics

remanufacturing [10]. They conclude that the inuence
of product recovery is very much context dependent.
Fleischmann et al. [28] nd that the return ow information required by proposed models consists of the
aggregate return volume in each period over the planning horizon. Toktay et al. [18] state that there are few
documented business examples of forecasting specically for RL. If one could exactly know how much is
going to be returned and when, one would benet from
incorporating this perfect information a priori in the
management of returns. de Brito and van der Laan [29]
report on the impact of misinformation on forecasting
performance and performance with respect to inventory
costs by analyzing four forecasting methods as proposed by Kelle and Silver [11]. They nd that estimates
of the return rate can be quite erroneous in practice.
Again, whether RL network is open loop or closed loop
depends on a host of factors. In general, open-loop
systems are common in commodity-based industries
and are usually characterized by the existence of a seconds market [6]. Again, Mitra [30] says that quality
level of these will draw different prices in the seconds

540

S.K. Srivastava / Omega 36 (2008) 535 548

market The problem of choosing an appropriate reverse

channel is addressed by Savaskan et al. [27].
So, from the literature, it emerges that RL is not
a symmetric picture of forward distribution [16]. It is
much more reactive (supply driven). It is difcult to estimate supply-related parameters such as the unit operational costs directly from reported statistical data. Hu
et al. [31] use interviews with high-level decision makers of high-technology manufacturers for estimation of
supply-related parameters for their RL cost minimization model for the treatment of hazardous wastes. The
interviews included both open and closed-ended questions. The analytical results of the interview data are
then aggregated to identify the unit operational costs
and appear in the objective function and constraints of
their model.
Despite the success of a few organizations such as
Grameen Bank in Bangladesh and Unilever worldwide,
most organizations continue to mistakenly assume that
poor markets possess no value recovery opportunities
and have yet to understand the possibilities of serving the markets they are used to ignoring. Coordination
requirement of two markets (market for products made
from virgin materials and market for products made
from earlier-used products), supply uncertainty of returns and disposition task (further action to be taken on
returns) are identied as major challenges in RL literature [24].

4. Conceptual model
To address some of the issues related to designing RL
networks for product returns, we conceptualize a threeechelon (consumers returns collection centers
rework sites) multi-period RL and value recovery network model as shown in Fig. 3. We try to address a
number of strategic and operational questions related
to disposition, location, capacity and customer convenience using this conceptual model. The denitions used
in this conceptual model are described in Appendix A.
In our conceptual model, we assume that the consumers are the sources of product returns, similar
to some of the existing literature [15,23,25,32]. We
assume that the RL and value recovery network is
designed from scratch as a bring scheme i.e., the customers bring the used product to collection/buy-back
center (generally in a given time-window known a priori by telephone/Internet). For simplicity, we restrict
the choice to the existing distribution/retail outlets,
some or all of which may act as prospective collection
center [32].

Further, the differentiated complexity of operations

leads to two distinct rework sites: repair and refurbishing centers and remanufacturing centers. Repair
and refurbishing centers require lower capital investment, are more skill-based and repair/refurbish goods
in order to make them almost as good as new. Remanufacturing centers require very high capital investment,
are more technology-based and produce upgraded remanufactured goods. The rework facilities will come
up at some or all of collection centers. The disposition decisions are guided by prot motive and all the
returned goods are resold in primary or seconds market after necessary disposition. The rst disposition
(sell directly without rework) is carried out at collection centers themselves, as this involves no substantial
investment [32].
As customers do not prefer long distances, the collection centers need to be located within a certain
maximum distance from them. The customers receive
a resolution price on acceptance of the return. We assume no take-back obligation. Testing facilities and
product valuation charts are available at all collection centers. Manpower is skilled for inspection and
testing.
We use data from Srivastava and Srivastava [32] for
prospective locations of collection centers (a nite set
model comprising 117 locations) as well as quality
(grades), quantity and arrival-rates for three scenarios
of product returns for select product categories for next
10 years in the Indian context. We also use informal interviews with various stakeholders and secondary data
sources for estimating other input parameters such as
costs, distances, processing times, and conversion factors (including recovery rates) associated with the activities shown in Fig. 3. We assume unconstrained storage
capacity at each facility and negligible disposal costs.
The capabilities and capacities of rework sites differ.
Discrete capacity expansions are carried out at these
locations. No trans-shipment is allowed among rework
sites. A fraction of products sent to rework facilities
cannot be remanufactured/refurbished. Modules are recovered from these and sold in primary or seconds markets. Each product may have a few different type of
modules (some of them may be common).
As reworked goods supply is much less than their demand in the current Indian scenario, we assume that the
market for reworked products is unlimited. So, there is
no closing/decommissioning of facilities. We also assume that price of various products and modules in primary and seconds markets is known and suitable ICT
infrastructure is in place to support, analyze and coordinate RL activities.

