International Journal of Production Economics

Integrated foodbank network design: model and a case study

--Manuscript Draft--

Manuscript Number: PROECO-D-24-00033

Article Type: Research paper

Keywords: Foodbank; Integrated Network; Nutrition; Food Insecurity; Optimization

Corresponding Author: Jyotirmoy Dalal, Ph.D.

The University of Sheffield
SHEFFIELD, South Yorkshire UNITED KINGDOM

First Author: Ajinkya Tanksale, Ph.D.

Order of Authors: Ajinkya Tanksale, Ph.D.

Jyotirmoy Dalal, Ph.D.

Nistha Dubey, Ph.D.

Abstract: To address the UN’s zero hunger goal (SDG 2), scattered and isolated initiatives
by nonprofit organizations towards operating foodbanks are generally ineffective in
developing countries where the foodbank ecosystem is at a preliminary stage.
Establishing an integrated system comprising entities such as donors, foodbanks, food
recovery and redistribution agencies (FRRA), and beneficiaries can be quite complex
due to an underlying hierarchy, scale of operation, types of donors, and the severity of
food insecurity of the beneficiaries. In this work, we present a strategic mixed-integer
programming model to design an integrated foodbank network towards achieving an
efficient, effective, and equitable food distribution mechanism for food-insecure
beneficiaries while accounting for their age profile and nutritional requirements. We
ensure cost-efficiency by minimizing the total system cost, effectiveness by
discouraging food waste and unmet demand via charging penalties, and equity by
adopting five variants of an egalitarian approach. We conduct a case study with a mix
of real and realistically estimated data to design a foodbank network in Delhi (India)
and present detailed analyses with insights for the practitioners. Specifically, the effects
of budget and strategic-to-operational cost constraints on the solution are identified.
Among other observations, our analysis highlights when initiatives for collecting more
ready-to-eat foods might be taken to relieve the pressure on the integrated
system, and also helps in identifying the conditions when investment in capacity
building serves the beneficiaries’ interests better than direct spending.

Suggested Reviewers: Jitendra Kumar Jha, Ph.D.

Associate Professor, Indian Institute of Technology Kharagpur
jkjha@iem.iitkgp.ernet
Expertise in food supply chain in Indian context.

Irem Sengul Orgut, Ph.D.

Assistant Professor, The University of Alabama
isorgut@cba.ua.edu
Expertise in food bank operations management.

Halit Uster, Ph.D.

Professor, Southern Methodist University
uster@lyle.smu.edu
Expertise in mixed-integer programming and applications in real-world problem solving.

Manish Verma, Ph.D.

Professor, McMaster University DeGroote School of Business
mverma@mcmaster.ca
Expertise in developing mixed-integer programming models to apply on real-world
problems.

Vinay Ramani, Ph.D.

Associate Professor, Indian Institute of Technology Kanpur
vramani@iitk.ac.in
Expertise in economics and operations management interface.

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
 Priyanka Verma, Ph.D.
Associate Professor, Indian Institute of Management Mumbai
priyankaverma@iimmumbai.ac.in
Expertise in Optimization, Supply Chain Modelling, and Supply Chain Network design.

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Cover Letter

January 3, 2024

Editor-in-Chief,
International Journal of Production Economics

Dear Professor S. Minner,

Please find our submission of the original research article entitled, “Integrated Food Bank Network
Design: model and a case study” for publication consideration in the International Journal of Production
Economics Journal.

We are optimistic that the Editorial team will provide us with invaluable feedback on our submission, and
we look forward to the review process.

Thanking you,

Dr. Jyotirmoy Dalal

Lecturer in Decision Sciences
Sheffield University Management School
Conduit Road, Sheffield, S10 1FL, United Kingdom
Title page including author details

Integrated foodbank network design: model and a case study

Ajinkya Tanksale

atanksal@clarkson.edu

David D. Reh School of Business, Clarkson University, USA

Jyotirmoy Dalal*

j.dalal@sheffield.ac.uk

University of Sheffield, Management School, Conduit Road, S10 1FL, Sheffield, United Kingdom

Nistha Dubey

nistha.dubey.rs.mec18@itbhu.ac.in

Mechanical Engineering Department, Indian Institute of Technology BHU, Varanasi, India

Declarations of interest: none.

Acknowledgement: skipped for double-blind review.

CRediT author statement

Ajinkya Tanksale: Conceptualization, Methodology, Software, Formal analysis, Visualization,

Validation, Supervision. Jyotirmoy Dalal: Conceptualization, Investigation, Writing - Original
Draft, Formal analysis, Validation. Nistha Dubey: Conceptualization, Methodology, Investigation,
Software.

*Corresponding author. Contact : +44-7438881742 ; Email: j.dalal@sheffield.ac.uk

Highlights (for review)

Highlights

 Problem studied: strategic design of an integrated foodbank network.

 Objective: to achieve 3E’s i.e. efficiency, effectiveness and equity in the system.
 Approach: a mixed-integer programming model is formulated.
 Case study: a proposed foodbank network in Delhi, the capital of India, is presented.
 Managerial insights and policy implications are illustrated through experimentation.
Manuscript WITHOUT Author Identifiers

Integrated foodbank network design: model and a case study

Submitted to: International Journal of Production Economics

January 3, 2024

Abstract

To address the UN’s zero hunger goal (SDG 2), scattered and isolated initiatives by
nonprofit organizations towards operating foodbanks are generally ineffective in developing
countries where the foodbank ecosystem is at a preliminary stage. Establishing an integrated
system comprising entities such as donors, foodbanks, food recovery and redistribution agencies
(FRRA), and beneficiaries can be quite complex due to an underlying hierarchy, scale of
operation, types of donors, and the severity of food insecurity of the beneficiaries. In this
work, we present a strategic mixed-integer programming model to design an integrated foodbank
network towards achieving an efficient, effective, and equitable food distribution mechanism for
food-insecure beneficiaries while accounting for their age profile and nutritional requirements.
We ensure cost-efficiency by minimizing the total system cost, effectiveness by discouraging
food waste and unmet demand via charging penalties, and equity by adopting five variants of
an egalitarian approach. We conduct a case study with a mix of real and realistically estimated
data to design a foodbank network in Delhi (India) and present detailed analyses with insights for
the practitioners. Specifically, the effects of budget and strategic-to-operational cost constraints
on the solution are identified. Among other observations, our analysis highlights when initiatives
for collecting more ready-to-eat foods might be taken to relieve the pressure on the integrated
system, and also helps in identifying the conditions when investment in capacity building serves
the beneficiaries’ interests better than direct spending.
Keywords: Foodbank; Integrated Network; Nutrition; Food Insecurity; Optimization.

1 Introduction
To date, ‘zero hunger’ remains an extremely challenging goal to achieve by 2030 (United
Nations, 2023b) for both the developing and developed countries, exposed to extreme hunger

1
and malnutrition in a conflict-struck post-pandemic world. In 2021, 12.5% of U.S. households
with children and 10.2% of households in general remained food-insecure (U.S. Department of
Agriculture, 2023). In 2022, as per United Nations (2023b), while about 735 million people (9.2% of
the global population) were exposed to chronic hunger, an estimated 2.4 billion people experienced
moderate to severe food insecurity, which indicates an acute lack of access to required nourishment.
In the 2022 Global Hunger Index (GHI), India, one of the emerging and developing countries with
17.7% of global population in 2023, was ranked 107th out of the 121 countries with the score of 29.1
(level: serious) (Global Hunger Index, 2023). While the global food insecurity level is alarmingly
high, the Food Loss Index (FLI) [connected to UN SDG 12.3.1A] shows a staggering global food
loss of 13.8% in 2019 (i.e., around US $400 billion), making the food loss and waste reduction (UN
SDG target 12.3) (United Nations, 2023a) an associated challenge (Akkaş and Gaur, 2022).

To address the humanitarian crisis of hunger, the United Nations (2023b) urges for ‘coordinated
action and policy solutions’ that can be instrumental in making ‘a fundamental shift in trajectory’,
and help ‘to achieve the 2030 nutrition targets’. Foodbanks as nonprofit organizations can play an
important role in fighting hunger by becoming aggregators of surplus food having varying shelf lives
(ranging from fresh produce to cooked food) from different sources such as grocers, growers, and
supermarkets, and distribute the same to the needy population (Nair et al., 2017). Despite facing
challenges, the foodbank ecosystem has become mature over the decades in developed countries.
For example, Feeding America, the largest foodbank network in the US started in the late 1960s
(Feeding America, 2023). The Trussell Trust, with a network of over 1,200 foodbanks in the UK,
provide emergency food and support to the poor since 1997 (The Trussell Trust, 2023). In 1981,
Edmonton’s Food Bank started its journey as the first foodbank in Canada (Edmonton’s Food
Bank, 2023). However, despite facing food insecurity at a serious level in the GHI score, India
witnessed its first foodbank only in 2012 (The Hindu, 2012). Developments in the following years
have also been driven primarily by philanthropic initiatives of individuals and small groups to
address the local needs, lacking any broader plan or systems thinking (Dubey and Tanksale, 2022).

In a recent initiative, the Food Safety and Standards Authority of India [FSSAI] (2023) has
created the ‘Indian Food Sharing Alliance’ (IFSA), a platform to encourage ‘food donation, stop
food waste and food loss in the country’. Our research is motivated by the aim of IFSA to create
a ‘network by including food collection agencies, citizens, food businesses, corporates, civil society
organizations, volunteers, government and local bodies in a coordinated manner’ (Food Safety
and Standards Authority of India [FSSAI], 2023). In this paper, with a mathematical modeling
approach, we propose an integration of different isolated efforts that can help in creating a foodbank

2
ecosystem in developing countries where the concept is relatively new. Specifically, using a mixed-
integer programming (MIP) model (see §3) we present a case - with a mix of real and realistically
estimated parameters - of designing a strategic network in Delhi (India) by integrating two distinct
tiers of donors, beneficiaries, foodbanks, and food recovery and redistribution agencies (FRRA) (see
§4). While Tier-I includes large institutional donors and foodbanks with large capacities to provide
nutritional supplements to beneficiaries at certain schools, old-age homes, etc., Tier-II involves
small-scale donors, FRRAs with limited capacities to serve beneficiaries at slums/shelters exposed
to extreme food insecurity. Our proposed model can be used by the social planners as a centralized
decision making tool, which, in turn, can aid the authorities such as FSSAI in their policymaking
towards addressing the food insecurity and food waste issues effectively. Furthermore, the model
can aid in exploring the effects of imposing various practical conditions and restrictions on the
solution, thereby assist in gaining insights towards developing more impactful policies. Specifically,
our model can be instrumental in answering the following research questions.

1. Given the donation quantities from donors of different capabilities, beneficiaries’ demand, set
of existing and potential foodbanks and FRRAs with certain options of capacity increase,
which foodbanks and FRRAs to select for constructing an integrated network by connecting
them with appropriate donors and beneficiaries?
2. How the age profile and nutritional requirements of beneficiaries of different socio-economic
backgrounds can be included in strategic decision making?
3. How would the proposed integrated system respond to supply and demand uncertainties, and
a budget constraint?
4. What is the value of integrating two tiers that presently operate in silos?

Our base model designs the strategic network while addressing 3E’s, i.e., efficiency, effectiveness,
and equity. Cost-efficiency and effectiveness are addressed together by minimizing the system cost
and charging penalties for demand shortage (i.e., inability to address food insecurity), uncollected
donations as well as collected but unused donations (i.e., inability to address food waste) (see
§3.2). Specifically, the objective function minimizes the sum of fixed costs of adding foodbanks and
FRRAs, their capacity expansions (if needed), costs of food procurement and distribution among
different network entities, and three penalties for: (i) unmet demand, (ii) uncollected food at donors,
as well as (iii) collected but unused food at the foodbanks/FRRAs. We present a detailed analysis
of different egalitarian approaches of equity consideration in §5.7. Our extensive numerical analyses
(see §5.1 - §5.6) with various key observations should provide insights and help the policymakers in
appreciating the advantage of designing an integrated system to address food insecurity (see §5.4).

