Date: June 9 – 10 , 2026
Venue: CPD-LG.18, Centennial Campus, The University of Hong Kong
Participation: by invitation ONLY
Saif Benjaafar
University of Michigan
Over-Crowded and Under-Matched: Matching Queues with Strategic Agents
Abstract ▶
We study a dynamic matching environment in which agents strategically decide whether to join a queue and, conditional on joining, whether to accept a match or wait for a higher-quality opportunity. We show that equilibrium behavior can simultaneously exhibit excessive entry and insufficient matching relative to the socially optimal allocation. These two distortions operate on distinct margins—admission and selectivity—and reinforce one another, leading to overcrowded queues and reduced throughput. We characterize equilibrium and socially optimal policies in a tractable continuous-time model, establish a sharp comparison between them, and show that decentralization of the social optimum requires pricing both entry and match acceptance.
Bio ▶
Saif Benjaafar is the Seth Bonder Collegiate Professor of Industrial and Operations Engineering and Professor of Electrical Engineering and Computer Science at the University of Michigan, where he also serves as the Goff Smith Co-Director of the Tauber Institute for Global Operations, a joint institute of the Ross School of Business and the College of Engineering. Prior to joining the University of Michigan, he was McKnight Presidential Endowed Professor and Distinguished McKnight University Professor at the University of Minnesota. He is a founding member of the Singapore University of Technology and Design and served as Pillar Head of Engineering Systems and Design. He is a Fellow of INFORMS, MSOM, and IISE and a former Editor-in-Chief of the INFORMS journal Service Science.
Jeannette Song
Duke University
Innovator’s Edge in Supply Chain Transparency and Food Waste Reduction
Abstract ▶
New digital technologies allow retailers to track product freshness in real time and, in some cases, disclose that information directly to consumers. In the fresh produce retail industry, these tools can improve inventory decisions, reduce food waste, and influence consumer demand. Yet adoption is costly, and early movers often face higher implementation costs than later adopters. This raises a central question: when does moving first actually pay off? To address this question, we develop a dynamic model of competing retailers facing uncertain supply freshness and heterogeneous consumers. In our setting, transparency creates value both operationally, by improving inventory decisions, and strategically, by reducing consumers’ information frictions. We show that the value of transparency depends not only on the technology itself but also on how retailers deploy it, including whether they use it solely for internal decision-making, disclose it to consumers, or combine it with pricing actions. Our analysis yields new insights into when early adoption is advantageous, how transparency reshapes competition, and how digital adoption affects both profitability and food waste. (Joint work with Bora Keskin and Chenghuai Li.)
Bio ▶
Jeannette Song is the R. David Thomas Professor of Business Administration and a Professor of Operations Management at the Fuqua School of Business of Duke University. She studies supply chain management and operations strategy. Her recent research covers supply chain digitization, data-driven operational decision-making, resilient supply chain strategies, and responsible operations. She has published extensively in leading academic journals and has edited the “Research Handbook on Inventory Management” (Edward Elgar Publishing, 2023). Professor Song is an INFORMS Fellow, an MSOM Fellow, and a former President of MSOM; she is also a Department Editor for Management Science.
Day 1 (June 9)
9:30 – 9:40
Opening Remarks
Prof. Zhixi Wan
9:40 – 10:20
Hongyu Chen
Massachusetts Institute of Technology
Using LLMs for Service Systems Evaluation —— A Weak Shadow Variable Perspective
Abstract ▶
Estimating average customer ratings from voluntary feedback is a central problem in platform operations, but feedback is often missing not at random (MNAR), so standard estimators are biased and the target quantity is not identified without strong assumptions. Large language models (LLMs) create a new opportunity in these settings by transforming rich unstructured records, such as customer-service dialogues, into scalable predictions of otherwise missing outcomes. We develop a partial identification framework that uses these predictions as weak shadow variables and computes sharp bounds on the estimand through a pair of linear programs, without requiring the accuracy of the LLM prediction. For finite samples, we propose a set-expansion estimator and a bootstrap procedure that deliver valid inference for the identified set. In semi-synthetic experiments based on real customer-service dialogue data from JD.com, LLM-based weak shadow variables shrink identification intervals by 75–83% while maintaining valid coverage under realistic MNAR response patterns.
