About

Sébastien Martin is an Associate Professor of Operations at Kellogg School of Management. He received his Ph.D. in operations research from MIT and an M.Sc. in applied mathematics from Ecole Polytechnique. His research focuses on the interface between humans and algorithms in public-sector operations and online platforms, with an emphasis on prescriptive analytics and real-world impact. He has designed Lyft’s dispatch algorithm, which increased drivers’ yearly revenue by tens of millions, and optimized school transportation systems in Boston and San Francisco, enabling significant cost savings and increased fairness. Before joining Kellogg, he was a postdoctoral research fellow at Lyft. At Kellogg, he teaches courses such as AI Foundations for Managers and explores the use of generative AI in business and education, including creating Kellogg’s AI Teaching Assistant, Kai. His work has earned him multiple awards, including the Edelman Award twice, and he serves on the Board of Directors of ESAB Corporation.

Research topics

• Computer Science
• Engineering
• Operations research
• Economics
• Business
• Operations management
• Transport engineering
• Computer network
• Parallel computing

Selected publications

• Algorithmic Precision and Human Decision: A Study of Interactive Optimization for School Schedules

  Management Science · 2025-11-10

  articleSenior author

  In collaboration with the San Francisco Unified School District (SFUSD), this paper introduces an interactive optimization framework to tackle complex school scheduling challenges. The choice of school start and end times is an optimization challenge, as schedules influence the district’s transportation system, and limiting the associated costs is a computationally difficult combinatorial problem. However, it is also a policy challenge, as transportation costs are far from the only consequence of school schedule changes. Policymakers need time and knowledge to balance these considerations and reach a consensus carefully; past implementations have failed because of policy issues, despite state-of-the-art optimization approaches. We first motivate our approach with a microfoundation model of the interplay between policymakers and researchers, arguing that limiting their dependency is key. Building on these insights, we propose a framework that includes (1) a fast algorithm capable of solving the school schedule problem that compares favorably to the literature and (2) an interactive optimization approach that leverages this speed to allow policymakers to explore a variety of solutions in a transparent and efficient way, facilitating the policy decision-making process. The framework led to the first optimization-driven school start time changes in the United States, updating the schedule of all 133 schools in SFUSD in 2021, with annual transportation savings exceeding $5 million. A comprehensive survey of approximately 27,000 parents and staff in 2022 provides evidence of the approach’s effectiveness. This paper was accepted by Felipe Caro, Special Issue on the Human-Algorithm Connection. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.05834 .

  PublisherDOI

• Annuaire de droit de l'Union européenne

  Éditions Panthéon-Assas eBooks · 2025-10-30

  book-chapter
  PublisherDOI

• Relative Monte Carlo for Reinforcement Learning

  2025-07-02

  articleOpen access

  We propose and analyze a new policy gradient algorithm for reinforcement learning (RL), relative Monte Carlo (rMC). The method estimates policy gradients using relative returns between a root sample path and counterfactual simulated paths, instantiated by taking a different action from the root. The resulting gradient estimate is both unbiased and has low variance. rMC is compatible with any differentiable policy, including neural networks, and is guaranteed to converge even for infinite horizon tasks. The method utilizes common random number coupling of the simulated paths to reduce variance and increase the likelihood that paths merge, thereby reducing simulation complexity. It is particularly well suited to discrete event control problems where actions have a "local" effect, such as queueing, supply chain, or ride-hailing problems. Indeed, we show that it has provably low complexity for a family of inventory control problems. Numerical tests on a challenging inventory and fulfillment problem show that compared to traditional RL approaches, rMC converges in far fewer iterations (lower variance), has better policy performance (unbiased), and requires minimal hyperparameter tuning.

  PublisherOA PDFDOI

• A Better Match for Drivers and Riders: Reinforcement Learning at Lyft

  INFORMS Journal on Applied Analytics · 2024-01-01 · 13 citations

  article

  We used reinforcement learning to improve how Lyft matches drivers and riders. The change was implemented globally and led to more than $30 million per year in incremental driver revenue.

  PublisherDOI

• Frontiers in Operations: Employees vs. Contractors: An Operational Perspective

  Manufacturing & Service Operations Management · 2024-05-17 · 20 citations

  article

  Problem definition: We consider a platform’s problem of how to staff its operations given the possibilities of hiring employees and setting up a contractor marketplace. We aim to understand the operational difference between these two work arrangement models. Methodology/results: We consider a model where demand is not only stochastic but also evolving over time, which we capture via a state of the world that determines the demand distribution. In the case of employees, the platform controls the number of employee hours it uses for serving demand, whereas in the case of contractors, it sets the wage paid to them per utilized hour. We show that although the employee problem is equivalent to a standard newsvendor, the contractor one corresponds to an unusual version of the newsvendor model where utilization is the control variable. Managerial implications: This distinction makes the contractor model more flexible, allowing us to prove that it performs significantly better, especially if the order of magnitude of demand is unknown. Meanwhile, hybrid solutions that combine both employees and contractors have complex optimal solutions and offer relatively limited benefits relative to a contractor marketplace. History: This paper has been accepted in the Manufacturing & Service Operations Management Frontiers in Operations Initiative. Funding: This research was partially supported by the National Natural Science Foundation of China [Grant 71821002]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2023.0029 .