3
 This paper contributes to the literature in several ways. First, we present a new MIP model
to design an integrated network comprising donors, foodbanks, FRRAs, and beneficiaries with
different levels of nutritional needs. Our approach comprehensively includes both periodic (weekly
or monthly) bulk donations of large shelf-life foods by institutional donors and daily donations of
excess short self-life (cooked) food (more details on operational frequencies in §3.1) by small-scale
donors, thus, helps in addressing together the food insecurity and food waste issues. Second, our
study explicitly considers the conversion of different types of donated foods with different nutritional
values into predetermined configurations of food packets to fulfill the varied nutritional needs of
beneficiaries of different age groups with exposure to different socio-economic vulnerabilities. Third,
numerical analyses of our case study with the real and realistically estimated parameters, set at the
Delhi-National Capital Region (NCR) in India, elicit interesting insights that can aid a decision-
maker in the centralized, integrated, foodbank ecosystem.

The remainder of this paper is organized as follows. §2 presents a review of related literature,
followed by a discussion of our problem setting and the mathematical model in §3. Next, we present
a case study in §4 along with the base model’s solution. In §5, we discuss different numerical analyses
to illustrate the effects of changing various model parameters on the solution. Finally, we conclude
in §6 with a discussion on the usability of certain insights from this research to a decision-maker
and an indication of a few future research directions.

2 Related literature
Although the foodbank system has been well-established in Northern America since the 1960’s,
it gradually gained attention in the European and Asian countries over the past years. Due to
foodbank’s important role in addressing food insecurity, research interests have evolved in different
areas of foodbank operations and food aid distribution, for example, behavioral aspects of foodbank
users (Tarasuk and Eakin, 2005), nutritional analysis of the donation (Peters et al., 2021), prediction
of donation (Davis et al., 2016), donor-beneficiary matching (Dalal, 2022), optimizing the food
distribution effort by route planning (Govindan et al., 2014; Nair et al., 2017; Reihaneh and
Ghoniem, 2018), application of mobile pantries for distribution (Stauffer et al., 2022), food aid
modality selection (Rancourt et al., 2015; Sahinyazan et al., 2021).

Since we propose designing a strategic integrated foodbank network to enable donation collection,
packaging, and distribution by explicitly considering the varying needs of beneficiaries and the
nutritional values of donated and procured food, we primarily review the literature on foodbank
network design with the emphasis on cost-efficiency, effectiveness, and equity, while also keeping a

4
focus on the nutritional aspects of the distributed food. For reviewing diverse aspects of foodbank-
related research - which is out of our scope - we refer the reader to Dubey and Tanksale (2022) and
Mahmoudi et al. (2022).

Network design: On the foodbank network (re-)design problem, Martins et al. (2019) consider
a system comprising donors, foodbanks, and charitable agencies to serve the beneficiaries. Their
model’s decisions include opening foodbanks at potential locations, closing existing foodbanks,
storage and transportation capacity acquisition, food purchase, and transportation. The authors
adopt a triple bottom line approach with economic, environmental, and social objectives, and
analyze their tradeoffs using a lexicographic approach in a case study of foodbank network from
Portugal. Kaviyani-Charati et al. (2022) propose a multi-period, two-stage stochastic programming
model to design a similar network with donors, foodbanks, and demand zones experiencing demand
uncertainty, and apply the same on a case study on foodbanks in Tehran, Iran. They minimize food
waste and ensure the safety of surplus food through a cold chain while optimizing social performance
in addition to overall cost reduction. In a related study on Iran, Ghahremani-Nahr et al. (2023)
consider foodbank location, allocation, routing, and inventory decisions with the primary emphasis
on the nutritional value of a predefined food basket at foodbanks. They propose a multi-objective
robust optimization model and majorly contribute with novel solution approach. In a recent study,
Reusken et al. (2023) make capacity acquisition and network planning decisions for foodbanks,
while focusing on strategic utilization of available budgets to maximize the number of beneficiaries
that can be supported by the network. To the best of our knowledge, the above-mentioned four
literature focusing on foodbank network designing and planning are most relevant to our work.
Also, no foodbank network design problem has yet been studied in the Indian context - a gap that
we intend to fill.

Efficiency, effectiveness, and equity: Considering the importance of 3E’s, i.e., efficiency,
effectiveness, and equity (also aligned to UNSDG 10: reduced inequalities) in nonprofit operations,
several recent works emphasize these aspects. Martins et al. (2019) ensure equity in the distribution
of donated food among all charities by embedding it into social objectives while incorporating
efficiency and effectiveness in the economic and environmental objective functions. These 3E’s
also play a crucial role in food distribution (Orgut et al., 2016; Hasnain et al., 2021) and routing
problems for food rescue pickup/delivery operations (Nair et al., 2017; Rey et al., 2018). In foodbank
literature, increasing the number of beneficiaries served or the total quantity of food (re)distributed
is a common measure of effectiveness. Alkaabneh et al. (2021) undertake effectiveness as a measure
of the nutritional value of the allocated food to the agency. However, considering the prevalence of

5
the food waste problem, we address effectiveness by reducing waste by collecting surplus food at
the donors, penalizing any unused donation, and minimizing the unmet demand of beneficiaries by
charging different penalties. In literature, the 3E’s are commonly addressed by defining multiple
objectives and analyzing their tradeoffs using multi-objective optimization. In our work, efficiency
and effectiveness are captured in a single composite objective function. We conduct a separate
analysis on five equity variants by adopting the egalitarian approach of minimizing the maximum
value of some linear function of unmet demand experienced by beneficiaries (see details in §5.7).

Nutritional requirements: Food, being the central theme of the foodbank system’s fight
against hunger, must contain the nutrients that are essential to meet the recommended daily
intake requirement of the beneficiaries. However, many studies limit their scope to charitable
agencies and overlook the details of the end beneficiaries’ requirements by taking an aggregated
demand. Consequently, those studies cannot distinguish adults and children exposed to hunger
at different severities. In the context of foodbank network design, to the best of our knowledge,
only Ghahremani-Nahr et al. (2023) consider maximizing the nutritional values of the food packets.
Although some health and nutrition literature highlight the distinct needs of different age groups
concerning nutrients, this issue is inadequately addressed in the existing foodbank network design
and planning literature. Campbell et al. (2013) conduct a comprehensive survey on 137 US
foodbanks to document, understand, and analyze the culture, capacities, and practices related to
the nutrition of foodbanks. Bazerghi et al. (2016) investigate whether the beneficiaries’ nutritional
requirements are met by foodbanks. Both studies discover inadequacies in the foodbanks’
programs and recommend more attention towards aligning with the nutrition policies. This health
and nutrition literature helps us in devising the concept of preparing packets at foodbanks for
distribution from the available food following the predefined configurations to meet the diverse
nutritional needs of beneficiaries. Moreover, we assign priority weights to different beneficiary
groups to facilitate demand fulfillment in a resource-limited situation.

Geographical context: The literature contains several case studies on foodbank system
development in countries such as Portugal, Iran, Kenya, Netherlands, etc., however, no study has
yet been done on the network design of foodbanks in India, the most populous country experiencing
food insecurity at a serious level. We also fill this gap by conducting a realistic data-based case
study on Delhi, the capital city of India.

6
3 Problem setting and model formulation
In this section, we briefly present our proposed integrated foodbank system, followed by a MIP
model for designing the network. We also present an illustrative toy instance solution.

3.1 Problem description

Unlike several developed countries where a foodbank ecosystem is already established comprising
entities such as donors, foodbanks, partner agencies, and beneficiaries, in many developing
countries (e.g., India) this is an emerging domain with opportunities to strengthen the system by
systematically integrating the isolated and localized small social entrepreneurial initiatives. Based
on the scale, the scope of operation, capacity, and stakeholder characteristics, we broadly classify the
foodbank system into two tiers. Large foodbanks, equipped with large storage facilities, vehicles,
paid staff, and volunteers, can collect bulk donations (e.g., food grains, pulses, packaged foods
with long shelf life) from institutional donors such as retailers, growers, and grocers. The donation
stored at the foodbank’s warehouse is later converted to packets that are distributed to institutional
beneficiaries such as adult care homes, orphanages, and schools attended by economically backward
communities, to supplement their nutritional needs. We would call such a system comprising
“large donor → foodbank → institutional beneficiary” as Tier-I. Many smaller food recovery
and redistribution initiatives operate locally at towns and small cities to serve the marginalized
section of society such as malnourished slum-dwellers, homeless people staying at shelters, etc.
Such beneficiaries are in dire need of extensive food support, unlike supplements for the Tier-I
beneficiaries. These small agencies (often function informally) with a handful of volunteers get
engaged in primarily collecting various ready-to-eat and cooked food donations from sources with
excess food on regular basis (e.g., student hostels, office canteens, army barracks, restaurants, etc.),
and attempt to distribute their collection on the same day due to lack of storage/refrigeration
facilities and limited vehicular capacities (generally use small cars or motorbikes). We refer to these
entities as the Food Recovery and Redistribution Agencies (FRRA) by following the terminology
used in the IFSA (2023) website. Some of these FRRAs may also receive donations from a Tier-I
foodbank operating in their proximity. However, low capacities and extremely limited shelf life of
the donated food restrict them to work locally, where distribution can be completed within a few
hours. We would call this system comprising “small donor → FRRA → individual beneficiary”
as the Tier-II. These two tiers, representing two very different systems, often operate in silos in
practice, lacking any overarching framework.

Our research proposes a strategic integration of the aforementioned two tiers to make a holistic

7
 Food Bank

𝐵1

𝐽1
𝐼1
𝐵̅1

FRRA

𝐵2
𝐽2
𝐼2
𝐵̅2

Donor Set 𝑰 Food Bank/FRRA Set 𝑩 Beneficiary Set 𝑱

Figure 1: (Colour online) Schematic diagram of sets (notations in Tables 1 - 3)

foodbank ecosystem using an MIP model. Given the sets of donors, beneficiaries, existing as
well as potential (new) foodbanks and FRRAs (refer to Figure 1), our model makes strategic
decisions on selecting potential foodbanks/FRRAs for inclusion in an integrated network, capacity
expansion decisions for existing or potential foodbanks, and operational decisions of connecting
donors and beneficiaries to the operating foodbanks/FRRAs of the same tier, food procurement
(if needed), and flows among these network entities. Since nonprofit systems generally experience
tight budgets, our model finds the solution cost-effectively by minimizing the sum of strategic
network building cost, various tactical and operational costs associated with flows, and penalties to
discourage both food shortage and waste. This way, our model addresses both economic efficiency
(cost minimization) and operational effectiveness (reducing shortage and waste). Furthermore, in
§5.7, we extensively discuss equity, thus, our approach addresses the important 3E’s in nonprofit
operations management.