S.K. Srivastava / Omega 36 (2008) 535 548

COLLECTION CENTER

541

REPAIR &
REFURBISHING
CENTER

CUSTOMER

PRIMARY
MARKET

S
E
C
O
N
D
S

REMANUFACTURING
CENTER

M
A
R
K
E
T

Fig. 3. The conceptual reverse logistics model.

5. Mathematical model formulation

We formulate a multi-product, multi-echelon, prot
maximizing RL and value recovery network model covering activities from collection to rst stage of remanufacturing. We ensure that it is a good representation of
the real-life situation and is at the same time tractable.
The objective function and various parameters and constraints have been clearly dened. The problem has been
treated similar to a multi-stage resource allocation problem. Various decisions such as the disposition decisions,
the sites to be opened, the capacity additions at any period of time as well as the number of products of a particular grade that are to be processed or sold during a
particular period of time are decided by the model.
Our combinatorial problem resembles a multicommodity network ow problem with a few sequentially dependent decisions for which no special algorithms are applicable apart from decomposition [33].
Besides, the large numbers of horizontal dimensions
(117 probable candidate collection center locations)
make our model very complex. For this, even for a
single time-period, there are as many as 1.66 1035
(2117 1) ways to open collection centers and 13 572
(117 P2 ) possible routings of returns. The problem complexity and size force us to device a simple strategy that
reduces computational complexity without any appreciable loss in the solution from practical perspectives.
We nd that there is no contradiction to our conceptual
model, if the opening decision for collection centers is
taken in a simple optimization model (based on certain
strategic and customer convenience constraints) and
thereafter impose these as rigid constraints for the main
model that determines the disposition decisions, loca-

Product Returns Data from Secondary Sources

Simple Optimization
[Investment Cost Optimization for locating
Collection Centers based on Strategic and
Customer Convenience related constraints]

Customer
Convenience
Constraints

Collection Center Locations

Parameters

Strategic
Constraints,
if any

Product Returns Collected

Main Optimization
[Profit Optimization for disposition,
location, capacity and flows based on
various input parameters and constraints]

Constraints

Detailed RL Network Design

(disposition, locations, capacities, flows etc.)

Fig. 4. Hierarchical optimization model.

tion and capacity addition decisions for rework sites

(remanufacturing centers and repair and refurbishing
centers) at different time periods as well as the ows
to them from the collection centers.
Our hierarchical optimization model is shown in
Fig. 4. The rst optimization decides the collection
center opening decision and all subsequent decisions
are carried out by the second optimization. First, we
use certain strategic and customer convenience constraints to determine the collection center locations.
This model, coded in GAMS (general algebraic modeling system) also calculates the product returns at these
locations. The main optimization model determines the
disposition decisions; location and capacity addition

542

S.K. Srivastava / Omega 36 (2008) 535 548

decisions for rework sites (remanufacturing centers

and repair and refurbishing centers) at different time
periods as well as the ows to them from collection
centers. The input parameters, variables and constraints
have been derived on the basis of informal interviews
with 84 stakeholders and the requirements of the mixed
integer linear program (MILP) formulation based on
the conceptual model.
The optimization model for collection center location
decision minimizes investment (xed and running costs
of facilities as well as transportation costs) subject to
following constraints:
(1) goods from a customer go to only open collection
center,
(2) goods from a customer go to only one collection
center,
(3) an open collection center remains open,
(4) distance between a customer and a collection center is within specied maximum limit to meet minimum customer service level requirements, and
(5) all parameters and variables are non-negative.
Number and grade of products originating at a particular
customer zone during a particular time-period, distance
between their origin and prospective collection centers,
transportation cost for products per unit distance (borne
by customer) and xed and running costs of collection
centers are some of the important input parameters used
in this model. Maximum distance limit is set between
origin of returns and their distance from collection centers for meeting customer convenience requirements.
The binary decision variables decide whether a collection center is opened during a particular time-period
and also ensure that a particular return goes to one open
collection center only.
The main optimization model maximizes prot
{realization from reselling (RL costs + resolution
price)}, subject to following constraints:
(1) the three disposition decisions are mutually exclusive and exhaustive,
(2) all goods with rst disposition decision are resold
at the collection centers,
(3) goods are sent to rework centers as per disposition
norms,
(4) goods from collection centers go to only open rework centers,
(5) capacity balance at rework centers,
(6) goods processed within capacity limits at open rework centers,
(7) inventory balance at rework centers, and
(8) all parameters and variables are non-negative.

