Seminars / Conferences

Justice systems are high-stakes, complex operations that shape societies and lives through the rule of law.  Limited judicial resources can lead to costly delays, stunted economic development, and failure to deliver justice. Using the Supreme Court of India as an exemplar of such resource-constrained settings, we apply ideas from service operations to study judicial delay. Specifically, we model court dynamics as a case-management queue, in which each case may experience multiple service encounters spread across time, all necessarily involving the same server.

We employ data-driven calibrated simulations to study this analytically intractable queue. We identify key sources of inefficiency that drive congestion. We demonstrate how operational interventions can reduce expected delay by as much as 65%.

We will also discuss our ongoing work on MSOM’s Grand Challenge on OM-informed Justice Design, highlighting open problems where the operations management community is well suited to improve access to justice and the timeliness and fairness of the judiciary.

Ilan Lobel is the Fauvelais Family Professor of Data-Driven Operations at New York University’s Stern School of Business. His work sits at the intersection of artificial intelligence, data science, economics, and operations, with a particular focus on how data, algorithms, and AI shape digital marketplaces. Professor Lobel serves as the director of the Andre Koo Technology and Entrepreneurship MBA Program and as the coordinator of the Stern’s MBA AI specialization. His research has received the 2018 MSOM Young Scholar Prize, the 2021 INFORMS TIMES Award, and the INFORMS Revenue Management and Pricing Section Prize in both 2014 and 2024. He currently serves as area editor of Operations Research. His teaching has been recognized by the 2017 Poets & Quants Top 40 MBA Professors Under 40 list and the 2018 Stern Distinguished Teaching Award.

Ilan’s take will present his paper, Pricing AI as a Service.

We study data portability in a model of dynamic price competition where product quality improves for each customer as personal data accumulates with usage. This can reflect AI-driven personalization based on customer data or customers embedding their own data into the firm’s product through usage. We show that when consumers are forward looking, a firm strictly benefits from offering portability if and only if doing so increases total future surplus per initial customer. We identify conditions ensuring that at least one firm strictly prefers to offer portability, with both firms preferring to do so when product differentiation is sufficiently strong relative to differences in net value. By contrast, when consumers are myopic, there are equilibria in which neither firm offers portability. We identify conditions under which a portability mandate harms or benefits consumers. We extend these results to settings with personalized prices and with imperfect data portability.

LLM-based digital twins are increasingly proposed as substitutes for human participants across social science, market research, and policy analysis. I identify fundamental epistemic and structural challenges to using them as replacements for human evidence, and even to the weaker claim that they can help explore human behavior in data-thin regions involving new people or tasks. After a brief tour of the LLM research frontier and its peculiarities, I connect these developments to longstanding arguments in the philosophy of AI: LLMs learn from vast, noisy, and epistemically contaminated data and lack the embodied experience and survival-driven objectives that shape human cognition and behavior. Human behavior itself is heterogeneous, non-objective, changing, and often unstable—quite different from the systems traditional digital twins are designed to model. I organize these issues into a framework for understanding the resulting failure modes, theoretical limits, and empirical dilemmas of LLMs-as-human-surrogates, with particular attention to the validation problem and the conditions and limits of transportability across populations, tasks, and contexts. I conclude with two new empirical studies that add to the critical examination of LLMs-as-human-surrogates.

Motivated by advances in AI agents, we study the impact of improved search technology on learning and welfare in agentic markets. Consumers engage in costly search to acquire signals of product fit prior to purchase. The market aggregates observations — indications of fit for searched products and indications of quality for chosen products — thereby guiding future searches. We characterize the long-run steady-state of the resulting dynamics and study the impact of improving search technology.  We find cheaper search improves learning and consumer surplus, whereas more informativeness can degrade both {\em unless} the market learns as much as consumers about the products, e.g., by “reading the transcript” of agentic conversations.  Finally, we consider the impact of search on how businesses set prices.  In symmetric markets, market efficiency improves when search is cheaper and/or more informative. However, more informative search may decrease consumer surplus as it weakens competition.

In classical mechanism design, buyers are assumed to know their private valuations, while the seller, who does not observe these valuations, designs allocations and payments to maximize revenue. In many modern online platform environments, however, these informational assumptions no longer hold. Buyers may have only partial information about their own valuations, or the seller may be able to acquire information about buyers’ valuations. In this talk, I present a few papers that incorporate these alternative informational structures and characterize the optimal selling mechanisms under each. 

First, I study a setting in which buyers do not know their valuations and must incur an inspection cost to learn them. I characterize the optimal selling mechanisms under both observable and unobservable inspection and show how the Magician’s Problem can be used to obtain a 1 – 1/\sqrt{k+3}-approximation when selling k items to multiple buyers.