Bio ▶
Hongyu Chen is a fourth-year PhD student at MIT, advised by Prof. David Simchi-Levi. His research develops methodologically rigorous frameworks for data collection and decision-making in complex environments, with a particular focus on leveraging generative AI as a data augmentation tool to improve efficiency, reliability, and economic value. Before joining MIT, he received his bachelor’s degree in Statistics and Economics from Peking University.
10:30 – 11:10
Peibo Zhang
Emory University
The Impact of AI Search on Online Content Ecosystem: Evidence from Google and Reddit
Abstract ▶
Search engines traditionally complement online content platforms by directing users seeking information to external websites. The emergence of generative AI search tools that summarize answers directly on the results page may disrupt this relationship by reducing the need to visit source platforms. We study this question using Google AI Overviews and Reddit, one of the largest online discussion platforms. Our identification exploits Google’s content moderation policy: Safe-for-Work (SFW) Reddit communities are indexed by Google organic search and surfaced in Google AI Overviews, while Not-Safe-for-Work (NSFW) communities, though indexed by organic search, are prohibited from being referenced in AI Overview summaries. Using a difference-in-differences design, we find that AI Overviews increase engagement in SFW communities: daily comments rise by 10.7 percent and the number of commenting users by 11.3 percent relative to NSFW communities. The effects are concentrated in experience-based discussions (opinions, advice, and personal experiences) rather than fact-based information. However, the subsequent introduction of Google AI Mode, which allows users to interact conversationally with the AI summary, largely eliminates these gains in experience-based content. These results suggest that the effects of AI search depend critically on interface design and types of content.
Bio ▶
Peibo Zhang is a PhD Candidate in Operations Management at Emory University. Her research sits at the intersection of platform operations, generative AI applications, and social impact.
11:20 – 12:00
Ruicheng Ao
Massachusetts Institute of Technology
Operational Evaluation of AI-Mediated Operational Decisions
Abstract ▶
As AI systems enter operational workflows, their outputs increasingly become inputs to consequential decisions: whether to release an action, repair an artifact, or send a case for expert review. This creates an operations problem that is not captured by static accuracy alone. The relevant question is how benchmark evidence, feedback, verification cost, and downstream operational consequences should jointly determine the next action. In this talk, I will develop a workflow-grounded view of operational evaluation for AI-mediated decision systems. Motivated by a real-world supply chain workflow, I will show why AI-generated recommendations require evaluation in the context of the operational decisions they trigger. I will then discuss how operational methods can help decide when such recommendations should be released, repaired, or sent for expert review. The broader message is that Operations Research can play a central role in making AI systems reliable in real-world decision workflows.
Bio ▶
Ruicheng Ao is a third-year Ph.D. candidate in the Institute for Data, Systems, and Society (IDSS) at MIT, working with Prof. David Simchi-Levi and Prof. Thomas Magnanti. His research studies how Operations Research, Statistics, and Large Language Models can jointly enable efficient and reliable AI-driven decision systems. On one side, he uses optimization, stochastic modeling, and statistical decision theory to improve LLM serving, evaluation, training, and human-AI decision-making. On the other hand, he develops LLM-based agents and interactive systems for optimization, service operations, and supply chain. This summer, he’s doing his internship at the Tencent Hunyuan Frontier AI lab, supervised by Shunyu Yao. Before joining MIT, he received his B.S. from the School of Mathematical Sciences at Peking University
12:00 – 14:00
Lunch
14:00 – 15:00
Max Shen
The University of Hong Kong
Emerging Topics on AI Applications
Abstract ▶
TBA
Bio ▶
Professor Zuo-Jun Max Shen currently serves as Senior Advisor to the President and Dean of Graduate School at The University of Hong Kong. He is also a Chair Professor jointly in the Faculty of Engineering and Faculty of Business and Economics. After obtaining his PhD from Northwestern University in 2000, he began his academic career at the University of Florida before joining UC Berkeley in 2004, where he rose through the ranks to become Chancellor’s Professor and Chair of Department of Industrial Engineering and Operations Research and Professor of Department of Civil and Environmental Engineering. He joined HKU as the Vice-President and Pro-Vice-Chancellor (Research) in 2021, where he has established several innovation centers and research institutes. He is also an INFORMS Fellow and POMS Fellow.