Here, realization from reselling is the sum of revenues

from sale of returned products with or without rework as well as sale of recovered modules. RL costs
comprise xed and running costs of facilities, transportation, processing and inventory holding costs. The
resolution price is the money paid to the customers
for returns. Number and grade of products arriving at
a particular collection center during a particular timeperiod, distance between collection centers and rework
sites, transportation cost for products per unit distance,
xed and variable costs of rework centers, processing
costs, processing times, capacity addition sizes and
fraction of returns that can be successfully reworked on
are some of the important input parameters used in this
model. The binary decision variables decide whether
capacity is added at a candidate rework center during
a particular time-period. Other decision variables are
related to disposition decision and various ows.
6. Data collection
For application of the proposed model, its input
data may be classied into two groups: (1) returns
data which include the types of returned products, and
the time-varying amount associated with each type of
product, and (2) operations and cost related parameters
such as costs of facilities, capacity block sizes, processing times, fraction recovery rates, average number of
recoverable modules, storage costs, processing costs,
distances, transportation costs, procurement costs, sale
prices and so on.
We take the product returns data from Srivastava and
Srivastava [32]. These are shown in Table 2. For certain parameters such as collection costs and xed and
running costs of collection centers, informal interviews
with various stakeholders are the main determinants.
We decide the maximum distance a consumer wont
mind traveling for returning a particular product by experimentation with the collection center model keeping
in mind the consumer preferences from informal interviews. Fixed and variable costs of rework sites and the
capacity sizes are based mostly on web-searches.
For most of other parameters, such as product grades,
transportation costs, number of modules, sale price of
modules, processing costs, resolution price paid, recovery rates, etc., we refer to secondary sources and arrive
at some good estimates prior to approaching the stakeholders for deciding their values. Our interaction with
stakeholders revealed that they use the resolution price
as a reference for determining sale prices after disposition decision. Therefore, we use resolution price as the
reference price for sale prices at various facilities.

S.K. Srivastava / Omega 36 (2008) 535 548

543

Table 2
Relevant data for various product categories
Product category

Televisions
Passenger cars
Refrigerators
Washing M/C
Cellular handsets
Computers

Products in use
in base year (in 000)

76 282
5873
26 104
4952
2312
5796

CAGR of
forecasted sales (%)

4.5
11.8
9.7
6.4
18.1
17.4

Most likely estimated returns (in thousands)

1st Year

10th Year

CAGR (%)

1254
80
498
259
223
540

6789
833
3614
1690
2474
4760

20.6
29.7
23.2
24.6
30.7
27.4

Source: Srivastava and Srivastava [32].

Table 3
Few relevant operations and cost related parameters and their ranges
Items

Television

Refrigerator

Computer

Washing m/c

Car

Handset

Product grades
Transportation cost
Fixed cost (C)
Running cost (C)
Maximum distance (km)
No. of modules
Capacity block (Rm )
Capacity block (Rw )
Collection costs
Variable cost (Rm )
Variable cost (Rw )
Sale price (Modules)
Fixed cost (Rm )
Fixed cost (Rw )
Processing costs (Rm )
Processing costs (Rw )
Recovery rate (Rm )
Recovery rate (Rw )
Processing times (Rm )
Processing time (Rw )
Sale price ratio (C)
Sale price ratio (Rw )
Sale price ratio (Rm )
Resolution price paid

4
0.6 (0.24)
400 000
1 200 000
300
3
10 000 000
500 000 h
50
60 000k (6)
600k (1.2)
1502500
500 000 000
5 000 000
3202790
150260
0.90.99
0.90.98
2.913.24
1.211.48
0.850.94
1.102.48
1.152.54
22108020

4
0.4 (0.16)
300 000
600 000
200
10
10 000 000
2 000 000
80
50 000k (5)
2000k (1.0)
504000
750 000 000
30 000 000
8003790
5003600
0.80.9
0.820.97
1.713.54
1.012.28
0.900.94
1.122.82
1.152.94
201010 020

4
0.3 (0.15)
300 000
300 000
250
15
10 000 000
50 000 h
40
50 000k (5)
100k (2)
504000
750 000 000
750 000
18006680
10006300
0.800.97
0.800.99
1.001.24
1.122.14
0.950.98
1.072.05
1.102.10
612018 020