  PublisherDOI

• Relative Monte Carlo for Reinforcement Learning

  SSRN Electronic Journal · 2024-01-01

  preprintOpen access
  PublisherDOI

• Detours in Shared Rides

  Management Science · 2024-05-17 · 12 citations

  articleSenior author

  Detours are considered key for the efficient operation of a shared rides service, but they are also a major pain point for consumers of such services. This paper studies the relationship between the value generated by shared rides and the detours they create for riders. We establish a limit on the sum of value and detour, and we prove that this leads to a tight bound on the Pareto frontier of values and detours in a general setting with an arbitrary number of requests. We explicitly compute the Pareto frontier for one family of city topologies and construct it via simulation for several more networks, including one based on ride-sharing data from commute hours in Manhattan. We find that average detours are usually small, even in low-demand-density settings. We also find that by carefully choosing the match objective, detours can be reduced with a relatively small impact on values and that the density of ride requests is far more important than detours for the effective operations of a shared rides service. In response, we propose that platforms implement a two-product version of shared rides and limit the worst-case detours of its users. This paper was accepted by Hamid Nazerzadeh, data science. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2020.03125 .

  PublisherDOI

• Trading Flexibility for Adoption: From Dynamic to Static Walking in Ride-Sharing

  Management Science · 2024-10-17 · 6 citations

  article

  On-demand ride-sharing aims to fulfill riders’ transportation needs whenever and wherever they want. Although this service level appeals to riders, overall system efficiency can improve substantially if riders are willing to be flexible. Here, we explore riders’ flexibility in space via walking to more accessible pickup locations. Ride-sharing platforms have traditionally implemented dynamic walking to optimize rider pickup locations and rider-driver assignment jointly. We propose an alternative that we call static walking, which presents a predetermined pickup location to the rider before optimizing rider-driver assignment. Although dynamic walking enables more efficient matching of riders and drivers, we hypothesize that riders prefer static walking because of the certainty of the pickup location before booking the ride. Using simulations on Lyft data, we show that static walking can capture up to 96% of the value of dynamic walking in congested urban networks at a fixed adoption rate. Furthermore, experimentation on Lyft’s user interface suggests that providing riders with information on pickup location before an opt-in decision can increase walking adoption—to the extent that static walking may outperform dynamic walking overall. More broadly, this study highlights the importance of carefully designing flexibility mechanisms on platforms: a little flexibility goes a long way, especially when flexibility presents a barrier to adoption. This paper was accepted by J. George Shanthikumar, data science. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.03201 .

  PublisherDOI

• Human-AI Interactions and Societal Pitfalls

  2024-07-08 · 5 citations

  articleSenior author

  When working with generative artificial intelligence (AI), users may see productivity gains, but content generated with the help of AI may not match their preferences exactly. The boost in productivity may come at the expense of users' idiosyncrasies, such as personal style and tastes, preferences we would naturally express without AI. To let users express their preferences, many AI systems let users edit their prompt (e.g., Midjourney) or allow more natural interactions (e.g., ChatGPT), and users can always review and edit the AI-generated output themselves. However, aligning a user's intentions with an AI's output can take time and may not always be worth it if the AI's first or default output "does the job." In short, users face a trade-off between AI output fidelity and communication cost. The purpose of this work is to examine the impact of this human-AI interaction on the AI-generated content we produce as a society.

  PublisherDOI

• Algorithmic Precision and Human Decision: A Study of Interactive Optimization for School Schedules

  2024-07-08

  articleSenior author

  In collaboration with the San Francisco Unified School District (SFUSD), this paper introduces an interactive optimization framework to tackle complex school scheduling challenges. The choice of school start and end times is an optimization challenge, as schedules influence the district's transportation system, and limiting the associated costs is a computationally difficult combinatorial problem. However, it is also a policy challenge, as transportation costs are far from the only consequence of school schedule changes. Policymakers need time and knowledge to balance these considerations and reach a consensus carefully; past implementations have failed because of policy issues despite state-of-the-art optimization approaches.

  PublisherDOI

Frequent coauthors

• Hubert Delzangles

  68 shared

• Arthur Delarue

  8 shared

• Garrett van Ryzin

  Amazon (United States)

  6 shared

• Dimitris Bertsimas

  6 shared

• Loïc Grard

  5 shared

• Francisco Castro

  Anderson University - South Carolina

  4 shared

• Patrick Jaillet

  Massachusetts Institute of Technology

  4 shared

• Ilan Lobel

  New York University

  4 shared

Awards & honors

• Edelman Award Laureate (2019 and 2023)
• George Dantzig and TLS Society dissertation awards
• Daniel H. Wagner Prize Finalist
• INFORMS Distinguished Service Award, Operations Research, IN…
• Poets and Quants Best 40 Under 40 MBA Professors