Technological advances enable new business models for heavy equipment manufacturers wherein customers access equipment without ownership. We seek to understand the profitability and environmental performance of different emerging business models in light of salient economic and operational factors. We develop a game-theoretic model to identify the optimal choice between a traditional ownership-based business model and two access-based models: servicization and peer-to-peer sharing. After-sales services, equipment characteristics, usage environments, and fuel prices affect this choice. We also provide a novel framework to analyze business models' environmental impact, which incorporates trade-offs between economic value and environmental costs and shows that all models may create win-win situations for the manufacturer and the environment.

With Niyazi Taneri (Cambridge University) & Sameer Hasija (INSEAD).
Published in Operations Research, 2024, Volume 72, Issue 6.

Family planning services play a central role in the sustainable development agenda by improving health and education while reducing poverty and gender inequality. Marie Stopes International (MSI) reaches underserved clients in rural areas with the help of mobile healthcare units (MHUs). This approach is gaining traction more generally to enable broad access to health services for marginalized rural populations, thus stepping closer to fulfilling SDG 3: Good Health and Well-Being. Our research addresses the complex challenge of allocating limited mobile healthcare unit resources to create the largest possible impact in the long-term, especially in light of intricate demand dynamics.
We model the MHU resource allocation problem as the optimization of a sum of sigmoidal functions and derive new theoretical results with operationalizable managerial insights. We further develop a machine learning approach to predict demand in such a setting that provides interpretable results, a key requirement in the humanitarian context. Our empirical application combines MSI's client demand data with rich, publicly available data sources. We use the resulting estimates to demonstrate the benefits of our analytically derived insights.

With Andres Alban (Frankfurt School of Finance & Management), Harwin de Vries (Erasmus University), & Luk Van Wassenhove (INSEAD).
Published in Manufacturing & Service Operations Management, 2022, Volume 24, Issue 6.
Finalist, 2022 MSOM Society Award for Responsible Research in Operations Management.

We propose a new linear programming (LP) framework for the Guaranteed Service Model (GSM), one of the most widely applied approaches for optimizing safety stock placement in supply chain networks. Unlike existing approaches, our framework optimizes the GSM on any acyclic supply chain network of bounded treewidth — a graph-theoretic measure quantifying how "tree-like" a network is. The framework uses an exact LP reformulation of the GSM with size exponential in the network's treewidth but polynomial in other network parameters, such as the number of nodes. This makes it tractable for real-world supply chains, which typically exhibit low treewidth, and allows for the use of standard LP optimization software, unlike previous GSM optimization strategies, which rely on specialized algorithms or heuristics. It also enables new results and insights for the GSM, such as duality-based sensitivity analysis, principled upper and lower bounds, and the ability to incorporate operational constraints. We extend this framework to handle closed-loop supply chains with reverse flows, demonstrating that individual reverse flows only marginally increase computational complexity. This closed-loop GSM facilitates the analysis of how reverse flows impact safety stock placement in supply chains that involve recycling, remanufacturing, and product returns. Overall, our framework provides new theoretical foundations and practical tools for managing safety stock in complex modern supply chain networks.

With Andre Calmon (Georgia Tech), Georgina Hall (INSEAD), & Mohit Tawarmalani (Daniels School of Business).
Under review. A draft is available here.

Firms and policy makers face many resource allocation decisions. Often, these decisions cannot be made purely based on utilitarian considerations but capture some notion of fairness. Operations Management has provided a deep understanding of the trade-off between fairness and efficiency but has often ignored the uncertainty inherent in evaluating resource allocations. One fundamental reason for this uncertainty stems from imperfect information gathering. For example, in a push to achieve net-zero emissions, governments increasingly impose restrictions, such as congestion charges, without accurately understanding the effects of such restrictions on different sub-populations.
Building on a distributionally robust optimization framework, we study how the "price of fairness" is affected by uncertainty in the impact of different policies. We demonstrate that ignoring uncertainty can systematically bias decisions away from disadvantaged sub-populations. We then analyze how information-gathering activities can mitigate this bias and how these activities should optimally be targeted.

With Chengxin Yan (Nankai University) & Peter Zhang (Carnegie Mellon University).
Analysis in progress.

Everyday business tasks require sequential, interlinked decisions that make it difficult for workers to identify the best strategy. As a result, Artificial Intelligence (AI)-based recommendation systems are applied increasingly to reduce complexity. However, due to human biases, there remains a gap between what is technically possible and what is utilized. This work seeks to understand how recommendation systems can help induce efficient behavior in sequential decision-making and contribute to long-term learning.
We develop a sequential decision-making task in the form of a virtual electric vehicle driving game. Participants, equipped with algorithm-generated recommendations, need to make sequential charging decisions while facing uncertain traffic. With the help of inverse reinforcement learning, we identiy participants' underlying decision-making strategies and how these are affected by recommendations, which we differentiate based on their information content and precision. Our experimental results offer key insights into how humans make sequential decisions and respond to different types of recommendations, both immediately and in terms of long-term learning.

With Park Sinchaisri (UC, Berkeley).
Writing in progress.