3
0.5 (0.2)
300 000
600 000
200
4
5 000 000 h
200 000 h
40
20 000k (4)
200k (1)
1502000
300 000 000
3 000 000
18003700
15003600
0.940.99
0.930.98
1.011.42
0.921.44
0.900.95
1.55.08
1.555.10
10006000

6
1.0 (1.0)
2 400 000
2 400 000
200
30
2 000 000 h
100 000 h
200
120 000k (60)
20 000k (20)
10014 000
1 000 000 000
10 000 000
13 79035 650
10 65030 650
0.940.99
0.961.00
5.7111.91
6.7110.41
0.920.99
1.081.46
1.101.50
40 00020 002

4
0.1 (0.1)
180 000
180 000
150
3
1 000 000
50 000 h
20
5000k (5)
50k (1)
501500
50 000 000
500 000
8004300
5003900
0.810.99
0.850.98
0.210.45
0.220.50
0.900.98
1.054.30
1.064.35
12501000

Some important operations and cost related parameters used for our experimentation and analysis are given
in Table 3.In the table, times and capacities are in hours
and costs are in Indian national rupees (INR) [One US
$ 45 INR]. The transportation costs in brackets are
the ones that have been considered for deciding collection center locations in the GAMS model. These are
in INR/km. The costs in brackets are the per-unit variable costs for remanufacturing (Rf ) and refurbishing
(Rw ). Ranges for other parameters in the table give
lower and upper values. Their distribution within the
range is generally non-linear.

7. Results and discussion

In this section, we discuss the results of experimentation and analysis across the select category of products under various scenarios to gain insights into both
the modeling and solution aspects of the RL and value
recovery network design. Simultaneously, we present a
few generalizations of results and their derived managerial implications.
First, we experiment with the collection center
model to nd out the impacts of various factors such as
maximum distance limits for customer convenience,

544

S.K. Srivastava / Omega 36 (2008) 535 548

Table 4
Experimentation with collection center model for personal computers
Estimated returns scenario

% Per unit transportation cost

Collection centers to be opened (t1.t10)

Maximum distance constraint (km)
100

150

200

300

500

None

Pessimistic

0
20
40
60
80
100

80
80
80
80
8081
8082

59
5960
5961
5961
5961
5962

43
4347
4347
4348
4348
4348

23
2327
2327
2327
2331
2335

10
1015
1119
1119
1226
1230

1
1012
1016
1125
1130
1130

Most likely

0
20
40
60
80
100

80
80
81
8083
8186
8285

59
6063
6067
6372
6375
6376

43
4648
4755
4964
4970
4972

23
2635
2747
2863
3068
3369

10
1531
1645
1960
2065
2067

1
1131
1645
1660
2065
2067

Optimistic

0
20
40
60
80
100

80
80
8085
8287
8291
8392

59
6067
6374
6376
6483
6884

43
4756
4968
4972
5180
5283

23
2742
3066
3570
3977
4581

10
1639
2064
3068
3476
3980

1
1539
2064
3068
3376
3980

transportation costs and xed and running costs, so as

to arrive at reasonable parameter values that keep the
model simple (in terms of size) and do not put any
signicant unrealistic constraints on the main model.
As the transportation costs for bringing the returns to
collection centers are borne by the customers, their
maximum distance from consumers should be within
reasonable convenient limits, inputs from informal interviews with various stakeholders about maximum
distance limits are taken into consideration while designing the experiments. The results for one such
experiment for PCs are shown in Table 4.
We see that for lower volumes and lower percentage per unit transportation costs, the maximum distance
constraint is pre-dominant whereas percentage per unit
transportation cost becomes more signicant for higher
distances. In absence of this constraint, the percentage
per unit transportation cost is the only criterion and the
number of collection centers opened increases progressively. It also increases with time as the number of returns increase with time. In absence of both, the model
suggests opening just one collection center. Optimization for most likely returns with maximum distance
limit in the range of 100300 km and percentage per
unit transportation cost in the range of 4060% seems
rational.

