Andre Calmon

News

• Sep 2026 “From Trees to Treewidth” is published in Management Science and receives the 2026 INFORMS Computing Society Harvey J. Greenberg Research Award.
• May 2026 “The Value of Time and Location Commitment for Decentralized Emergency Medical Services” is published in Manufacturing & Service Operations Management.
• Dec 2025 “When Supply Chains Become Autonomous” appears in Harvard Business Review, with a companion AI Beer Game.

Research

My research interests include sustainable operations management, business model innovation in complex and resource-constrained settings, and AI and analytics in operations. My current projects look at how organizations source and deploy AI and at business models for commercializing deep tech.

For managers, see “When Supply Chains Become Autonomous” (Harvard Business Review, 2025) and its companion AI Beer Game.

Journal articles

• From Trees to Treewidth: Inventory Management in Complex Supply Chain Networks

  P. Blaettchen, A. P. Calmon, G. Hall, and M. Tawarmalani

  Management Science, Articles in Advance, 2026.

  Winner, 2026 INFORMS Computing Society Harvey J. Greenberg Research Award.

  doi ssrn

  abstract

  We propose an exact solution approach to the Guaranteed Service Model (GSM), one of the most widely applied models for optimizing safety stock placement in supply chain networks. Based on linear programming (LP), our approach handles any directed acyclic network and any cost function that depends on a stage’s incoming and outgoing service times. It scales polynomially in the number of nodes n in the network, and exponentially in its treewidth, which quantifies how “tree-like” a network is, and can be much smaller than n. This contrasts with existing approaches, which scale exponentially in n. The proof of exactness relies crucially on showing that the join of transportation-like polytopes remains integral, and it is more broadly applicable to other Operations Management problems. The use of linear programming makes for straightforward implementation, including when incorporating additional operational constraints. It also enables sensitivity analyses and the construction of principled bounds on the GSM’s optimal value. Finally, it allows for the use of standard LP optimization software, resulting in considerable gains in solving time.

• The Value of Time and Location Commitment for Decentralized Emergency Medical Services

  P. van den Berg, A. P. Calmon, A. K. Gernert, S. Lemmens, M. Rabinovich, and G. Romero

  Manufacturing & Service Operations Management, Articles in Advance, 2026.

  doi

  abstract

  Emergency medical services (EMS) in many low- and middle-income countries utilize decentralized platforms coordinating independent ambulance providers. However, significant operational challenges arise from uncertainty in provider time availability and unpredictable idle locations. These uncertainties hinder reliable service coverage and negatively impact patient outcomes. Using data from our partner Flare in East Africa regarding their operations in Nairobi, Kenya, we investigate the relative effectiveness of enhancing provider temporal commitment (time availability) versus spatial commitment (strategic location) to improve system coverage. We employ optimization models tailored to ambulance commitment uncertainty, a detailed case study, data-driven simulations, and a game-theoretic model. Our findings quantify a stark “cost of decentralization”: the coverage provided by Flare’s 340 loosely committed ambulances could potentially be matched by fewer than 15 optimally deployed fully committed units. We find that enhancing spatial commitment typically yields substantially larger coverage gains than increasing time availability alone, and that both dimensions are strongly complementary.

• Resilience of Agricultural Supply: The Impact of Climate Change on Optimal Seed Production and Allocation

  U. Serhatli, A. P. Calmon, and E. Yücesan

  Asia-Pacific Journal of Operational Research, Article 2650032, 2026.

  doi ssrn

  abstract

  Agricultural supply chains have been experiencing an increasing number of disruptions due to natural disasters, extreme weather events, geopolitical disturbances, and animal or plant diseases, amplifying concerns about the resilience of food supply. This paper focuses on the supply of commercial seeds, the key input into agricultural supply chains. We examine how supply chain disruptions affect the optimal production decisions of seed manufacturers, the expected revenue of farmers that are part of the production process, and the resilience of food supply. We model the seed production process as a discrete-time stochastic dynamic optimization problem where, in each period, the seed manufacturer solves an optimization problem with two stages: the planting stage, where the manufacturer chooses the quantity of hybrid seeds to produce through a network of independent farmers, and the allocation stage, where the manufacturer allocates the resulting yield to different markets with varying profit margins. We then examine how changes in the likelihood of a disruptive event affect the supply chain’s performance. We prove that, as the probability of disruption increases, (a) the manufacturer’s expected profits decrease while the optimal planting quantity grows; (b) the expected total yield decreases, reducing the expected profit of the farmers who are under contract with the manufacturer; (c) the expected allocation quantity to different markets decreases with a more significant drop in smallholder markets. Finally, we present a simulation model calibrated to industry data where we find that even a small increase in the likelihood of supply volatility can dramatically change the optimal production decisions and have significantly negative effects on the profitability of commercial seeds, on the availability of seeds in smallholder markets, on the profits earned by contracted farmers, and on the value of typical operational improvements such as delayed differentiation.