Prof. Shen is an internationally renowned scholar in intelligent decision-making for complex systems, spanning global manufacturing, logistics, and energy sectors. His breakthrough innovations in AI-driven supply chain optimization have earned unprecedented recognition while delivering measurable improvements in global sustainability and operational efficiency. Prof. Shen currently serves as the Chief Scientist of Supply Chain at JD.com and the Chief Scientist of Industrial AI at TCL. The body of research work collaborating with industries has earned him Franz Edelman Laureate twice, Gartner Award, Wagner Prize, and INFORMS Prize.
15:00 – 15:20
Coffee Break
15:20 – 16:00
Tong Xie
University of Chicago
Speed of Intervention in Algorithmic Markets: Controlling Collusion and Stability
Abstract ▶
We study how platform interventions shape markets with algorithmic collusion. The effectiveness of intervention depends not only on how strong it is, but also on how fast it is implemented. While intervening too aggressively can destabilize the system, the timing of the intervention also matters. Sudden interventions may induce instability when pushing algorithms out of the “basin of attraction”, leading the market into persistent oscillation. Our results show that effective market regulation requires careful design of both intensity and speed of intervention.
Bio ▶
Tong Xie is a fourth-year Ph.D. candidate in Operations Management at the University of Chicago Booth School of Business. Her research focuses on algorithmic markets, platform design, and learning dynamics, with an emphasis on how platform interventions and algorithmic decision rules influence collusion, stability, and long-run market outcomes. She holds a B.B.A. in Economics from CUHK-Shenzhen in 2020, and an M.Phil. in Industrial Engineering from HKUST in 2022.
16:10 – 16:50
Renfei Tan
Massachusetts Institute of Technology
Multi-agent Adaptive Mechanism Design
Abstract ▶
We study a sequential mechanism design problem where a principal elicits truthful reports from multiple rational agents without prior knowledge of their beliefs. We propose Distributionally Robust Adaptive Mechanism, which integrates mechanism design and online learning to jointly ensure truthfulness and cost efficiency. The mechanism iteratively estimates agents’ beliefs and solves a distributionally robust program with shrinking ambiguity sets. It achieves high-probability truthfulness and low regret, with a matching lower bound.
Bio ▶
Renfei Tan is a fourth-year PhD student from MIT, Institute for Data, Systems, and Society. His research focus is on the combination of mechanism design problems with statistical learning and modern AI techniques. His research work has recently won the Best Paper at MIT LIDS 31st Annual Student Conference.
17:00 – 17:40
Jiacheng Chang
The University of Hong Kong
Adaptive Design for In-App Advertising Games: A Data-Driven Methodology Validated with Field Experiments
Abstract ▶
Rewarded ads, in which players voluntarily watch videos in exchange for in-game benefits, have become an important monetization format in both mobile gaming and digital advertising. Yet they create an operational tension: designs that increase immediate ad usage may frustrate players and reduce future retention. We study this intertemporal trade-off in level-based puzzle games and develop a data-driven framework to maximize players’ expected lifetime rewarded-ad views through dynamic difficulty control. The problem is challenging because firms observe only limited early behavior, while player response exhibits heterogeneity in skill, ad tolerance, and churn propensity, as well as path dependence driven by difficulty reference and prior ad exposure. We formulate the firm’s decision problem as a partially observable Markov decision process with latent player types and instantiate it with a parsimonious behavioral model of win probability, ad watching, and churn. For estimation, we exploit the model structure to develop a multi-step procedure based on recursive logistic regressions. For deployment, we compute policies offline using partially observable Monte Carlo planning and compile them into compact lookup tables that satisfy sub-second latency requirements in live game systems. In collaboration with Tencent, China’s largest digital advertising network, we implement the framework in commercial games and evaluate it through two large-scale randomized field experiments. Relative to the industry baseline, the deployed policy increases long-run rewarded-ad views by 19.23% and 13.58%, while also improving engagement and retention. These results show how dynamic, data-driven experience design can serve as an implementable revenue-management lever in ad-supported digital platforms.