Next, we generate various scenarios with the optimization models using variations in processing times,
processing costs, return rates and other sensitive and
signicant input parameters for our select categories of
products. This helps us to draw some useful implications and managerial insights regarding characteristics
of RL networks for these. Here, we do not present detailed results for different scenarios and present only a
few signicant results and interpretations derived from
them in Table 5.
The disposition decisions are affected by a complex
interplay of various input and cost-related parameters.
Small changes in many input parameters and variables
change the decisions but do not affect the overall prots
appreciably. Thus, the prots for near-optimal policies
fall within a narrow range. Very few goods are sold
at a discount without rework. Refurbishing decision is
the dominant decision. The relatively high xed and
running capital costs of remanufacturing facilities act
against them vis--vis relatively cheaper and laborintensive repair and refurbishing facilities. The facility
sites keep on increasing with time period from year
to year and are somewhat proportional to returns for
a given maximum distance limit for customer convenience. Further, we see that the RL networks for refrigerators and televisions and the ones for cellular handsets

S.K. Srivastava / Omega 36 (2008) 535 548

545

Table 5
Few signicant decisions for the most likely scenarios
Decisions

TV

Car

Refrigerator

Washing M/c

Handset

Computer

Disposition 1 (resell)
Disposition 2 (remanufacture)
Disposition 3 (refurbish)
No. of collection centers
No. of repair and refurbishment centers

0.6%
0.0%
99.4%
4968
518

0.2%
0.0%
99.8%
1140
526

0.5%
0.0%
99.5%
2137
527

0.0%
0.0%
100.0%
3447
726

0.0%
0.0%
100.0%
5272
2658

0.8%
0.0%
99.2%
2644
1735

Table 6
Some signicant outputs for different products
Context

Television

Cars

Refrigerators

Washing machines

Handsets

Computers

Prot (million INR) [most likely]

Prot (million INR) [pessimistic]
Prot (million INR) [optimistic]
Break-even value [Inexion point]
Avg. capacity utilization (W ) [most likely]

15 651
3560
33 222
4%
24.3%

2173.1
184.2
4771.4
6%
96.9%

881.0
96.6
1763.9
4%
22.9%

220.2
18.9
451.1
17%
25.3%

732.7
115.3
1355.2
2%
65.1%

193.9
27.1
346.6
6%
81.3%

Table 7
Average prot per returned item for most likely scenarios
Product category

Prot
(million INR)

Returns
(000)

Average prot (INR)

Average resolution (INR)

% Avg. prot/avg. resolution (%)

TV (combined)
Passenger cars
Refrigerators
Washing M/C
Cellular handsets
Computers

15 651
21 731
8810
2202
7327
1939

42 282
4151
19 761
9685
81 470
21 086

370
5236
446
227
90
92

4766
11 7664
5947
1395
4108
11 513

8
4
7
16
2
1

and computers exhibit similar characteristics. These

are different and distinct for the other two product categories, namely passenger cars and washing machines.
We also determine the absolute prots (in million of
INR) for the three scenarios (most likely, pessimistic
and optimistic) and the break-even values of returns for
setting up various facilities for product returns for a
10-year period time-horizon. Average capacity utilization of the rework facilities for the most likely scenario
is also calculated. These are shown in Table 6. All categories show prots even for the pessimistic scenarios.
The break-even values for establishing value recovery
from product returns for these product categories, except for washing machines, is quite low. Further, we
observe that the average capacity utilization is higher
for cars and computers and relatively lower for refrigerators and washing machines.
Stakeholders are interested in the average prot per
item if appropriate RL network (as suggested by our
approach) are set up. Table 7 summarizes results for
the most likely scenarios. The average prot per item is

quite high for passenger cars and fair enough for televisions, refrigerators and washing machines. The last
column shows percentage of average prot per average resolution, a better indicator. These gures may be
used for relating risk with returns. For example, the
returns (as prots) for the risk-free cellular handsets
and computers are lower than that for the risky washing machines. Other product categories lie in-between.
Thus, setting up RL networks is protable for all the
select category of products; only the risks and returns
differ.
The above discussions show that designing effective
and efcient RL networks has important ramications
on OEMs and their consortia, local remanufacturers,
third party service providers as well as markets. It may
provide them many useful inputs and managerial insights (such as decisions regarding facility locations,
dispositions and various ows as well as break-even
values for investments and prots for a given scenario,
etc.). By determining the factors that most inuence a
rms RL undertakings, it can concentrate its limited