• Traceability Technology Adoption in Supply Chain Networks

  P. Blaettchen, A. P. Calmon, and G. Hall

  Management Science 71(1): 83–102, 2025.

  doi ssrn

  abstract

  Modern traceability technologies promise to improve supply chain management by simplifying recalls, increasing visibility, and verifying sustainable supplier practices. Initiatives leading the implementation of traceability technologies must choose the least-costly set of firms—or seed set—to target for early adoption. Choosing this seed set is challenging because firms are part of supply chains interlinked in complex networks, yielding an inherent supply chain effect: benefits obtained from traceability are conditional on technology adoption by a subset of firms in a product’s supply chain. We prove that the problem of selecting the least-costly seed set in a supply chain network is hard to solve and even approximate within a polylogarithmic factor. Nevertheless, we provide a novel linear programming-based algorithm to identify the least-costly seed set. The algorithm is fixed-parameter tractable in the supply chain network’s treewidth, which we show to be low in real-world supply chain networks. The algorithm also enables us to derive easily computable bounds on the cost of selecting an optimal seed set. We leverage our toolbox to conduct large-scale numerical experiments that provide insights into how the supply chain network structure influences diffusion.

• Business Model Innovation for Ambulance Systems in Low- and Middle-Income Countries: “Coordination and Competition”

  A. K. Gernert, A. P. Calmon, G. Romero, and L. N. Van Wassenhove

  Production and Operations Management, New Business Models Special Issue, 1–17, 2024.

  doi ssrn

  abstract

  Several low- and middle-income countries’ emergency transportation systems (ETSs) do not have a centralized emergency number. Instead, they have many independent ambulance providers, each with a small number of ambulances. As a result, ETSs in these contexts lack coordination and ambulances. Using a free-entry equilibrium model, we show that in such decentralized systems, the probability that any given call can be served by at least one ambulance, that is, its coverage, is at most 71.54%, regardless of the ETS’s profitability. We examine three business models that can address the ETS’s lack of coordination and ambulances: (i) a competitor-only business model, where an entrepreneur enters the ETS and acquires ambulances to compete with existing providers; (ii) a platform business model, where an entrepreneur coordinates existing providers; and (iii) an innovative platform-plus business model, where an entrepreneur combines (i) and (ii): setting-up a platform and acquiring platform-owned ambulances. We also examine a government-run platform that takes no commissions from providers. Using a game-theoretic approach, we find that it is optimal for all platform models to incentivize all providers to join. However, only the government-run platform may incentivize providers to acquire additional ambulances.

• Operational Strategies for Distributing Durable Goods in the Base of the Pyramid

  A. P. Calmon, D. Jue-Rajasingh, G. Romero, and J. Stenson

  Manufacturing & Service Operations Management 24(4): 1887–1905, 2022.

  doi ssrn

  abstract

  Novel life-improving products, such as solar lanterns and energy-efficient cookstoves address essential needs of consumers in the base of the pyramid (BoP). However, their profitable distribution is often difficult because BoP customers are risk-averse, their ability to pay (ATP) is lower than their willingness to pay, and they face uncertainty regarding these products’ value. We examine two practical strategies from distributors in the BoP: (1) improving the product’s affordability through a discount and (2) increasing awareness of the product’s value. We introduce a supply chain model for the BoP and analyze the distributor’s pricing problem with refunds as well as the distributor’s optimal budget allocation between strategies (1) and (2). We find that, in the BoP, the distributor’s profit-maximizing budget allocation often yields the lowest consumer surplus. This misalignment between profits and consumer surplus disappears if customers’ ATP is high. Moreover, the misalignment can be resolved if the distributor offers free product returns and commits to a maximum retail price.