Figure 2 presents a schematic diagram of the proposed system with the model parameters and
decision variables. For notations, we refer the reader to Tables 1 - 3. The beneficiary nodes have
different beneficiary types π ∈ Π, e.g., child and adult. While Dπj represents the demand of
type π beneficiary at the node j, unmet demand at that node in terms of the number of type π
beneficiaries, is represented by λπj . If needed for fulfilling the beneficiaries’ demand, the model can
open potential foodbanks/FRRAs b ∈ B̄1 ∪ B̄2 , i.e., at the respective tier, incurring fixed cost Cb
(associated decision variable xb ). A donor i1 having total available donation quantity Qf i1 , sends

8
 Tier-I: Food Bank 𝒃𝟏
Tier-I Beneficiary

Storage Transportation Staff 𝐷𝜋𝑗1 ; 𝜆𝜋𝑗1

Parameters
𝑞𝑓𝑖1 𝑏1 Capacity 𝑆𝑏01 ; 𝑆𝑙 𝑇𝑏01 ; 𝑇𝑘 𝑉𝑏01 ; 𝜌𝑏01 ; 𝜙; 𝜙̂ 𝜗𝜋,𝑏1, 𝑗1 ,𝑝
𝑗1
𝑖1
Cost 𝐶𝑙𝑏1 𝐶𝑘 𝐶𝑀

𝑄𝑓𝑖1 , 𝜃𝑓𝑖1 Decision variables 𝑦𝑙𝑏1 ; 𝑢𝑏1 𝑡𝑘𝑏1 ; 𝑣𝑏1 𝑝𝑏1 ; 𝑤𝑏1

Donor Type I
If not already open, Fixed Cost for opening = 𝐶𝑏1

𝛼𝑓 , Γ𝑓
𝑞𝑓𝑏1 𝑏2
Food Purchase

Tier-II: FRRA 𝒃𝟐
Tier-II Beneficiary
Transportation Staff 𝐷𝜋𝑗2 ; 𝜆𝜋𝑗2
Parameters

Capacity
𝑞𝑓𝑖2 𝑏2 𝑇𝑏02 ; 𝑇𝑘 𝑉𝑏02 ; 𝜌𝑏02 ; 𝜙; 𝜙̂ 𝜗𝜋,𝑏2, 𝑗2, 𝑝
𝑖2 𝑗2
Cost 𝐶𝑘
𝐶𝑀
𝑄𝑓𝑖2 , 𝜃𝑓𝑖2
Decision variables 𝑡𝑘𝑏2 ; 𝑣𝑏2 𝑝𝑏2 ; 𝑤𝑏2
Donor Type II
If already not open, Fixed Cost for opening = 𝐶𝑏2

Figure 2: Schematic diagram of proposed system: parameter and decision variables in Tables 1 - 3.

qf i1 b1 to an operative foodbank b1 . If θf i1 quantity remains uncollected at i1 , as mentioned before,

to discourage it, a penalty is charged. If donation received at a Tier-I foodbank b1 is inadequate, the
model can suggest procuring µf b1 quantity of permitted food type f (indicator αf = 1) at the unit
cost Γf . Note that a Tier-II FRRA b2 does not purchase food, but, can receive qf b1 b2 quantity of
food type f transferred from a Tier-I foodbank b1 . Note here, since Tier-I mostly handles uncooked
food donations and purchases, we consider a Tier-I foodbank b can have an initial storage capacity
Sb0 , with possible increments Sl at levels l ∈ L at the cost of Clb (associated decision variable ylb ).
However, Tier-II FRRAs being small and primarily handling cooked/ready-to-eat food with low
shelf life, do not have storage capacity. Foodbanks and FRRAs have initial transport capacities
Tb0 with possible increase Tk at the cost of Ck (associated decision variable tkb ). Similarly, the
workforce capacity of a foodbank/FRRA b can be increased from its initial level (volunteers Vb0
and paid staff ρ0b ), considering the staff hiring cost C M (associated decision variable pb ).

Figure 3 further elaborates on the food flow through the proposed system. Upon donation
receipt, purchase (if needed), and transfer, the net quantity qbf b is allocated as ωf πb among different
beneficiary types (π ∈ Π). Next, discounting any unused food quantity γf πb at the foodbank/FRRA

9
 𝑏1

𝜗𝜋1 𝑏1 𝑗1 𝑝
𝜋1
𝑞𝑓𝑖1 𝑏1 Foodbank 𝑗1
𝑖1
Child: 𝜋1 𝜋2
𝑞̂𝑓𝑏1
𝑄𝑓𝑖1 , 𝜃𝑓𝑖1

𝜋1
Adult: 𝜋2
𝑗2
𝜋2

𝛼𝑓 , Γ𝑓
𝑞𝑓𝑏1 𝑏2
Food Purchase 𝑏2

𝜗𝜋1 𝑏2 𝑗3 𝑝
𝜋1
𝑗3
FRRA
Child: 𝜋1 𝜋2
𝑞𝑓𝑖2 𝑏2
𝑖2 𝑞̂𝑓𝑏2

𝑄𝑓𝑖2 , 𝜃𝑓𝑖2 𝜋1 𝑗4
Adult: 𝜋2
𝜋2

Figure 3: (Colour online) Schematic diagram of flows through proposed system (notations in Tables
1 - 3).

b (discouraged by penalty), the net available food allocated for beneficiary π is converted to τπbp
packets (shown in the rectangles within Figure 3), prepared as per predefined packet composition
(represented by βf πmp ) to meet the beneficiaries’ nutritional needs (further elaborated in §4). From
τπbp packets, our MIP model further determines the appropriate allocation of ϑπbjp packets of type
p for beneficiary type π living at node j (shown along the arrows from foodbanks to beneficiaries).

Finally, we mention the differences in operational frequencies between the two tiers (also in Table
3). While Tier-I foodbanks handle bulk inflows on a weekly basis, Tier-II FRRAs receive smaller
IN ). Moreover, all the foodbank → FRRA transfers also occur on
donations daily (parameter Tm
daily basis (parameter T1T R ). Therefore, donation inflow-handling frequencies (parameter Tm
H)

are also on a weekly and daily basis for the Tier-I foodbanks and Tier-II FRRAs, respectively.
OU T ), the
However, as outflows from foodbanks/FRRAs of both tiers occur daily (parameter Tm
P is also on a daily basis. Thus, the centralized decision maker of the
packaging frequency Tm
proposed foodbank network can run our mathematical model on every weekend to plan ahead for

10
the upcoming week. Next, we present our mathematical model with all notations.

3.2 Notations and mathematical model

We present the notations for sets and indices, decision variables, and parameters for the proposed
mathematical model in Tables 1, 2, and 3, respectively.

Entity Notation Description Index

Network M Set of Tiers m
I1 Set of donors who deliver food to Tier-I foodbanks
Donor I2 Set of donors who deliver food to Tier-II FRRAs i
I Set of all donors
B1 Set of existing Tier-I foodbanks
B2 Set of existing Tier-II FRRAs
Foodbank B̄1 Set of candidate Tier-I foodbanks b
B̄2 Set of candidate Tier-II FRRAs

B Set of all foodbanks and FRRAs B1 ∪ B2 ∪ B̄1 ∪ B̄2
L Storage capacity levels (Only at Tier-I foodbanks) l
Capacity
K Transport capacity alternatives (type of vehicles) k
Π Set of beneficiary type π
J1 Set of Tier-I beneficiary nodes
Beneficiary J2 Set of Tier-II beneficiary nodes
j
J Set of all beneficiary nodes
Food Items F Set of food items f
P Set of food packet types p

Table 1: List of sets and indices for the mathematical model.

xb = 1 if foodbank/FRRA b ∈ B̄1 ∪ B̄2 is open, 0 otherwise

ylb = 1 if storage capacity level l ∈ L is installed at Tier-I foodbank b ∈ B1 ∪ B̄1 , 0 otherwise
tkb number of vehicles of type k ∈ K hired at foodbank/FRRA b ∈ B
pb number of personnel hired at foodbank/FRRA b ∈ B
ub effective storage capacity at Tier-I foodbank b ∈ B1 ∪ B̄1
vb effective transportation capacity at foodbank/FRRA b ∈ B
wb effective staff capacity at foodbank/FRRA b ∈ B
qf ib quantity of food f ∈ F flow from donor i to foodbank/FRRA b (across-tier flow not allowed, i.e.,
qf ib := 0 if i ∈ I1 , b ∈ B2 ∪ B̄2 and qf ib := 0 if i ∈ I2 , b ∈ B1 ∪ B̄1 )
µf b quantity of food f ∈ F procured at Tier-I foodbank b ∈ B1 ∪ B̄1
qf bb′ quantity of food f ∈ F transferred from Tier-I foodbank b ∈ B1 ∪ B̄1 to
Tier-II FRRA b′ ∈ B2 ∪ B̄2 (i.e., qf bb′ := 0 if b ∈ B2 ∪ B̄2 , b′ ∈ B)
q̂f b quantity of food f ∈ F available for distribution at foodbank/FRRA b ∈ B
θf i uncollected food type f ∈ F at donor location i ∈ I
ωf πb quantity of food f ∈ F for beneficiary type π ∈ Π at foodbank/FRRA b ∈ B
τπbp food packets of type p ∈ P for beneficiary type π ∈ Π made at foodbank/FRRA b ∈ B
ϑπbjp packet type p ∈ P sent from foodbank/FRRA b ∈ B for beneficiary type π ∈ Π staying at node j ∈ J
γf πb food f ∈ F for beneficiary type π ∈ Π remained unused at foodbank/FRRA b ∈ B
(i.e., collected from donor but not distributed)
λπj Unmet demand of beneficiary type π ∈ Π at beneficiary location j ∈ J

Table 2: List of decision variables for the mathematical model.

11
 Capacity
Sb0 Initial storage capacity at Tier-I foodbank b ∈ B1 ∪ B̄1
Sl Storage capacity addition at level l ∈ L (only for a Tier-I foodbank)
Tb0 Initial transport capacity at foodbank/FRRA b ∈ B
Tk Transport capacity addition by including a vehicle of type k ∈ K
Vb0 Number of volunteers at foodbank/FRRA b ∈ B
ρ0b Number of paid staff at foodbank/FRRA b ∈ B
U t , U p upper bounds on transport and staff capacity addition, respectively
φ Donation quantity handled by one person per day
φb Number of food packets handled by one person per week
Demand and donation
Dπj Demand at beneficiary location j ∈ J for beneficiary type π ∈ Π
Qf i Quantity of food f ∈ F available for donation at donor i ∈ I
Nutrition
βf πmp Quantity of food f ∈ F in a packet of type p ∈ P for beneficiary type π ∈ Π
belonging to tier m ∈ M
αf = 1, if food f ∈ F is allowed to be procured by foodbank, else 0
Costs
Cb Cost of opening a new foodbank/FRRA
Clb Cost of installing storage capacity of level l ∈ L at Tier-I foodbank b ∈ B1 ∪ B̄1
Ck Cost of adding a vehicle of type k ∈ K
CM Cost of hiring a staff at foodbank/FRRA
Γf Unit procurement cost for food item f ∈ F
α Unit transportation cost
dij Distance between node pair (i, j)
Cij = αdij , transportation cost between appropriate node pairs (i, j) in network
(1)
Pm Penalty for uncollected donation from donor of Tier m ∈ M
(2)
Pm Penalty for collected but unused donation at a foodbank/FRRA of Tier m ∈ M
(3)
Pπm Penalty for unmet demand of beneficiary type π ∈ Π belonging to Tier m ∈ M
Frequency of Operations
IN
Tm Inflow frequency at foodbank/FRRA of Tier m ∈ M; T1IN = 1 (weekly), T2IN = 7 (daily)
TR
T1 Transfer frequency from foodbank of Tier m = 1; T1T R = 7 (daily)
OU T OU T
Tm Foodbank/FRRA → beneficiary distribution frequency; Tm = 7 (daily), m ∈ M
H
Tm H
Inflow handling frequency at foodbank/FRRA; T1 = 1 (weekly), T2H = 7 (daily)
P P
Tm Packet sending frequency from foodbank/FRRA of Tier m; Tm = 7 (daily), m ∈ M

Table 3: List of parameters for the mathematical model.