546

S.K. Srivastava / Omega 36 (2008) 535 548

resources in those areas; rest may be outsourced. The

insights drawn from various scenarios help concerned
stakeholders in identifying appropriate strategies. For
example, for lower-grade used products of relatively
high transportation effort, decentralization of facilities
may be more successful.
8. Conclusions
This paper highlights and re-inforces the importance
of RLan important area for practitioners which has
been under-explored by academics. We carry out RL
and value recovery network optimization and explore
the implications of setting up of remanufacturing and
repair and refurbishing centers by OEMs or their
consortia for certain categories of products in Indian
context. The major contribution of this research lies in
developing a formal framework for analyzing the network model and providing useful managerial insights.
Experimentation with variations in processing times,
processing costs and recovery rates provides insights
for various decisions. Insights into implementation issues can be drawn and if necessary, the model may
be modied (by using rigid constraints about various ows and facility opening decisions) for further
analysis.
Our model determines the disposition decision for
various grades of different products simultaneously with
locationallocation and capacity decisions for facilities
for a time horizon of 10 years. We develop our conceptual model from scratch borrowing from existing literature and industry practices and model the problem in an
operations research (OR) framework. We use a bi-level
program with decomposition to solve our prot maximization problem. Williams [33] suggests that a good
model can be developed by focusing on ease of understanding, ease of detecting errors and ease of computing the solution. We do the same. The basic problem is
formulated as a Mixed Integer Linear Program (MILP).
We use GAMS 21.2 with full version of the CPLEX 7.5
Solver. We use informal interviews with various stakeholders, similar to Hu et al. [31] for estimating various
operational parameters and justify the tools used and
explain methodology for estimation of parameters.
As in the existing literature, we nd a signicant impact of quantity, quality and timing of returns on decisions as well as overall prot [22,28]. There is space
and time dependency of various decisions, which is in
agreement with the basic rationale behind our approach.
The customer convenience distance constraints and the
per unit transportation costs also impact the network design to a considerable extent. Further, we nd that data

assumptions have direct implications on the construction of the underlying scenario.

Our work is a rst step towards RL network design for
product returns and value recovery in the Indian context.
It has been able to highlight various risks and opportunities. The ndings show that presently remanufacturing
is not a viable economic proposition. Underdeveloped
remanufacturing technologies and high capital investment in remanufacturing facilities are the bottlenecks.
Besides, product returns in most categories are still
lower than the scale of operations needed to reach the
critical mass to make remanufacturing economically
viable.
Kroon and Vrijens [25] consider wide domain of
GrSCM in a simple context of recycling of empty containers. Jayaraman et al. [10] consider remanufacturing
costs per unit and storage limitations of facilities in their
closed loop supply chain model. Jayaraman et al. [23]
consider remanufacturing facilities capacities in their
RL model. Our model tries to take the work further covering a wider domain of GrSCM. It considers variables
and parameters till the rst stage of remanufacturing
while optimizing facilities design. The model decides
even the disposition decision for a particular grade of
product at a particular location during a particular period of time. Thus, it integrates RL and value recovery. There are instances in literature [10,22] suggesting multi-period models as avenues for future work. Hu
et al. [31] have used multi-time period models in slightly
different contexts. We develop and experiment with a
full-edged multi-period model. It considers resolution
price as well as sale of recovered modules. This is a
step further to the consideration of revenue from sale of
reclaimed material and similar to component recovery
suggested by de Brito and Dekker [19]. We agree with
Toktay et al. [18] that return ow parameters should be
updated with time.
This paper has its own limitations. We deal with
supply side (returns) and returns disposition but do not
consider the co-ordination of the two markets. We still
follow a push system where the volumes of returns
drive the decisions and do not consider controlling product returns. The paper by Savaskan et al. [27] considers
many of these issues explicitly, assuming closed loop
supply structures as given. Thus, our work is complementary to their paper. We choose facilities from given
location options; there is no free choice. During formulation, we carry trade-off between scope and utility,
while in the solution we trade off between computation
time and optimality. The MILP model assumes that only
one block of capacity may be added during one timeinterval at a prospective site. Again, as our model has a

S.K. Srivastava / Omega 36 (2008) 535 548

very big measure of time-period (1 year), we do not consider lead-times in transportation, as they are negligible
with respect to a single time-period. We do not consider
integrated logistics (forward and RL) in a single model.
Further, we assume that all refurbished/remanufactured
products can be sold (unconstrained market).
The cost of rework vis--vis cost of production from
virgin materials too has not been considered. We carry
out estimations and optimization for product categories
and not brands or OEMs per se; however, inferences can
be drawn for them by simply using percentage of returns
equal to the market share of the brand or OEM. Finally,
our approach is highly exible and has scope for further
enhancement and enlargement. It can easily incorporate
multiple cost structures, market side considerations and
constraints related to resource conservation and regulatory perspective. It may be used for other potential products such as tires and batteries. Similarly, though our
study was done in the Indian context, it can be adapted
and applied to situations in other developing countries.
The paper opens a number of avenues for future research such as considering integrated logisticsunder
which circumstances should returns be handled, stored,
transported, processed jointly with forward ows and
when should they be treated separately; comparing cost
of remanufacturing with cost of production from virgin materials; potential attractiveness of postponement
strategies in value recovery; changes in a rms RL and
value recovery strategy for a particular product over the
course of the products life; and modeling for situation
when customer returns cannot be turned down.
The existing model may be congured for certain
operational as well as strategic decisions to develop a
customized decision support system. Penalties may be
imposed in the model instead of rigid customer convenience distance constraint. More detailed aspects of remanufacturing too may be included. Assumptions such
as unrestrained markets and innite storage capacity
may also be relaxed. Similarly, cost of remanufacturing
vis--vis cost of production from virgin materials may
be taken into consideration. Comparing/contrasting to
similar issues in countries like Brazil, China, Japan,
Mexico, Russia and USA, may throw light on how the
demographic makeup impacts RL network design. Considering the co-ordination of the two markets offers another very rich and fertile avenue for future work.
Finally, we agree with de Brito and Dekker [19] that
learning more about the practice of RL network design
for product returns can complement some of the latent
modeling difculties. So, survey-based research methods may be used to explain current practices, predominant and critical issues and managerial techniques used