• Warranty Matching in a Consumer Electronics Closed-Loop Supply Chain

  A. P. Calmon, S. C. Graves, and S. Lemmens

  Manufacturing & Service Operations Management 23(5): 1314–1331, 2021.

  doi ssrn

  abstract

  We examine a dynamic assignment problem faced by a large wireless service provider (WSP) that is a Fortune 100 company. This company manages two warranties: (i) a customer warranty that the WSP offers its customers and (ii) an original equipment manufacturer (OEM) warranty that OEMs offer the WSP. The WSP uses devices refurbished by the OEM as replacement devices, and hence their warranty operation is a closed-loop supply chain. Depending on the assignment the WSP uses, the customer and OEM warranties might become misaligned for customer-device pairs, potentially incurring a cost for the WSP. We identify, model, and analyze a new dynamic assignment problem that emerges in this setting called the warranty matching problem. We introduce a new class of policies, called farsighted policies, which can perform better than myopic policies. We also propose a new heuristic assignment policy, the sampling policy, which leads to a near-optimal assignment. We show that our assignment policies reduce the average uncovered time and the expected number of out-of-OEM-warranty returns by more than 75% in comparison with our industrial partner’s current assignment policy.

• Revenue Management with Repeated Customer Interactions

  A. P. Calmon, F. D. Ciocan, and G. Romero

  Management Science 67(5): 2944–2963, 2021.

  doi ssrn

  abstract

  Motivated by online advertising, we model and analyze a revenue management problem where a platform interacts with a set of customers over a number of periods. Unlike traditional network revenue management, which treats the interaction between platform and customers as one-shot, we consider stateful customers who can dynamically change their goodwill toward the platform depending on the quality of their past interactions. Customer goodwill further determines the amount of budget that they allocate to the platform in the future. These dynamics create a trade-off between the platform myopically maximizing short-term revenues, versus maximizing the long-term goodwill of its customers to collect higher future revenues. We identify a set of natural conditions under which myopic policies that ignore the budget dynamics are either optimal or admit parametric guarantees. We also show that, if these conditions do not hold, myopic and finite look-ahead policies can perform arbitrarily poorly in this repeated setting.

• Inventory Management in a Consumer Electronics Closed-Loop Supply Chain

  A. P. Calmon and S. C. Graves

  Manufacturing & Service Operations Management 19(4): 568–585, 2017.

  doi ssrn

  abstract

  The goal of this paper is to describe, model, and optimize inventory in a reverse logistics system that supports the warranty returns and replacements for a consumer electronic device. The reverse-logistics system is a closed-loop supply chain: failed devices are returned for repair and refurbishing; this inventory is then used to serve warranty claims or sold through a side-sales channel. Managing inventory in this system is challenging due to the short life-cycle of these devices and the rapidly declining value for the inventory. We examine an inventory model that captures these dynamics, characterize the structure of the optimal policy for stochastic demand, and introduce an algorithm to calculate optimal sell-down levels. We also provide a closed-form policy for the deterministic version of the problem and use this policy as a certainty-equivalent approximation to the stochastic optimal policy.

Working papers

• Child Labor as an Operational Lever: Heterogeneous Effects of Interventions in Cocoa Supply Chains

  A. P. Calmon, A. K. Gernert, D. A. Iancu, and L. A. Manea

  Major revision at Manufacturing & Service Operations Management.

  pdf

  abstract

  More than 1.5 million children engage in child labor in cocoa production in Côte d’Ivoire and Ghana despite decades of interventions. Most of this work occurs within smallholder farm households, where child labor is one of the levers adults use to balance tradeoffs among production, consumption, and finances. These tradeoffs are household-specific, creating a design problem for responsible-sourcing programs: an intervention that reduces child labor in some households may increase it in others. We develop a model of a cocoa household that chooses its consumption, borrowing, saving, and child labor use under harvest uncertainty, and use it to predict how common interventions affect child labor. We then test the predictions using survey data collected by an NGO partner in Ghana. Our model shows that interventions affect child labor through three forces: resource relief, labor-productivity effects, and consumption expansion. Cash transfers and improved savings access operate primarily through resource relief and strictly reduce child labor when the household can borrow or save; otherwise, child labor is unchanged. Improved credit access can either lower or raise child labor depending on whether repayment relief dominates debt-financed consumption. Production support can reduce or increase child labor depending on whether resource relief dominates the labor-productivity and consumption-feedback forces. Under multiplicative production risk, labor-substituting production support weakly reduces child labor for all households, whereas labor-complementary support can either strictly lower or strictly increase child labor. All of these interventions raise household welfare but have nuanced effects on consumption. The survey evidence is consistent with our framework, though descriptive rather than causal: cash transfers show no systematic increase in child labor, while loan repayment ability and production-related benefits are associated with lower child labor among more constrained households and higher child labor among less constrained or higher-capacity households. Responsible-sourcing programs should evaluate and target interventions based on household characteristics—such as income, spending, loan repayment capacity, or production scale—that are already measured in standard surveys and can help predict when a given intervention might decrease child labor or backfire.