Formulation

X X X X X 
[P] Min Z = Cb x b + Clb ylb + Ck tkb
b∈B¯1 ∪B¯2 l∈L b∈B1 ∪B¯1 k∈K b∈B
X X X X
+ C M (pb + ρ0b xb ) + C M (pb + ρ0b ) + αf Γf µf b
b∈B¯1 ∪B¯2 b∈B1 ∪B2 f ∈F b∈B1 ∪B̄1
X X X X X X XX 
+ Cib qf ib + Cbb′ qf bb′ + Cbj ϑπbjp
f ∈F {(i,b):i∈Im ; f ∈F b∈B1 ∪B̄1 b′ ∈B2 ∪B̄2 {(b,j):b∈Bm′ ∪B̄m′ ; π∈Π p∈P
b∈Bm′ ∪B̄m′ j∈Jm ;m,m′ ∈M,m̸=m′ }
m,m′ ∈M,m̸=m′ }

12
 X X X  X X X X  X X X 
(1) (2) (3)
+ Pm θf i + Pm γf πb + Pπm λπj (1)
m∈M f ∈F i∈Im m∈M f ∈F π∈Π b∈Bm ∪B̄m π∈Π m∈M j∈Jm

subject to
X
ylb ≤ 1 ∀b ∈ B1 (2)
l∈L
X
ylb ≤ xb ∀b ∈ B̄1 (3)
l∈L
X
ub = Sb0 + Sl ylb ∀b ∈ B1 (4)
l∈L
X
ub = Sb0 xb + Sl ylb ∀b ∈ B̄1 (5)
l∈L
X
vb = Tb0 + Tk tkb b ∈ B1 ∪ B2 (6)
k∈K
X
vb = Tb0 xb + Tk tkb b ∈ B̄1 ∪ B̄2 (7)
k∈K

tkb ≤ U t ∀k ∈ K, b ∈ B1 ∪ B2 (8)

tkb ≤ U t xb ∀k ∈ K, b ∈ B̄1 ∪ B̄2 (9)

wb = (Vb0 + ρ0b ) + pb ∀b ∈ B1 ∪ B2 (10)

wb = (Vb0 + ρ0b )xb + pb ∀b ∈ B̄1 ∪ B̄2 (11)

pb ≤ U p ∀b ∈ B1 ∪ B2 (12)

pb ≤ U p xb ∀b ∈ B̄1 ∪ B̄2 (13)

X
qf ib + θf i = Qf i ∀f ∈ F, i ∈ Im , m ∈ M (14)
b∈Bm ∪B̄m
XX
IN
qf ib ≤ Tm vb ∀b ∈ Bm ∪ B̄m , m ∈ M (15)
f ∈F i∈Im
X X
qf bb′ ≤ T1T R vb ∀b ∈ B1 ∪ B̄1 (16)
f ∈F b′ ∈B2 ∪B̄2
X
OU T
q̂f b ≤ Tm vb ∀b ∈ Bm ∪ B̄m , m ∈ M (17)
f ∈F
XX X
qf ib + αf µf b ≤ T1H (φwb ) ∀b ∈ B1 ∪ B̄1 (18)
f ∈F i∈I1 f ∈F

13
XX X X
qf ib + qf b′ b ≤ T2H (φwb ) ∀b ∈ B2 ∪ B̄2 (19)
f ∈F i∈I2 f ∈F b∈B1 ∪B̄1
XX
P
τπbp ≤ Tm φ̂wb ∀b ∈ Bm ∪ B̄m , m ∈ M (20)
π∈Π p∈P
X X
q̂f b = qf ib + αf µf b − qf bb′ ∀f ∈ F, b ∈ B1 ∪ B̄1 (21)
i∈I1 b′ ∈B2 ∪B̄2
X X
q̂f b = qf ib + qf b ′ b ∀f ∈ F, b ∈ B2 ∪ B̄2 (22)
i∈I2 b′ ∈B1 ∪B̄1
X
q̂f b = ωf πb ∀b ∈ B, f ∈ F (23)
π∈Π
XX X
qf ib + αf µf b ≤ ub ∀b ∈ B1 ∪ B̄1 (24)
f ∈F i∈I1 f ∈F
X
βf πmp τπbp + γf πb = ωf πb ∀f ∈ F, π ∈ Π, b ∈ Bm ∪ B̄m , m ∈ M (25)
p∈P
X
ϑπbjp = τπbp ∀π ∈ Π, p ∈ P, b ∈ Bm ∪ B̄m , m ∈ M (26)
j∈Jm
X X
ϑπbjp + λπj = Dπj ∀π ∈ Π, j ∈ Jm , m ∈ M (27)
p∈P b∈Bm ∪B̄m

xb ∈ {0, 1} ∀b ∈ B̄1 ∪ B̄2 (28)

ylb ∈ {0, 1}; ub ≥ 0 ∀b ∈ B1 ∪ B̄1 (29)

tkb ∈ Z+ ; vb ≥ 0 ∀k ∈ K, b ∈ B (30)

pb ∈ Z+ ; wb ≥ 0 ∀k ∈ K, b ∈ B (31)

qf ib ≥ 0 ∀f ∈ F, i ∈ Im1 , b ∈ Bm2 ∪ B̄m2 , m1 , m2 ∈ M, m1 ̸= m2 (32)

µf b ≥ 0; qf bb′ ≥ 0 ∀f ∈ F, b ∈ B1 ∪ B̄1 , b′ ∈ B2 ∪ B̄2 (33)

τπbp ∈ Z+ ∀π ∈ Π, b ∈ B, p ∈ P (34)

ϑπbjp ∈ Z+ π ∈ Π, b ∈ B, j ∈ J, p ∈ P (35)

θf i ≥ 0 ∀f ∈ F, i ∈ I (36)

γf πb ≥ 0; ωf πb ≥ 0 ∀f ∈ F, π ∈ Π, b ∈ B (37)

qbf b ≥ 0 ∀f ∈ F, b ∈ B (38)

λπj ≥ 0 ∀π ∈ Π, j ∈ J. (39)

14
 Objective function: The first two terms in (1) represent the fixed cost of adding a
foodbank/FRRA to the network and the storage capacity expansion cost at an existing or new Tier-
I foodbank. The next three terms correspond to costs for transportation capacity, and staff capacity
expansion, respectively, at the new and existing foodbanks/FRRAs. The sixth term captures the
cost of procuring the permitted food (αf = 1) at a Tier-I foodbank. The next three terms represent
flow costs corresponding to (i) donated food from donors to foodbanks/FRRAs, (ii) Tier-I to Tier-
II transfers, and (iii) packets from foodbanks/FRRAs to beneficiary nodes. Finally, the last three
terms in (1) capture the penalties for unmet demand at beneficiary nodes, uncollected donation at
donor nodes, and collected but undistributed food at foodbank/FRRA nodes, respectively.

Constraints: We ensure that at most one of the predetermined storage capacity levels is chosen
at an existing (see constraint (2)) or a new (see constraint (3)) Tier-I foodbank. The effective
storage capacity (i.e., initial plus expansion) is calculated in (4) and (5) for the existing and new
Tier-I foodbanks, respectively. Similarly, effective transport capacities for the existing and new
foodbanks/FRRAs of both tiers are presented by (6) and (7), respectively, while (8) and (9) set
the upper limits to those transport capacity expansions. In the same manner, while (10) and (11)
present the effective staff capacities at the existing and new foodbanks/FRRAs, (12) and (13) set
their upper limits. Next, (14) is the donors’ supply constraint, while (15), (16), and (17) ensure
that the inflow, transfer, and outflow quantities at the foodbanks/FRRAs do not exceed their
respective frequency-weighted (i.e., daily vs. weekly) transportation capacities. Similarly, (18)
and (19) for the two tiers ensure that the frequency-weighted effective workforce capacities are
adequate for handling the net of donation, procurement, and transfer. Constraint (20) represents
the workforce requirement for packaging. Constraints (21) and (22) calculate the net food available
for distribution at individual Tier-I and Tier-II foodbanks/FRRAs. Equation (23) handles the
allocation of net quantities of different food types to different beneficiary types. Next, (24) is the
effective storage capacity constraint for Tier-I foodbanks. With the predetermined compositions
to fulfil the nutritional needs of different beneficiary types of both tiers, the total allocated food is
packed into beneficiary-specific packets following equation (25), and it also captures any collected
but unused food (waste) at a foodbank/FRRA. Next, (26) – (27) together enforce the outflow of
packets from foodbanks/FRRAs to beneficiaries. The demand constraint (27) also captures the
unmet demand of a beneficiary node via the shortage variable λπj . Finally, (28) – (39) present all
the decision variables.

15
3.3 A Numerical Illustration
We now present a toy instance along with a solution in Figure 4 to illustrate how our proposed
system works. Tier-I contains Donor 1, an existing foodbank FB1 , a potential foodbank FB3 , and
two beneficiary nodes (school with 200 children; old age home with 50 adults). Similarly, Tier-II
contains Donor 2, one existing and one potential FRRA (FRRA2 and FRRA4 , respectively), and
two beneficiary nodes (slum with 50 children, 100 adults; and shelter with 50 adults). While Donor
1 donates 10,000 units of F1, the Donor 2 sends 4,500 units of F2 food, while leaving behind 50
units as the uncollected food quantity. In Figure 4, each beneficiary node’s weekly demand (=

Procured:
School
F1 = 391,000
FB1 (Existing Food Bank)
1400 (0)
𝐒𝐭𝐨𝐫𝐚𝐠𝐞 𝐂𝐚𝐩𝐚𝐜𝐢𝐭𝐲 = 𝟒𝟎𝟏, 𝟎𝟎𝟎. 773 (P1)
773 (P1) NIL
Net F1
F1 = 10,000
Donor 1 = (391,000 +10,000 NIL 627 (P1)
– 285,050) Old age home
F1: 10,000
= 115,950 NIL
FB3 (New)
350 (0)
Procured: 𝐒𝐭𝐨𝐫𝐚𝐠𝐞 𝐂𝐚𝐩𝐚𝐜𝐢𝐭𝐲 = 𝟒𝟎𝟎, 𝟎𝟎𝟎.

F1 = 399,900 Net F1 350 (P3)

627 (P1)
F1: 285,050 = (399,900 – 198,500

- 2350) 350 (P3)

TIER-I
F1: 198,500 = 199,050

Slum
TIER-II FRRA2 (Existing FRRA)
F1: 2,350

316 (P1) 316 (P1) 350

𝐒𝐭𝐨𝐫𝐚𝐠𝐞 𝐂𝐚𝐩𝐚𝐜𝐢𝐭𝐲 = 𝟎
F2 = 4,950
Donor 2 Net F1 = (285,050 + 198,500) 33 (P2) 700 (125)
33 (P2)
F2: 5,000 = 483,550
570 (P1) Shelter
Net F2 = 4,950 920 (P1) 1 (P1) 5 (P1)
Uncollected: F2 = 50 350 (P1) NIL

Legends: 350 (0)

FRRA4 (𝐍𝐞𝐰)
Packet for Child beneficiary
𝐒𝐭𝐨𝐫𝐚𝐠𝐞 𝐂𝐚𝐩𝐚𝐜𝐢𝐭𝐲 = 𝟎
1 (P1)
Packet for Adult beneficiary

Demand (Shortage)
Net F1 = 2350 5 (P1)

Figure 4: (Colour online) Illustration of a toy problem instance’s solution.

number of beneficiaries x 7) and unmet demand (in parenthesis) are shown.

An optimal solution of this problem prescribes opening of FB3 and FRRA4 , adding storage
capacities at both Tier-I foodbanks, procure F1 type food of 391,000 units at FB1 and 399,900 units
at FB3 , and transfer (follow downward dark arrows) certain amount of F1 from Tier-I foodbanks
to Tier-II FRRAs. After all these transfers, the workforce at foodbanks/FRRAs convert the net

16
food (consisting of F1 and/or F2) into P1, P2, or P3 types of packets by following the predefined
compositions of F1 and F2 (see Table 4), and then, send those to meet the beneficiaries’ demands.

PKT 1 PKT 1 PKT 2 PKT 3

Tier-I Tier-II
F1 F2 F1 F2 F1 F2 F1 F2
Child 150 0 Child 350 0 150 150 0 450
Adult 300 0 Adult 400 0 250 250 0 500

Table 4: Toy example parameters: Tier-I and Tier-II food packet compositions.

4 Case study
We conduct a case study focusing on supply chain network design for the foodbanks and FRRAs
operating in Delhi and the National Capital Region (NCR) of India. The choice of the study region
is motivated by the fact that India’s first foodbank was opened in Delhi in 2012 (The Hindu, 2012),
and this region has several foodbanks that can be categorized into two tiers as per our problem
setting. Figure 5 shows the spatial distribution of the donors, foodbanks and FRRAs, beneficiaries
of Tier-I & II for the proposed network design at our case study region. We discuss data sources
for the case study in §4.1, followed by presenting the base model’s solution in §4.2.

Legends:

Figure 5: (Colour online) The case study region in Delhi, India.

17
4.1 Data Sources
We now discuss how the model parameters for the donor, foodbank/FRRA, and beneficiary node
sets, nutritional requirements and food packet composition are arranged for our case study.