547

to manage the RL networks. Again, empirical studies in

other countries as mentioned above may be very useful.
For example, research towards best practices may help
decision makers to come up with solutions for various
strategic, tactical and operational aspects. Simultaneous
desk and eld research will aid such decision-making.
Acknowledgments
This work would not have been possible without the
co-operation of 84 respondents who were informally
interviewed for this study and shared their knowledge,
experience and expertise We also express our thanks to
Srivastava and Srivastava [32] for permitting us to use
their data. Finally, we thank the three anonymous referees for providing many pertinent and useful comments
that helped in making the manuscript more focused,
precise and useful.
Appendix A. Denitions used in the conceptual
model
Collection center: A facility where customers bring
their products for resolution.
Disposition option: The decision about what is to be
done next to the accepted returns. There are three types
of disposition options: sell them without rework at the
collection center; sell after repair and refurbishing; sell
after remanufacturing.
Modules: Particular set of items (assembly/subassembly) that serve a particular purpose and may be
used in generally more than one product category.
Primary market: Market for sale of new and premium
goods.
Product category: The types of products along with
their different models.
Product grade: The classication of various returned
product categories based on their quality. It is a nominal
measure of the condition of a returned product.
Recovery rate: Fraction of products for which product recovery is possible. Modules are recovered from
them. The remaining fraction (1-recovery rate) is sold
in primary or seconds markets.
Remanufacturing center: A rework facility using
advanced technology for processing returned products.
Repair and refurbishing center: A rework facility
using appropriate level of technology and skills for repairing/refurbishing returned products.
Resolution price: Monetary value paid for a product
return.
Reverse logistics (RL): The process of planning,
implementing, and controlling the efcient, effective

548

S.K. Srivastava / Omega 36 (2008) 535 548

inbound ow, inspection and disposition of returned

products and related information for the purpose of
recovering value.
Rework center: A facility where returned products
are refurbished/remanufactured. There are two types of
such centers depending upon the disposition decision
and the level of technology and skills for processing returns into nal products: repair and refurbishing center
and remanufacturing center.
Seconds market: Market for sale of repaired and discounted goods.
References
[1] Thierry M, Van Wassenhove LN, van Nunen JAEE, Salomon
M. Strategic issues in product recovery management. California
Management Review 1995;37(2):11435.
[2] Bloemhof-Ruwaard JM, van Beek P, Hordijk L, van
Wassenhove LN. Interactions between operational research and
environmental management. European Journal of Operational
Research 1995;85(2):22943.
[3] Klassen RD, Vachon S. Collaboration and evaluation
in the supply chain: the impact on plant-level environmental
investment. Production and Operations Management
2003;12(3):33652.
[4] Srivastava SK. Green supply chain management: a state-ofthe-art literature review. International Journal of Management
Reviews 2007;9(1):5380.
[5] Srivastava SK. Logistics and supply chain practices in India.
Vision: The Journal of Business Perspective 2006;10(3):6979.
[6] Vachon S, Klassen RD, Johnson PF. Customers as green
suppliers. In: Sarkis J, editor. Greener manufacturing and
operations: from design to delivery and back. Shefeld: UK
Greenleaf Publishing Limited; 2001. [chapter 9].
[7] Amini MM, Retzlaff-Roberts D, Bienstock CC. Designing a
reverse logistics operation for short cycle time repair services.
International Journal of Production Economics 2005;96(3):
36780.
[8] Hart SL. A natural-resource-based view of the rm. Academy
of Management Review 1995;20(4):9861014.
[9] Rogers DS, Tibben-Lembke RS. Going backwards: reverse
logistics trends and practices. Pittsburgh: RLEC Press; 1999.
[10] Jayaraman V, Guide VDR, Srivastava R. A closed-loop logistics
model for remanufacturing. Journal of the Operational Research
Society 1999;50(5):497508.
[11] Kelle P, Silver EA. Forecasting the returns of reusable
containers. Journal of Operations Management 1989;8(1):
1735.
[12] Min H, Ko JH, Ko SC. A genetic algorithm approach to
developing the multi-echelon reverse logistics network for
product returns. Omega 2006;34(1):5669.
[13] Mont O, Dalhammar C, Jacobsson N. A new business model
for baby prams based on leasing and product remanufacturing.
Journal of Cleaner Production 2006;14(17):150918.
[14] Carter CR, Ellram LM. Reverse logistics: a review of the
literature and framework for future investigation. Journal of
Business Logistics 1998;19(1):85102.
[15] Guide VDR, van Wassenhove LN. The reverse supply chain.
Harvard Business Review 2002;18(2):256.