• Reliability and Effectiveness of Autonomous AI Agents in Supply Chain Management

  C. X. Long, D. Simchi-Levi, F. Zhu, H. Su, A. P. Calmon, and F. P. Calmon

  Submitted to Management Science.

  arXiv beer game

  abstract

  This paper studies autonomous generative AI agents in multi-echelon supply chains using the MIT Beer Game. We identify four inference-time levers that shape performance: model selection, policies and guardrails, centralized data sharing, and prompt engineering. Model capability is the dominant factor: an out-of-the-box reasoning model exceeds human-level performance, and optimized reasoning models reduce costs by up to 67% relative to human teams. However, strong average performance masks substantial reliability risks. We introduce the agent bullwhip effect, the amplification of decision unreliability across echelons, manifesting along two dimensions: decision variance increases both across facilities at the same point in time and within the same facility across time. We develop a mathematical framework showing that this phenomenon is inherent to multi-agent systems that involve coordination and information delays, and we demonstrate that repeated sampling fails to meaningfully reduce it. To address this limitation, we propose a Group Relative Policy Optimization (GRPO)-based reinforcement-learning post-training framework that trains a shared base LLM using system-level supply-chain rewards. GRPO post-training substantially reduces tail events, curtails agent bullwhip, and improves the reliability of autonomous supply-chain agents.

• A Random Model of Supply Chain Networks

  P. Blaettchen, A. P. Calmon, and G. Hall

  Under review at Management Science.

  ssrn RG4SC

  abstract

  Supply chain problems are frequently formulated as optimization problems over graphs representing complex networks of interlinked input-output relationships. Frequently, these problems are hard, so researchers rely on analyzing stylized structures or developing heuristic solutions. Yet, a scarcity of real-world data has hindered our understanding of how these exact and heuristic solutions perform in practice and whether managerial insights carry over from these simplified settings. We address this critical gap by introducing RG4SC, a versatile random graph model for creating “test tracks” for supply chain management research. RG4SC’s simple micro-foundations and interpretable input parameters allow for systematically generating diverse and realistic network structures. We demonstrate its empirical validity and that it more adequately represents real-world supply chain networks than existing random models. We then showcase RG4SC’s utility for research through a case study on the Guaranteed Service Model, a widely used framework for safety stock optimization.

• Data-Driven Failure Time Estimation in a Consumer Electronics Closed-Loop Supply Chain

  A. P. Calmon, S. C. Graves, and S. Lemmens

  Major revision at Production and Operations Management.

  ssrn

  abstract

  We examine and analyze a strategy for forecasting the demand for replacement devices in a large Wireless Service Provider (WSP) that is a Fortune 100 company. The Original Equipment Manufacturer (OEM) refurbishes returned devices that are offered as replacement devices by the WSP to its customers, and hence the device refurbishment and replacement operations are a closed-loop supply chain. We introduce a strategy for estimating failure time distributions of newly launched devices that leverages the historical data of failures from other devices. The fundamental assumption that we make is that the hazard rate distribution of the new devices can be modeled as a mixture of historical hazard rate distributions of prior devices. The proposed strategy uses the empirical hazard rates from other devices to form a basis set of hazard rate distributions. We then use a regression to identify and fit the relevant hazard rates distributions from the basis to the observed failures of the new device.

• Specialize or Balance? Salesforce Allocation Across Differentially Vulnerable Retailers at the Base of the Pyramid

  O. Fatunde-Iloeje, A. Calmon, J. de Zegher, and G. Romero

  Major revision at Manufacturing & Service Operations Management.

  Second Place, 2021 POMS College of Sustainable Operations Best Student Paper Competition.

  ssrn

  abstract

  Distributors reach Base of the Pyramid (BoP) consumers through sales agents who build personal relationships with retailers. Agent turnover disrupts these relationships, and informal retailers, who depend heavily on personal ties, are particularly vulnerable. How should a distributor allocate its salesforce across formal and informal retailers when vulnerability to disruptions differs between them? The salesforce management literature recommends specialization, but we show that specialization concentrates vulnerable retailers with agents who earn lower commissions and face elevated turnover risk. We develop an analytical model showing that specialization is optimal when both retailer types face proportional revenue losses after a disruption. When informal retailers are disproportionately vulnerable, however, the optimal allocation shifts toward balance, with each agent serving a mix of formal and informal retailers. Using data from Essmart, a durable goods distributor serving over 2,000 retailers in southern India (2016–2024), we estimate that informal retailers experience incremental revenue loss on the order of one-third of their pre-treatment Essmart-derived revenue, confirming that differential vulnerability is the empirically relevant case.