Donors: Our study region (see Figure 5) includes several Tier-I institutional donors (e.g., KFC,
Cargill India, Kellogg India, Britannia Industries, Nestle India, ITC, and Hindustan Unilever) that
are primarily multinational companies from the FMCG and food industry (India FoodBanking
Network, 2023). Tier-II donors primarily include restaurants, small food joints, hostels, office
canteens, religious organizations, etc. Obtaining authentic data on the donation quantity and
food type from all donors is challenging due to the current ad-hoc nature of foodbank operations
that heavily rely on the volunteers’ day-to-day availability for donation collection and distribution,
lacking enough resources for conducting record-keeping. To resolve this data-collection challenge,
we divide our study region into six zones based on the pin codes (ZIP code or postcode equivalent in
India) of the institutional donors. Our correspondence with two foodbank personnel from each tier
provides us with representative aggregate-level estimates of donation quantities from those zones.
Thus, we obtain two sets of donors, six in Tier-I and another six in Tier-II. Furthermore, instead
of tracking itemized donation, we consider six food groups, namely, Cereals & millets (grains),
Pulses, Dairy & poultry, Fruit & vegetables, Snacks, and Cooked food, that are in compliance with
the dietary guidelines issued by National Institute of Nutrition (2011). We present the donation
quantities from all donor node in Table 5.

Tier-I Cereals Pulses Dairy & Fruit & Snacks Tier-II Cooked
Donors & millets poultry vegetables Donors food
1 117 114 39 275 142 7 213
2 164 125 58 254 0 8 392
3 234 67 0 0 0 9 704
4 217 78 28 168 171 10 36
5 249 94 119 234 75 11 433
6 – – 184 219 76 12 223

Table 5: Donation quantities (in Kilograms) of the Donors in Tier-I and II.

Foodbanks/FRRAs: We identify and classify a total of 15 organizations in our study region as

five foodbanks (Tier-I) and 10 FRRAs (Tier-II) that currently operate independent of one another
(see Table 6). For the case study, we consider all these 15 entities as potential as opposed to existing
foodbanks/FRRAs since they operate in silos and our model would determine which of those should
be selected for inclusion in our proposed integrated network. If a node is selected for inclusion, it
can start operating with its existing storage (if it is a foodbank of Tier-I), transport, and workforce
capacities as the initial capacities. We, however, anonymize the organizations and add small

18
 No. Foodbank/FRRA Website Specific Tier
Characteristics
1 India FoodBanking https://www.indiafoodbanking.org/ Distributes packaged I
Network (IFBN) foods
2 Mera Pariwar https://www.meraparivar.org/ Feeds school children I
3 Delhi Food Banking https://www.responsenet.org/ Foodbank I
Network
4 SYPM https://spym.org/ Focus on homeless I
shelters, accepts
surplus food; cold
store and logistics
available.
5 National Organization https://www.nationalngo.org/ Distributes surplus I
For Social Empower- food
ment
6 Robin Hood Army https://robinhoodarmy.com/ Food recovery & redis- II
tribution
7 Sangarsh India https://sangharshindia.org/ Food recovery & redis- II
tribution
8 Roti Bank (AIBRT) https://www.rotibankdelhi.org/ Food recovery & redis- II
tribution
9 Feeding India https://www.feedingindia.org/ Food recovery & redis- II
tribution
10 Ashray Adhikar Ab- https://www.homelesspeople.in/ Food recovery & redis- II
hiyan tribution
11 Indian Roti Bank https://www.rotibank.co.in/ Food recovery & redis- II
roti-bank-in-delhi/ tribution
12 Jeevan Hi Udeshya https://www.facebook.com/ Food recovery & redis- II
jeewanhiudeshya/ tribution
13 Project Jeevan https://indianlocalfoods.com/ Food recovery & redis- II
ngo-for-food-donation-in-india/ tribution
14 Khana Daan https://www.facebook.com/daanthali/ Freshly cooked food II
15 Little India Founda- https://little-india-foundation. Food recovery & redis- II
tion business.site/ tribution; cooking

Table 6: Foodbanks in the study region classified into Tier-I & Tier-II. Source: IFSA (2023)

perturbations in their latitude/longitude data while constructing the potential foodbank/FRRA

node-set. To obtain capacity data, we reached out to the foodbanks/FRRAs, however, did not
receive reliable information on their effective capacities. One foodbank personnel explained that
although they had about 200 registered volunteers, most are irregular, and often the number of
‘active’ volunteers reduces to single digits. With some realistic inputs, we estimate the initial
capacities of the potential foodbanks as shown in Table 7.

Beneficiaries: We consider 76 beneficiary locations in our study region, out of which 16 are
categorized as Tier-I and the remaining 60 as Tier-II. While Tier-I locations include schools, child
welfare centers, and old age homes, Tier-II locations represent certain areas with underprivileged
populations, e.g., slums, shelters, immigrant neighborhoods, etc. Considering the significant

19
 Tier (entity) Initial Capacity
Storage Transport Volunteers Staff
I (Foodbank) 1,000 kg 1,000 kg 5 2
II (FRRA) – 200 kg 5 2

Table 7: Initial capacity estimates of the potential foodbanks/FRRAs.

difference in the nutritional requirement of children and adults, and the higher socio-economic
vulnerability of Tier-II beneficiaries, we create four distinct beneficiary types, specifically, child
at Tier-I, adult at Tier-I, child at Tier-II, and adult at Tier-II. Estimating the demand of Tier-I
beneficiaries is relatively straightforward from their institutional enrollment records. However,
for the Tier-II beneficiary nodes, as we obtained aggregate level demand estimates through
correspondence with foodbank personnel, to ensure a consistency in parameter estimation, we
assumed that the numbers of child and adult beneficiaries in Tier-II were in the proportion of 44.3%
and 55.7%, respectively, following these age groups’ representation in the Census of India (2011).
Then, for the Tier-II beneficiary nodes, we multiplied their population with these percentages to
obtain the estimate of each beneficiary type, as presented in Table 8.

Tier Beneficiary locations Number of Total beneficiary count

locations
Child Adult Total
Schools and Child welfare 12 6,377 – 6,377
I
centers
Old age homes 4 – 594 594
II Slum, Shelter, and other 60 5,058 6,353 11,411
underprivileged areas
Total 76 11,435 6,947 18,382

Table 8: Estimated population of different beneficiary types in Tier-I & II.

Nutrition-related data: Following the dietary guideline from National Institute of Nutrition
(2011), we show the nutritional requirements of different beneficiary types in Table 9, with the
purchase price of each food group obtained from AgMarkNet (2023). These nutritional requirements
are considered in full (i.e. 100%) for the more vulnerable Tier-II beneficiaries, as they require
extensive support in terms of all meals per day. However, since Tier-I beneficiaries receive this
support in the form of a supplement to the primary nutrition from their institutions (e.g., school,
care home, etc.), we set their requirements at 50% of the values shown in Table 9.

Packet types: With guidance from the nutritional requirements data (National Institute of
Nutrition, 2011), we consider three food packet types (see Table 10) that are made from the
heterogeneous mix of uncooked and cooked food donations. Note, these packet types differ in the
proportion of cooked food ranging from 0% (packet type 1) to 100% (packet type 3). Packet type

20
 Gm/portion Recommended Portion Total Requirement (gm)* Price (|/ kg)
Food group
Child Adult Child Adult Child Adult Source:
AgMarkNet
(2023)
Cereals, millets 30 30 10 15 300 450 30
Pulses 30 30 2 2.5 60 75 140
Dairy, poultry 100 100 5 3 500 300 50
Fruit, vegetables 100 100 5 6 500 600 90
Snacks (sugar, fat) 5 10 15 5 75 50 100
* For Tier-I, take 50% of these values; for Tier-II take as-is.

Table 9: Beneficiary-wise food requirements and purchase prices.

Packet Type 1 Packet Type 2 Packet Type 3

Food group
Child Adult Child Adult Child Adult
Cereals & millets 300 450 150 225 0 0
Pulses 60 75 30 37.5 0 0
Dairy & poultry 500 300 250 150 0 0
Fruit & vegetables 500 600 250 300 0 0
Snacks (sugar & fat sources) 75 50 37.5 25 0 0
Cooked 0 0 717.5 732.5 1435 1465

Table 10: Configuration of food packets (all figures are in gms).

1 consists of only the foods with extended shelf-life, raw, ready-to-cook/eat type items, received
as donations or procured at Tier-I. As per our problem setting, since no donation from Tier-II
(containing cooked food) enters the Tier-I flow, the beneficiaries of Tier-I receive only packet type
1. However, a Tier-II beneficiary can receive any of the three packet types. Packet type 2 is a
mix of both tiers’ donations, therefore, the nutritional requirements given in Table 9 for each food
group, have been halved. Finally, packet type 3 covers a beneficiary’s daily nutritional need entirely
with cooked food received as donation at Tier-II.

Miscellaneous parameters: Apart from those discussed above, our model uses several other
capacity-, cost-, and logistics-related parameters. Table 11 lists them along with their sources for
estimation, wherever possible.

4.2 Base model solution

With parameter settings as explained in §4.1, we solve the case study problem instance. This
solution, referred to hereafter as the ‘base case’, suggests five Tier-I foodbanks (at nodes 1, 2, 3, 4,
5) and three Tier-II FRRAs (at nodes 9, 10, 13) to include in the integrated network. Tables 12
and 13 present the base case solution, illustrating capacity and flow decisions, respectively.

In Tier-I, except for one out of the five foodbanks, all others need storage capacity extensions
(three at their highest levels). FRRAs have no storage capacity, as per our problem setting (shown

21
 Parameter Specifics Capacity Cost (|) Comment/ Source
Small 2,000 kg 15,000
Storage level https://www.99acres.com/
Large 6,000 kg 40,000
Car 200 kg 5,250 Wagon-R, a commonly used vehicle for such
transport job in the study region, has cargo
capacity around 400 kg (source: https://www.
Transport
car.info/). Giving volumetric allowance, we
assume half of its capacity.
Cost of renting: realistic assumption.
Truck 700 kg 14,000 Capacity: https://trucks.cardekho.com/.
Cost of renting: realistic assumption.
Donation 100 kg 3,500 per In India |500 is approx daily
handled per person minimum wage. (Source: https:
person per week //www.india-briefing.com/news/
guide-minimum-wage-india-2023-19406.
html/)
Workforce
Packets 1,000 Estimated based on correspondence with food-
handled/ packets bank staff.
person/week
Cost of 10,000 Cost incurred due to miscellaneous initiatives,
adding new e.g., advertising, office setup, utilities etc.
foodbank
/FRRA

Table 11: Estimation of miscellaneous model parameters.

Tier-I Storage (gm) Transport (gm) Workforce (# persons)

Foodbanks
Initial Extension (level) Initial Extension Initial Extension
1 1,000,000 0 1,000,000 0 7 3
2 1,000,000 6,000,000 (large) 1,000,000 0 7 63
3 1,000,000 6,000,000 (large) 1,000,000 0 7 63
4 1,000,000 2,000,000 (small) 1,000,000 0 7 23
5 1,000,000P 6,000,000 (large) 1,000,000 P 0 7 P 59
= 25,000,000 = 5,000,000 = 246
Tier-II Storage (gm) Transport (gm) Workforce (# persons)
FRRAs
9 – – 200,000 900,000 (1 car, 1 truck) 7 3
10 – – 200,000 700,000 (1 truck) 7 1
13 – – 200,000 P 400,000 (2 cars) 7 P 0
= 2,600,000 = 25

Table 12: Base case solution: capacity decisions.

by ‘–’ in Table 12). A significant workforce extension is observed at almost all Tier-I foodbanks
(totaling 246 persons), compared to Tier-II FRRAs (25 persons), because Tier-I foodbanks require
a larger workforce to handle weekly inflows of donated and procured bulk foods (packing and
distribution are done daily). On the other hand, Tier-II FRRAs handle daily inflow (donations at
Tier-II plus transfers from Tier-I), packaging, and distribution, thereby requiring less workforce. No
transportation capacity addition is needed at Tier-I foodbanks but up to two vehicles of different

22
capacities are deployed at all the three chosen Tier-II FRRAs. Although Tier-I is involved in large
quantities of food purchases, as delivery of those is managed by sellers, foodbanks need not increase
transportation capacity to handle bigger purchases.
SUPPLY SIDE: Flow of food items (in gms.) DEMAND SIDE: Packet (#)
Donation Purchase Transfer (out) Transfer (in) Net for PKT PKT PKT Total
(in) (in) to Tier-II from Tier-I Distribution 1 2 3 PKT
Tier-I FB
1 799,000 201,000 1,240 – 998,760 0 511 185 696
2 687,000 6,313,000 70 – 6,999,930 9 4,868 1 4,878
3 472,000 6,513,580 5,770,970 – 1,214,610 674 80 80 834
4 942,000 2,058,000 3,000,000 – 0 0 0 0 0
5 601,000 5,999,000 5,792,095 – 807,905 1 0 562 563
10,021,205 6,971
Tier-II FRRA
9 0 – – 6,905,770 6,905,770 4,738 0 0 4,738
10 994,545 – – 4,604,925 5,599,470 2,484 1,374 0 3,858
13 1,006,455 – – 3,053,680 4,060,135 2,128 0 687 2,815
16,565,375 11,411

Table 13: Base case solution: flow decisions.