Next, I turn to inspection on the seller’s side. I show how a seller’s ability to condition payments on partial information about buyers’ valuations can fundamentally alter not only the structure of the optimal mechanism but also the analytical tools required to derive it. In particular, I demonstrate how methods from the calculus of variations, combined with convex analysis, can be used to characterize the optimal selling mechanism.

We examine how artificial intelligence impacts three core pillars of collaboration—performance enhancement, expertise integration, and social engagement—through a pre-registered field experiment with 791 professionals at Procter & Gamble, a global consumer packaged goods company. Working on real product innovation challenges, professionals were randomly assigned to work either with or without AI, and either individually or with another professional in new product development teams. Our findings reveal that (1) AI significantly enhances performance: individuals with AI matched the performance of teams without AI, demonstrating that AI can effectively replicate certain benefits of human collaboration; (2) AI helps bridge functional silos: without AI, R&D professionals tended to suggest more technical solutions, while Commercial professionals leaned toward commercially oriented proposals, whereas professionals using AI produced more balanced solutions regardless of background; and (3) AI’s language-based interface prompted more positive self-reported emotional responses, suggesting it can fulfill part of the social and motivational role traditionally offered by human teammates. Finally, decomposing the innovation process reveals that AI primarily enhances the quality of generated ideas—elevating the entire distribution of creative output—while human judgment retains value in evaluative selection. This finding highlights the multiple and complementary roles that human and AI partners can play in new product development and creative problem-solving tasks. More generally, our results suggest that AI adoption in knowledge work affects not only performance but also how expertise and sociality manifest within teams, compelling organizations to rethink the structure of collaborative work.

As operations increasingly rely upon flexible labor models – such as gig, part-time, and remote work – it has become commonplace for individuals to work multiple jobs. Across two descriptive studies, relying on a combination of transaction-level data from 90,232 customers of a nationwide retail bank and primary survey data, we study whether people with multiple jobs live their off-the-clock lives materially differently from equivalently compensated people who rely on a single job. We find that people who rely on multiple jobs spend 18.1 percentage points (p.p.) less of their labor income overall, which is driven by a 16.5 p.p. decrease in the share of income spent on necessities and a 2.0 p.p. decrease in the share of income spent on indulgences. They spend meaningfully more on education and transportation, but notably less in all other key spending areas, including categories traditionally associated with enhanced physical wellbeing, such as healthcare and food, and experiential categories often associated with mental wellbeing, such as travel and entertainment. These patterns converge with responses from the General Social Survey, in which equivalently-compensated individuals who rely on multiple jobs report lower financial satisfaction and lower family satisfaction than their single-income counterparts. Together, the findings describe systematic variation in off-the-clock outcomes across different working structures and surface considerations for job design that may promote greater sustainability for employees.

Recent scholarly work has extensively examined the phenomenon of algorithmic collusion driven by AI-enabled pricing algorithms. However, online platforms commonly deploy recommender systems that influence how consumers discover and purchase products, thereby shaping the reward structures faced by pricing algorithms and ultimately affecting competition dynamics and equilibrium outcomes. To address this gap in the literature and elucidate the role of recommender systems, we propose a novel repeated game framework that integrates several key components. We first develop a structural search model to characterize consumers’ decision-making processes in response to varying recommendation sets. This model incorporates both observable and unobservable heterogeneity in utility and search cost functions, and is estimated using real-world data. Building on the resulting consumer model, we formulate personalized recommendation algorithms designed to maximize either platform revenue or consumer utility. We further introduce pricing algorithms for sellers and integrate all these elements to facilitate comprehensive numerical experiments. Our experimental findings reveal that a revenue-maximizing recommender system intensifies algorithmic collusion, whereas a utility-maximizing recommender system encourages more competitive pricing behavior among sellers. Intriguingly, and contrary to conventional insights from the industrial organization and choice modeling literature, increasing the size of recommendation sets under a utility-maximizing regime does not consistently enhance consumer utility. This counterintuitive “more is less” effect arises because expanding the recommendation set reduces the recommender system’s ability to effectively influence the pricing strategies of AI-driven algorithms through dynamic rewards. Moreover, the degree of horizontal differentiation moderates this phenomenon in unexpected ways. The “more is less” effect does not arise at low levels of differentiation, but becomes increasingly pronounced as horizontal differentiation increases. This is because, in such contexts, consumers’ decisions are driven more by their inherent preferences than by price considerations. As a result, when the recommendation set is large, higher-priced products experience relatively smaller declines in their rewards and can sustain elevated prices, supported by the persistent demand engendered by horizontal preferences. These findings provide valuable insights for regulators, platform designers, and market participants, and they open up a range of promising avenues for future research.