• Data Analytics for Creative Processes: Designing the Next Delicious Product

  A. P. Calmon, F. P. Calmon, R. Goodwin, and J. Zazo

  Major revision at Manufacturing & Service Operations Management.

  ssrn

  abstract

  Over the past decades, firms that develop new flavor and fragrance products (a $30 billion per year market) have amassed large amounts of data from their own internal creation processes. Yet, new product creation still mostly relies on human expertise and arduous experimentation. We present a data analytics framework to aid in the flavor and fragrance creation process. The framework consists of two steps. In the first step, we use data to fit a metric over products, a procedure we refer to as curation. The metric of choice is the Earth Mover’s Distance (EMD), where distances between products are quantified in terms of the pairwise similarity between ingredients. The curation process is formulated as a metric learning problem, where the cost matrix used in EMD computations is directly fit from labeled data. In the second step, we develop analytical tools to aid in the design of new products by sampling and evaluating new products within an EMD neighborhood of a seed product, a procedure we refer to as creation.

• Pricing and Job Allocation in Online Labor Platforms

  V. F. Araman, A. Calmon, and K. Fridgeirsdottir

  Working paper.

  ssrn

  abstract

  We model, analyze, and optimize the operations of an online labor platform that matches jobs to workers. The arrival of jobs and workers to the platform is stochastic and the job processing time is random. The platform chooses the fees per job and assigns jobs to workers with the goal of (i) maximizing platform revenues, (ii) minimizing the unpredictability in workers’ profits, and (iii) minimizing any delay in processing the incoming jobs. Workers are sensitive to the revenue they make in the platform and, therefore, the worker arrival rate depends on the platform’s pricing and job allocation strategy. We introduce a policy class called Uniform Allocation (UA) and provide an analytical characterization of the platform’s behavior and performance under this policy class. Then, we design a UA policy that simultaneously optimizes objectives (i), (ii), and (iii) as the system scales.

Practitioner articles and book chapters

• When Supply Chains Become Autonomous

  C. Long, D. Simchi-Levi, A. P. Calmon, and F. P. Calmon

  Harvard Business Review, December 2025.

  article beer game

Teaching

Executive education

My executive teaching covers innovation, supply chains, operations strategy, AI and analytics, and sustainability. Since 2025 I have been Faculty Co-Director of the Coca-Cola Supply Chain Leader Excellence (SCLX) Program.

Courses

• Tech2Market Capstone (MBA and Executive MBA, Georgia Tech)
• Business Sustainability (MBA and executive programs, Georgia Tech and INSEAD)
  Page Prize for Excellence in Sustainable Business Education, Grand Prize, 2018
• Operations Management (undergraduate and MBA core, Georgia Tech and INSEAD)
  Best MBA Core Course Teacher Award, INSEAD, 2015, 2017, 2019; Student Recognition of Excellence in Teaching (CIOS Honor Roll), Georgia Tech, 2021, 2023, 2024
• Business Lab / New Business Models (undergraduate, Georgia Tech; MBA, INSEAD), a Create-X gateway course run as a tournament on Torneio

Published cases

Teaching tools

• Torneio, a web platform for running innovation tournaments in the classroom (with S. Chick and R. Ballestiero)
• AI Beer Game, an interactive simulation of AI agents making supply chain decisions; companion to the Harvard Business Review article
• Better Place Negotiation Game (PDF), an in-class role-play on introducing and scaling green energy business models

Leadership and service

• Faculty Director, Ray C. Anderson Center for Sustainable Business, Georgia Tech, 2025–present. The Center works with over 70 companies.
• Faculty Co-Director, Coca-Cola Supply Chain Leader Excellence (SCLX) Program, 2025–present.
• Co-creator and Director, Sustain-X (Georgia Tech’s program for sustainability-focused founders, now part of the Anderson Center), 2022–2025.
• Associate Editor, Manufacturing & Service Operations Management. MSOM Meritorious Service Award, INFORMS, 2022.

Advising

Ph.D. students

• Donghyun (Daniel) Choi, Georgia Tech, expected 2027 (co-advised with B. Toktay)
• Philippe Blaettchen, now at Singapore Management University (co-advised with S. Hasija); 2022 INFORMS TIMES Best Dissertation Award
• Utku Serhatli, now at Nova SBE (co-advised with E. Yücesan)

Postdoctoral researchers

• Andreas Gernert, now at Kühne Logistics University
• Stef Lemmens, now at Rotterdam School of Management, Erasmus University

Contact

I welcome inquiries about executive education, speaking, and research collaborations.