Table 13, presenting the flow decisions, is organized into two parts. The supply side (left) shows
flows of donated, purchased, and transferred food within and between the two tiers. The column
‘Net for Distribution’ represents the total amount of food available at a foodbank/FRRA for
distribution to the downstream beneficiary nodes over a week’s span. As per our base model
parameters, Tier-II with more beneficiaries, handles around 16.5 thousand kilograms of food, in
comparison with about 10 thousand kilograms in Tier-I. The demand side (right) of Table 13
provides a breakup of different packet types in two tiers. Note that the total number of packets
18,382 (=6,971 + 11,411) matches the total number of beneficiaries in each tier (see Table 8), i.e.,
no shortage occurs.

5 Numerical analysis
To better understand the effects of demand and donation (supply) changes, we undertake
three experiments by systematically varying demand, donation, and both. We discuss the main
observations and insights in §5.1 – §5.3. Moreover, we observe the effects of adding a total budget
constraint in §5.5, and linked with that, a constraint enforcing strategic-to-operational cost ratio
in §5.6.

5.1 Experiment 1 - effects of demand increase

Since in real life most foodbanks face the issue of meeting demand with limited supply, we
study the effects of demand increase. To this end, we systematically increase demands of all

23
 60
(A) Capacity with Demand change
50 (C) Percent of shortage for each beneficiary type with Demand increase
40 Beneficiary Priority Base
Percentage change in capacity

1.2x 1.4x 1.6x 1.8x 2x

Type case
30
Tier-II Child 1.00 0 0 0.00 0.01 0.01 0.01
20

Tier-II Adult 0.95 0 0 0.03 0.06 0.04 0.04

10
Tier-I Child 0.90 0 0 0.01 27.82 55.92 78.85
0 0.85
Tier-I Adult 0 0 9.76 100.00 100.00 100.00
Base case 1.2 1.4 1.6 1.8 2

Storage Transport Workforce

(B) Food availability with Demand change (D) Tier-II Proportion of packets with Demand change
40000
Thousands

100%
90%
30000

Percentage of packets
80%
70%
Quantity (gms)

20000 60%
50%
40%
10000 30%
20%
0 10%
Base case 1.2 1.4 1.6 1.8 2 0%
Base case 1.2 1.4 1.6 1.8 2
Total donation Total purchase
Total food in the system Total food for distribution with Tier I Packet type 1 Packet type 2 Packet type 3
Total food for distribution with Tier II

Figure 6: Effects of demand increase.

beneficiary nodes of both tiers in steps of 20%, while keeping donation quantities from all the
donors unchanged. Exhibits (A) to (D) of Figure 6 show these steps as ‘Base case’, followed
by ’1.2x’, ‘1.4x’, . . . , ‘2x’, representing beneficiaries’ demand increase by 20%, 40%, . . . , 100%.
Figure 6(A) illustrates the capacity increase in storage, transport, and workforce categories due to
a strain in the system triggered by a demand increase. However, the changes are not linear and
dissimilar for different capacity categories in our case study parameter settings. While the ‘1.2x’
case causes an approximate 20% increase in both storage and workforce capacities, less than 5%
transport capacity addition is observed. Case ‘1.4x’ and beyond, no further storage capacity increase
occurs. At this point, with all five Tier-I foodbanks forced to operate with their maximum possible
level of storage capacities, storage capacity becomes the bottleneck. Transportation and workforce
capacities increase, however, at slightly different rates as the beneficiary demand further increases.
Figure 6(B) represents the system-wide food availability through donations and purchases. While
donation quantities are kept fixed, food purchase increases at Tier-I to satisfy the rising demand.
However, as procured food needs storage, whose capacity becomes the bottleneck at the ‘1.4x’ case,
the food purchase level becomes constant as demand further increases. This also explains why

24
shortage is observed at this level and with further demand increase since procurement increase
no longer helps. Furthermore, two diverging lines in Figure 6(B), showing the changes in food
availability at Tier-I and Tier-II foodbanks and FRRAs respectively, illustrate some interesting
aspects of sharing limited food between two tiers. Note that with the same demand increase,
Tier-II receives a higher share of available food (steady increase) while the Tier-I receives less
(steady decrease after ‘1.4x’ case). This is primarily dictated by the higher priorities for the Tier-II
beneficiaries compared to their Tier-I counterparts (see Figure 6(C)). With demand increase and
donations remaining unchanged, the shortage profile expressed as percentages of the total number
of beneficiaries in each of the four types (i.e., Tier-I child, Tier-II adult, etc.), is presented in Figure
6(C). As discussed above, no shortage occurs at ‘1.2x’, and thereafter, it gradually increases while
obeying the relative priorities of different beneficiaries. Hence, almost 99.99% of Tier-II children’s
demands are met in all cases up to ‘2x’. On the other hand, the needs of adults in Tier-I remain
totally unmet from the ‘1.6x’ case. A similar pattern is observed for the children of Tier-I, however,
due to their priority weight of 0.90, unlike the adults, their shortage remains around 79% even in
the ‘2x’ case. We summarize this as a key observation below.

Key observation 1: In the absence of any budget constraint, demand increase can be handled
by an increase in procurement and different capacity additions up to a point, after which Tier-I
foodbanks’ storage capacities become bottleneck, and shortage is observed as per relative priorities
of the beneficiary types.

As demand increases to ‘1.6x’ and above, despite shortages, part of the available donation is not
collected from Tier-II donors, and penalties for both shortage and uncollected food are charged. This
interesting phenomenon occurs because demand satisfaction is not linearly related to the donation
quantity, it rather depends on the compositions of packet types (see Table 10), and, in turn, the
nutritional content of the donated food. With demand increase, our previous observation on the
increase in purchase (essential to meet nutritional needs) until the Tier-I foodbank’s storage capacity
is exhausted (see Figure 6(B)), explains leaving some donated food as uncollected at Tier-II donor
sites even when there are shortages in the system. Figure 6(D), showing the increased proportions
of packet type 1 at Tier-II with increasing demand further highlights the importance of considering
nutritional aspect in the decision-making process. Although donation quantity remains unchanged
at Tier-II, its relative contribution (low shelf-life food) gradually reduces with the demand increase.
Therefore, food collected and purchased at Tier-I, being nutritionally more significant and the major
composition of packet type 1, largely contributes in Tier-II’s demand fulfillment.

Key observation 2: After one level of demand increase, the increased procurement at Tier-

25
I forces making more of packet types 1 and 2, following their predetermined compositions.
Consequently, donation at Tier-II becomes less important, leaving them partly uncollected.

5.2 Experiment 2 - effects of donation change

To observe the effects of donation quantity change, we conduct an experiment where donations
in both tiers are systematically changed (increased or decreased) from the base case values in steps
of 20%. No change is made in demand or other model parameters.

As donation varies, interesting changes in different capacities are captured in Figure 7(A). Observe
that storage capacities are insensitive to changes in donation between ‘0.8x’ and ‘1.6x’. With
donation decrease ‘0.6x’ or below, purchase increases to compensate for the shortfall, thereby
increasing the storage requirement in Tier-I. In donation increase of ‘1.6x’ and above, (which
includes Tier-II donation as well, requiring no storage) we observe a gradual decrease in storage
requirement. A similar pattern is observed for workforce capacity. However, transport capacity
changes in the other way: after a 40% donation increase, more capacity is needed to collect the
increased donation at both tiers and to conduct the Tier-I to Tier-II transfers. Beyond ‘2x’ donation
increase, we observe uncollected donations (not shown in exhibits), therefore, do not explore further.

Figure 7 (B) depicts the changes in food donation and purchase quantities, and availability in each
tier and in the system. We observe a linear decrease in purchases as donations increase, keeping
the total quantity unchanged. Also, Figure 7(C) shows that donation increase causes a small
decrease in strategic cost (foodbank opening and capacity building) and a significant decrease in
the operational cost (transport and purchase). While an increase in donation at both tiers reduces
the system-wide purchase requirement (operational cost savings), some strategic cost savings occur.
With the increase in Tier-II’s donation, the dependency on Tier-I is relieved to some extent, which,
in turn, reduces Tier-I’s storage requirements, thereby, saving on the strategic cost. In our case
study’s parameter setting, no shortage is incurred at any of the cases between ‘0x’ - ‘2x’, because
a donation decrease is always compensated by increased purchase at Tier-I. Figure 7(D) provides
interesting observations about packet composition change in Tier-II with the changes in donation.
In the extreme case of zero donation, only Packet Type 1 is formed from all the procured food
at Tier-I. As donations increase at both tiers, Packet Types 2 and 3 are also made with the
cooked/ready-to-eat food donation received at Tier-II. However, this leads to a key observation
with a practical recommendation as follows.

Key observation 3: Although packet type 1 with the highest proportions of uncooked food can
be instrumental in addressing demand increase via additional procurement at Tier-I, it would pose

26
 (A) Capacity change (%) with donation change (C) Cost proportions with donation change
25 3000

Thousands
Percentage change in capacity

20
2500
15
2000

Cost (INR)
10
1500
5 Donation multiplier
0 1000

-5 500
-10
0 0.2 0.4 0.6 0.8 Base 1.2 1.4 1.6 1.8 2 0
case 0 0.2 0.4 0.6 0.8 Base 1.2 1.4 1.6 1.8 2
case
Storage Transport Workforce
Strategic cost Operational Cost

(B) Food availability with donation change (D) Tier-II proportion of packets with donation change
30000 100
Quantity (gm) Thousands

25000
80

Percentage of packets
20000
60
15000
40
10000

20
5000

0 Donation multiplier 0
0 0.2 0.4 0.6 0.8 Base 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 Base 1.2 1.4 1.6 1.8 2
case case
Donation multiplier
Total donation Total purchase
Packet type 1 Packet type 2 Packet type 3
Total food in the system Total food for distribution with Tier I
Total food for distribution with Tier II

Figure 7: Effects of donation change (increase and decrease from Base case).

practical issues to the poorest Tier-II beneficiaries without provisions for self-cooking at shelters.
For those, packet types 2 and 3, reducing the challenges of on-site cooking and feeding by limited
foodbank volunteers, would be preferred. Therefore, initiatives for encouraging cooked or ready-to-
eat food donation from Tier-II donors can help in reducing the pressure on Tier-I as well as the
integrated system.

5.3 Experiment 3 - simultaneous donation decrease and demand increase

The joint effects of donation decrease and demand increase are studied by changing both
parameters simultaneously in steps of 20% (e.g., demand ’1.2x’ and donation ’0.8x’). Although the
pattern of changing food availability is similar to Figure 6(B), we note that to compensate for the
donation decrease in both tiers, the purchase quantity in Tier-I does not increase proportionately
because the storage capacity of Tier-I foodbanks becomes bottleneck at demand ’1.4x’ and donation
’0.6x’ (similar to Experiment 1). Additionally, as shown in Table 14, demand increase triggers
shortage at this point to a much higher extent compared to Experiment 1. Specifically, shortages
in Tier-I adults aggravate sharply (shown in boldface) from the ’1.4x demand, 0.6x supply’ case.

27
 Multipliers (Demand, Supply)
Beneficiary Priority Base case (1.2x, 0.8x) (1.4x, 0.6x) (1.6x, 0.4x) (1.8x, 0.2x) (2x, 0x)
Type
Tier-II Child 1.00 0 0 0 0.04 0.03 0.01
Tier-II Adult 0.95 0 0 0 0.03 0.03 0.02
Tier-I Child 0.90 0 0 0 36.01 65.65 89.79
Tier-I Adult 0.85 0 0 76.02 100.00 100 .00 100.00

Table 14: Percent of shortage for each beneficiary type in Experiment 3.