[16] Fleischmann M, van Wassenhove LN, van Nunen JAEE, van

der Laan E, Dekker R, Bloemhof-Ruwaard JM. Quantitative
models for reverse logistics: a review. European Journal of
Operational Research 1997;103(1):117.
[17] Horvath PA, Autry CW, Wilcox WE. Liquidity implications of
reverse logistics for retailers: a Markov chain approach. Journal
of Retailing 2005;81(2):191203.
[18] Toktay B, van der Laan EA, de Brito MP. Managing product
returns: the role of forecasting. In: Dekker R, Fleischmann M,
Inderfurth K, Van Wassenhove LN, editors. Reverse logistics:
quantitative models for closed-loop supply chains. Germany:
Springer; 2004. [chapter 3].
[19] de Brito MP, Dekker R. Modelling product returns in
inventory control-exploring the validity of general assumptions.
International Journal of Production Economics 2003;8182:
22541.
[20] Mitra S. Analysis of a two-echelon inventory system with
returns. Omega 2008; in press, doi:10.1016/j.omega.
2006.10.002.
[21] Stock J, Speh T, Shear H. Many happy (product) returns.
Harvard Business Review 2002;80(7):168.
[22] Fleischmann M, Beullens P, Bloemhof-Ruwaard JM, van
Wassenhove LN. The impact of product recovery on
logistics network design. Production & Operations Management
2001;10(2):15673.
[23] Jayaraman V, Patterson RA, Rolland E. The design of
reverse distribution networks: models and solution procedures.
European Journal of Operational Research 2003;150(1):
12849.
[24] Fleischmann M, Krikke HR, Dekker R, Flapper SDP. A
characterization of logistics networks for product recovery.
Omega 2000;28(6):65366.
[25] Kroon L, Vrijens G. Returnable containers: an example of
reverse logistics. International Journal of Physical Distribution
& Logistics Management 1995;25(2):5668.
[26] Louwers D, Kip BJ, Peters E, Souren F, Flapper SWP. A
facility location allocation model for reusing carpet materials.
Computers & Industrial Engineering 1999;36(4):85569.
[27] Savaskan RC, Bhattacharya S, Van Wassenhove LN. Closed
loop supply chain models with product remanufacturing.
Management Science 2004;50(2):23952.
[28] Fleischmann M, Bloemhof-Ruwaard JM, Beullens P, Dekker R.
Reverse logistics network design. In: Dekker R, Fleischmann M,
Inderfurth K, Van Wassenhove LN, editors. Reverse logistics:
quantitative models for closed-loop supply chains. Germany:
Springer; 2004. [chapter 4].
[29] de Brito MP, van der Laan EA. Inventory management with
product returns: the value of information. Research series
report in management, EI 2003-060-LIS, Erasmus University
Rotterdam, The Netherlands, 2003.
[30] Mitra S. Revenue management for remanufactured products.
Omega 2007;35(5):55362.
[31] Hu TL, Sheu JB, Huang KH. Reverse logistics cost
minimization model for the treatment of hazardous wastes.
Transportation Research, Part E: Logistics and Transportation
Review 2002;38(6):45773.
[32] Srivastava SK, Srivastava RK. Managing product returns for
reverse logistics. International Journal of Physical Distribution
and Logistics Management 2006;36(7):52446 [special issue
on 3PL, 4PL and reverse logistics].
[33] Williams HP. Model building in mathematical programming.
2nd ed., Chichester, UK: Wiley; 1985.