Foodbanks and FRRAs opened in Experiment 1-3: The systematic change in demand
and donation quantities made in experiments 1 - 3 influences the network structure as presented in
Table 15. In Experiment 1 (donation constant), Tier-I procurement increases to satisfy the growing
demand. The need for more storage to accommodate this increased food has forced more foodbanks
to open, mostly with their highest levels of storage capacities. In Experiment 2, to compensate
for donation change, a systematic adjustment occurs in purchase quantity. Since demand stays the
same, the storage need is also unchanged, keeping the same eight foodbanks/FRRAs open in all the
cases we present. In Experiment 3, we observe a gradual increase in the number of foodbanks, all
operating at their highest levels of storage capacities (similar to Experiment 1), to accommodate
the increased purchases compensating for the donation decrease.

Experiment 1: demand increase, donation same

Demand multiplier Base case 1.2x 1.4x 1.6x 1.8x 2x
# Foodbanks/FRRAs 8 8 12 13 13 14
Experiment 2: donation increase, demand same
Donation multiplier Base case 1.2x 1.4x 1.6x 1.8x 2x
# Foodbanks/FRRAs 8 8 8 8 8 8
Experiment 3: donation decrease, demand increase
Demand multiplier Base case 1.2x 1.4x 1.6x 1.8x 2x
Donation multiplier Base case 0.8x 0.6x 0.4x 0.2x 0x
# Foodbanks/FRRAs 8 9 9 13 13 14

Table 15: Number of opened Foodbanks in Experiments 1 - 3.

5.4 The value of integration

The discussion on the effects of demand and donation variations in §5.1 – §5.3 would help in
appreciating the value of integration of the two tiers in our proposed network. For a quantitative
understanding, we cut the connection between two tiers by fixing all transfer variables qf bb′ to
zero, and re-solve our model for (i) base case, and (ii) simultaneous demand increase and donation
decrease. In (i), we observe a 40%, 26%, and 55% decrease in storage, transport, and workforce

28
capacities, respectively, from the base case solution. Clearly, the disintegrated system - having Tier-
II FRRAs with no storage capacities - requires much less of total capacity than the integrated one,
and incurs 65% less cost (excluding penalties). However, in the absence of the integrated system’s
transfer mechanism, as opposed to ‘no shortage’ in the base case of the integrated setting, Tier-II
of the disintegrated system exhibits a staggering shortage of 72.5% and 100% of child and adult
beneficiaries, respectively. The shortage worsens under the Experiment 3 setting (simultaneous
demand increase and donation decrease) as shown in Table 16, and a comparison with Table 14
entries would emphasize the value of integration.

Multipliers (Demand, Supply)

Beneficiary Priority Base case (1.2x, 0.8x) (1.4x, 0.6x) (1.6x, 0.4x) (1.8x, 0.2x) (2x, 0x)
Type
Tier-II Child 1.00 72.47 81.55 88.16 93.11 96.94 100
Tier-II Adult 0.95 100.0 100 100 100 100 100
Tier-I Child 0.90 0 0 0.01 0.02 0.01 0
Tier-I Adult 0.85 0 0 0.12 0 0.19 0

Table 16: Shortage % with demand increase and donation decrease in disintegrated system.

5.5 Adding budget constraint

While our proposed model [P] without a budget constraint, determines the (baseline) cost for
establishing an integrated foodbank network, understanding the impact of a tight budget can be
valuable to a decision-maker. To better understand this, we add a simple budget constraint by
restricting total cost to an upper bound B, whose initial value is set as the sum of strategic and
operational costs obtained from the base case solution. Next, this B value is gradually tightened by
multiplying its initial value with 0.8, 0.6, ..., 0, and the model is re-solved every time. In our case
study’s parameter setting, we find the budget constraint is binding in all these settings, indicating
the available budget would be fully utilized.

Budget Multipliers:= 0x 0.2x 0.4x 0.6x 0.8x Base case

Network No. of foodbanks/FRRAs 0 6 6 7 7 8
Storage (kg) 0 7,000 13,000 17,000 21,000 25,000
Capacity Transportation (kg) 0 5,900 6,100 6,500 6,800 7,600
Workforce (#) 0 78 126 174 222 271
Tier-II Child (1.0) 100 13.2 3.6 0 0 0
Shortage (%) Tier-II Adult (0.95) 100 100 99.9 80.2 42.6 0
Tier-I Child (0.9) 100 71.8 28.1 2.7 0 0
Tier-I Adult (0.85) 100 100 100 100 35.7 0
Costs Strategic (%) 0 53.3 48.8 45.7 44.7 44.1
Operational (%) 0 46.7 51.2 54.3 55.3 55.8

Table 17: Budget sensitivity analysis.

29
 Table 17 shows interesting changes in the solutions from different aspects such as network
structure, capacity, shortage percentage for different beneficiary types, and proportion of strategic
and operational costs. Additionally, Figure 8 depicts the changes in donation, food availability in
each tier as well as in the system, while the budget tightens (from right to left).

Food availability with Budget change

30000
Thousands

25000
Quantity (gm)

20000

15000

10000

5000

0
0x 0.2x 0.4x 0.6x 0.8x Base case

Budget multiplier

Total donation Total purchase Total food in the system

Total food with Tier I Total food with Tier II

Figure 8: (Colour online) Effects of budget change on food availability.

With budget reduction, overall capacity decreases and although the total number of food-
banks/FRRAs does not change much, the Tier-I foodbanks operate with minimal storage capacities.
At the ’0.2x’ setting, Figure 8 shows a significant reduction in storage and workforce capacities
(highlighted in red). Table 17 presents the shortage percentage for different beneficiary types.
Since capacity is closely associated with donation collection, distribution, and transfer, Figure 8
further explains the almost linear decrease in food quantity from right to left. Note that although
until ’0.6x’ setting, all available donations are collected, the ’0.4x’ onwards some donations remain
uncollected (at Tier-II), indicating the capacity crunch. A sharp decrease in storage and workforce
capacities (more than 70 % of base) is observed at ’0.2x’, whose reflection is also evident in Figure
8 between ’0x’ and ’0.2x’. Finally, with the tightening of budget, the strategic and operational cost
proportions also change. Corroborating with our above observations for the ’0.2x’ case, we observe
that although a minimal investment in capacity is made to sustain the system, budgets are reduced
for food purchase and its distribution.

Key observation 4: Although in non-profit settings, expenditure towards tangible services to

beneficiaries is preferred over capacity building, for a system with inadequate infrastructure and
running on a tight budget, the beneficiaries cannot be served without capacity building.

30
5.6 Adding strategic-to-operational cost proportionality constraint
We conduct this analysis to further explore the effect of changing strategic vs operational cost
proportions, by adding a cost proportionality constraint to the base model to ensure: “strategic
cost ≤ µ× operational cost”, where µ is a fraction. The budget constraint discussed in §5.5
is removed before running this experiment. Through this new constraint, we establish a clear
dominance of operational cost over strategic cost. While gradually varying µ and re-solving our
model, we present important components from the optimal solution in Table 18. Since in the
base case solution of [P], we observe the ratio of strategic to operational cost to be 0.79, for
this experiment, we only consider a systematic reduction of µ from that value. In Table 18,
note that with µ = 0.4, all capacities (storage, transport, and workforce) reduce drastically and
all beneficiaries experience huge shortages. In the line of our discussion in §5.5 leading to key
observation 4, this experiment again underlines the danger of exercising frugality in infrastructure
building, which is essential to ensure the serving of the beneficiaries.

µ := 0.3 0.4 0.5 0.6 0.7 Base case (0.79)

Network No of foodbanks/FRRAs 6 6 8 8 8 8
Capacity Storage (kg) 5,000 5,000 35,000 27,000 25,000 25,000
Transportation (kg) 5,400 5,400 7,400 7,400 7,400 7600
Workforce (#) 42 57 367 295 274 271
Shortage (%) Tier-II Child 61.45 67.18 0 0 0 0
Tier-II Adult 99.98 100.00 0 0 0 0
Tier-I Child 90.58 62.86 0 0 0 0
Tier-I Adult 98.65 100.00 0 0 0 0

Table 18: Effects of adding a Strategic-to-operational cost constraint.

5.7 Equity consideration with different granularities

When demand overwhelms donation quantity and system’s capacity, even the increased
procurement at Tier-I becomes inadequate to shortages as the storage capacity bottleneck hits.
In such resource-constrained situations, it is critical for the foodbank system (and any non-profit
organization) to maintain equity or fairness in distribution. Among different ways of incorporating
equity in an optimization model, we adopt the egalitarian approach of adding a minimax term in
the objective function of our model [P]. Specifically, we remove the last term from the objective
function expression (1) that represents the penalty for beneficiaries’ shortages, and call it Z1 . Then
we add to Z1 , the appropriate minimax term (see second column of Table 19) with a multiplier M
(to make the minimax term’s magnitude comparable with Z1 ). As the addition of the minimax
term makes the objective function nonlinear, we linearize it by the standard technique of replacing

31
the minimax expression with an auxiliary variable λ
e with appropriate indices and moving the latter

in constraint. Table 19 presents five minimax variants and their linearization schemes. Since our
base case solution for model [P] does not incur shortage for any beneficiary (see §4.2), we build
a comprehensive case of system-wide shortage by making demand to be twice the base case value
(i.e., ‘2x’ setting of Experiment 1). Table 20 presents the detailed shortage analysis for five equity
variants, and in the last column, adds the solution without equity consideration (i.e., solution of [P]
with ‘2x’ demand). Observe that although total shortage is not too different across the six columns,
their distribution among tiers and beneficiary types change interestingly. Below, we explain equity
variants representing different granularities and their effects on the shortage distribution.
# Objective Function (minimax) Linearized Objective Constraints added
E1.  
Min Z1 + M × max λπj Min Z1 + M × λ
e e ≥ λπj
λ j ∈ J, π ∈ Π.
π∈Π,j∈J

E2. P
P
λπj
π∈Π
π∈Π λπj
 
X X em ≥ P
λ ∀m ∈
Min Z1 + M × max Min Z1 + M × λ
em P
Dπ′ j ′
π ′ ∈Π j ′ ∈J
P P
m∈M
j∈Jm
π ′ ∈Π j ′ ∈J m
Dπ ′ j ′ m∈M m
M, j ∈ Jm ; λ
em ≥ 0 ∀m ∈ M.
E3.
eπ ≥ P λπj
 
X λπj X
λ ∀j ∈ J, π ∈ Π;
Min Z1 + M × max P Min Z1 + M × λ
eπ
j ′ ∈J Dπj ′
j ′ ∈J Dπj ′
j∈J π∈Π
π∈Π eπ ≥ 0 ∀π ∈ Π.
λ
E4.
emπ ≥ λπj ∀m ∈ M, j ∈ Jm , π ∈
 
X X λπj X X
λ
Min Z1 + M × max Min Z1 + M × λ
emπ
Dπj
m∈M π∈Π
j∈Jm Dπj m∈M π∈Π emπ ≥ 0 ∀m ∈ M, π ∈ Π.
Π; λ
E5.
emπ ≥ λπj ∀m ∈ M, j ∈ Jm , π ∈ Π;
 
X X λπj X X
λ
Min Z1 + M × νπm max Min Z1 + M × νπm λ
emπ
Dπj
m∈M π∈Π
j∈Jm Dπj m∈M π∈Π emπ ≥ 0 ∀m ∈ M, π ∈ Π.
λ

Table 19: Five equity variants - corresponding objective function and constraints.

Through experiments E1 to E5, we exhibit gradual progress in the equity representations. E1

represents the most basic approach among the five variants of equity by considering shortages at
both tiers’ beneficiary nodes and beneficiary types equally. This is reflected in Table 20 entries,
where we see very close shortage values for child and adult (5,778 and 5,402), and all positive entries
under ‘Tier and type jointly’. Note that the shortages are distributed much better compared to
corresponding entries under the ‘No Equity’ column. The variant E2 ensures in-tier equity, i.e., the
model attempts to treat shortage proportions (i.e., shortage to demand ratio) of child and adult
beneficiaries at par. We observe some extreme albeit opposite shortage allocations under the E2
and ‘No Equity’ columns. Particularly, E2 ends up allocating all shortages to Tier-II beneficiaries,
ignoring their higher nutritional needs. Therefore, this solution is unacceptable for our case study
problem instance. ‘In-type’ equity is presented by E3, where shortage proportions of children or
adults from different tiers are treated at par. Although we observe a higher shortage in Tier-II

32
(9,334) compared to Tier-I (1,856), and also a higher shortage for children (7,221) compared to
adults (3,969), the values are less extreme than the ‘No Equity’ column entries. In the last two
variants, while E4 represents an in-tier-in-type joint equity without relative penalties to shortages
at specific tiers and beneficiary types, E5 includes that information vis νπm parameter. The higher
relative weights of Tier-II child and adult beneficiaries, in conjunction with the in-tier-in-type equity
constraint, force most shortages to Tier-I (lower priority) in E4 and all shortages in E5.

This experiment shows that the effects of equity considerations on the solution are not
straightforward, and the decision-maker should analyze the pros and cons of different equity variants
before adopting one.
Equities E1 E2 E3 E4 E5 –
P1 P2
Child Child Child Child Child Child Child Child Child Child
Tier-I Tier-II
Tier-I Tier-II Tier-I Tier-II Tier-I Tier-II Tier-I Tier-II hild

P3 P4
Adult Adult Adult Adult
Adult Adult Adult Adult Adult Adult
Tier-I
Tier-I Tier-II Tier-I Tier-II Tier-I Tier-I Tier-II Tier-II
Tier-II
No Equity
Shortage (#) 11,180 11,099 11,190 11,043 11,270 11,251
% shortage 30.41 30.19 30.44 30.04 30.65 30.60
Beneficiary Type
Child 5,778 2,767 7,221 166 11,270 10,058
Adult 5,402 8,332 3,969 10,877 0 1,193
Tier-wise
Tier-I 1,536 0 1,856 166 11,270 11,245
Tier-II 9,644 11,099 9,334 10,877 0 6
Tier and type jointly
Tier-II Child 4,626 2,767 5,625 0 0 1
(νπm = 1.00)
Tier-II Adult 384 0 260 0 0 1,188
(νπm = 0.95)
Tier-I Child 1,152 0 1,596 166 11,270 10,057
(νπm = 0.90)
Tier-I Adult 5,018 8,332 3,709 10,877 0 5
(νπm = 0.85)

Table 20: Analysis of shortages for five equity variants.

6 Concluding remarks and future work

We present an approach of integrating two tiers having distinctly different levels of donors and
foodbanks serving beneficiaries with quite different nutritional requirements. We develop a MIP
model that determines - given the sets of potential foodbanks (in Tier-I; with storage capacity) and
‘food recovery and redistribution agencies’ [FRRA] (in Tier-II; without storage capacity) - which
entities from each tier should be chosen, transport, and workforce capacities, to form an integrated
network, connected with donor and beneficiary nodes of the same tier. Our model estimates the
optimal cost of building such an ecosystem by minimizing the sum of several fixed, capacity-building,

33
procurement, donation collection, and distribution-related costs. In our case study with a mix of
real and realistically estimated parameters, the model produces a base solution (see §4.2), illustrates
the benefit of integration (see §5.4)), and presents several insights by conducting a detailed analysis
of the effects of changing demand (see §5.1), donation quantities (see §5.2), and both of those
simultaneously (see §5.3). Furthermore, we examine the effects of adding a budget constraint (see
§5.5), a strategic-to-operational cost constraint (see §5.6), and equities of different granularities in
the egalitarian approach (see §5.7).

Insights from our numerical analyses would help the centralized system’s decision-maker in
recognizing the right courses of action when demand increases (Key observation 1), when it is better
to leave some donation uncollected (Key observation 2), when to take initiatives for encouraging
cooked food donation from the Tier-II donors (Key observation 3), when investing in capacity
building (strategic cost) becomes important than direct spending for the beneficiaries to serve
them better (Key observation 4).

We end our discussion by indicating some future research directions. First, while we address
donation (supply) and demand fluctuations in numerical analysis, embedding different sources of
uncertainties in the optimization model itself by adopting stochastic programming or a robust
optimization framework is a possibility. Second, the impact of an overall budget constraint and
equity consideration together can be interesting. While this work adopts the egalitarian approach of
expressing equity at different granularities, some alternate equity representations can be examined.
Third, in a multi-objective setting, equity, cost efficiency, and effectiveness (i.e., the 3E’s in nonprofit
operations management) can be modeled as three objective functions. While our work, being a
strategic model, considers donation collection and food packet distribution costs to be proportional
to the distances between node pairs, the associated routing decisions may also be integrated, albeit
with additional computational complexities. Finally, in addition to receiving donations as food
items, the inclusion of financial donations can be considered along with its overheads.

References
AgMarkNet. 2023. Agricultural commodity prices. https://agmarknet.gov.in/
PriceAndArrivals/DatewiseCommodityReport.aspx. Accessed online: 28 September,
2023.

Akkaş, A., V. Gaur. 2022. Om forum—reducing food waste: An operations management research
agenda. Manufacturing & Service Operations Management 24(3) 1261–1275.

Alkaabneh, F., A. Diabat, H. O. Gao. 2021. A unified framework for efficient, effective, and fair
resource allocation by food banks using an approximate dynamic programming approach. Omega
100 102300.

34
Bazerghi, C., F. H. McKay, M. Dunn. 2016. The role of food banks in addressing food insecurity:
a systematic review. Journal of community health 41 732–740.

Campbell, E. C., M. Ross, K. L. Webb. 2013. Improving the nutritional quality of emergency
food: a study of food bank organizational culture, capacity, and practices. Journal of hunger &
environmental nutrition 8(3) 261–280.

Census of India. 2011. Office of Registrar General and Census Commissioner, India, Ministry
of Home Affairs, Govt. of India, New Delhi. www.censusindia.gov.in. Accessed online: 28
August, 2023.

Dalal, J. 2022. A strategic donor-beneficiary assignment problem under supply and demand
uncertainties. International Journal of Operational Research 43(1-2) 188–207.

Davis, L. B., S. X. Jiang, S. D. Morgan, I. A Nuamah, J. R. Terry. 2016. Analysis and prediction
of food donation behavior for a domestic hunger relief organization. International Journal of
Production Economics 182 26–37.

Dubey, N., A. Tanksale. 2022. A study of barriers for adoption and growth of food banks in
india using hybrid dematel and analytic network process. Socio-Economic Planning Sciences 79
101124.

Edmonton’s Food Bank. 2023. Our Humble Beginnings & Present Work. https://www.
edmontonsfoodbank.com/. Accessed online: 9 October, 2023.

Feeding America. 2023. Our history - History of Food Banks. https://www.feedingamerica.

org/about-us/our-history. Accessed online: 9 October, 2023.

Food Safety and Standards Authority of India [FSSAI]. 2023. Save food, share food, share
joy. https://fssai.gov.in/upload/knowledge_hub/696995b34c094e60f8Indian%20Food%
20Sharing%20Alliance.pdf. Accessed online: 9 October, 2023.

Ghahremani-Nahr, J., A. Ghaderi, R. Kian. 2023. A food bank network design examining food
nutritional value and freshness: A multi objective robust fuzzy model. Expert Systems with
Applications 215 119272.

Global Hunger Index. 2023. Global Hunger Index - India. https://www.globalhungerindex.org/

india.html. Accessed online: 9 October, 2023.

Govindan, K., A. Jafarian, R. Khodaverdi, K. Devika. 2014. Two-echelon multiple-vehicle location–

routing problem with time windows for optimization of sustainable supply chain network of
perishable food. International Journal of Production Economics 152 9–28.

Hasnain, T., I. Sengul Orgut, J. S. Ivy. 2021. Elicitation of preference among multiple criteria in
food distribution by food banks. Production and Operations Management 30(12) 4475–4500.

IFSA. 2023. Indian Food Sharing Alliance. https://sharefood.eatrightindia.gov.in/

agency-list.html. Accessed online: 9 October, 2023.

India FoodBanking Network. 2023. IFBN in action. https://www.indiafoodbanking.org/about/

network. Accessed online: 28 August, 2023.

35
Kaviyani-Charati, M., M. Ameli, F. H. Souraki, A. Jabbarzadeh. 2022. Sustainable network design
for a non-profit food bank supply chain with a heterogeneous fleet under uncertainty. Computers
& Industrial Engineering 171 108442.

Mahmoudi, M., K. Shirzad, V. Verter. 2022. Decision support models for managing food aid supply
chains: A systematic literature review. Socio-Economic Planning Sciences 82 101255.

Martins, C. L., M. T. Melo, M. V. Pato. 2019. Redesigning a food bank supply chain network in a
triple bottom line context. International Journal of Production Economics 214 234–247.

Nair, D. J., D. Rey, V. Dixit. 2017. Fair allocation and cost-effective routing models for food rescue
and redistribution. IISE Transactions 49(12) 1172–1188.

National Institute of Nutrition. 2011. Dietary Guidelines for Indians - A Manual. https://www.
nin.res.in/downloads/DietaryGuidelinesforNINwebsite.pdf. Accessed online: 28 August,
2023.

Orgut, I. S., J. Ivy, R. Uzsoy, J. R. Wilson. 2016. Modeling for the equitable and effective
distribution of donated food under capacity constraints. IIE Transactions 48(3) 252–266.

Peters, K., S. Silva, R. Gonçalves, M. Kavelj, H. Fleuren, D. den Hertog, O. Ergun, M. Freeman.
2021. The nutritious supply chain: optimizing humanitarian food assistance. INFORMS Journal
on Optimization 3(2) 200–226.

Rancourt, M., J. Cordeau, G. Laporte, B. Watkins. 2015. Tactical network planning for food aid
distribution in kenya. Computers & Operations Research 56 68–83.

Reihaneh, M., A. Ghoniem. 2018. A multi-start optimization-based heuristic for a food bank
distribution problem. Journal of the operational research society 69(5) 691–706.

Reusken, M., F. Cruijssen, H. Fleuren. 2023. A food bank supply chain model: Optimizing
investments to maximize food assistance. International Journal of Production Economics 261
108886.

Rey, D., K. Almi’ani, D. J. Nair. 2018. Exact and heuristic algorithms for finding envy-free
allocations in food rescue pickup and delivery logistics. Transportation Research Part E: Logistics
and Transportation Review 112 19–46.

Sahinyazan, F. G., M. Rancourt, V. Verter. 2021. Food aid modality selection problem. Production
and Operations Management 30(4) 965–983.

Stauffer, J. M., M. Vanajakumari, S. Kumar, T. Mangapora. 2022. Achieving equitable food

security: How can food bank mobile pantries fill this humanitarian need. Production and
Operations Management 31(4) 1802–1821.

Tarasuk, V., J. M. Eakin. 2005. Food assistance through “surplus” food: Insights from an
ethnographic study of food bank work. Agriculture and Human Values 22 177–186.

The Hindu. 2012. India’s first food bank inaugurated in Delhi as part
of new network. https://www.thehindu.com/todays-paper/tp-national/
indias-first-food-bank-inaugurated-in-delhi-as-part-of-new-network/
article3507561.ece. Accessed online: 28 August, 2023.

36
The Trussell Trust. 2023. Our story - How we got to where we are today. https://www.
trusselltrust.org/about/our-story/. Accessed online: 9 October, 2023.

United Nations. 2023a. Goal 12: Ensure sustainable consumption and production patterns. https:
//www.un.org/sustainabledevelopment/sustainable-consumption-production/. Accessed
online: 12 October, 2023.

United Nations. 2023b. Sustainable development goals - Goal 2: Zero Hunger. https://www.un.
org/sustainabledevelopment/hunger/. Accessed online: 9 October, 2023.

U.S. Department of Agriculture. 2023. Food Security and Nutrition Assistance. https://
www.ers.usda.gov/data-products/ag-and-food-statistics-charting-the-essentials/
food-security-and-nutrition-assistance. Accessed online: 9 October, 2023.

37