Bio ▶
Jiacheng Chang is a PhD student at HKU Business School. His research focuses on revenue management and digital platforms, using dynamic programming and data-driven methods. He won Second Place in the 2025 INFORMS Behavioral Operations Management Best Working Paper Competition and has deployed his data-driven algorithms at Tencent.
17:40 – 18:00
Closing Remarks
Prof. Zhixi Wan
Day 2 (June 10)
9:10 – 9:20
Opening Remarks
Prof. Zhixi Wan
9:20 – 10:00
Xiao Lei
The University of Hong Kong
Point Redemption Design in Gamified Education
Discussant: Baozhuang Niu (South China University of Technology)
Abstract ▶
Points are a common feature of digital learning platforms, but they are typically used as symbolic feedback rather than as redeemable assets. We examine whether converting points into a functional currency improves student learning behavior, and whether the design of the redemption mechanism matters. In a randomized field experiment conducted with a large K-12 education provider in China, students in all groups earned points for learning activities, while treated students could redeem points for non-monetary digital collectible cards through one of three mechanisms: store, gacha, or time-limited redemption. Using matched difference-in-differences analysis on 7,359 student-week observations from 388 ninth-grade students, we show that point redemption increases assignment completion and error correction relative to a points-only control. The gacha design produces the strongest and most consistent effects. Mechanism analyses suggest that redemption increases motivation by giving points functional value, and that gacha outperforms deterministic designs because uncertainty prolongs pursuit of preferred rewards. The gains are concentrated among lower-performing students, female students, and non-STEM subjects. Our findings highlight point redemption as a consequential design lever in gamified education and show that the effectiveness of gamification depends not only on whether rewards exist, but also on how they are redeemed.
10:00 – 10:40
Zhengli Wang
The University of Hong Kong
Strategies for Milestone-driven Start-ups in Multi-activity Settings
Discussant: Dongyuan Zhan (University of Science and Technology of China)
Abstract ▶
New venture start-ups need to “survive” through multiple stages of reaching milestone targets. We investigate the strategies for start-ups in a milestone-oriented setting by examining a model of an entrepreneurial start-up firm, where its state is captured by a diffusion process. The entrepreneur can choose between multiple activities (or controls), which incur different cost and determine the drift and the variance of the process. Depending on whether the process reaches a fixed upper boundary or a fixed lower one, the start-up firm succeeds or fails. We completely solve for the optimal policy and provide an explicit characterization of its structure. In particular, the optimal policy only uses activities from a set characterized by a so-called efficient frontier curve that orders the activities by two intuitive measures: riskiness (drift-to-volatility ratio) and cost-effectiveness (drift-to-cost ratio). A unique feature of our model is that depending on the model parameters, the efficient frontier curves can be of different types, resulting in qualitatively different structures of the optimal policy. As far as we know, this is the first study that analyzes a stochastic control model which admits efficient frontier curves of different types. Our work provides start-up firms with intuitive measures to evaluate their activities and offers valuable insights on how the optimal strategies in a milestone-oriented setting change qualitatively contingent upon the specific scenario. We believe the results provide a foundational block in the study of entrepreneurial decision-making.
10:40 – 11:00
Coffee Break
11:00 – 12:00
Saif Benjaafar
University of Michigan
Over-Crowded and Under-Matched: Matching Queues with Strategic Agents
Abstract ▶
We study a dynamic matching environment in which agents strategically decide whether to join a queue and, conditional on joining, whether to accept a match or wait for a higher-quality opportunity. We show that equilibrium behavior can simultaneously exhibit excessive entry and insufficient matching relative to the socially optimal allocation. These two distortions operate on distinct margins—admission and selectivity—and reinforce one another, leading to overcrowded queues and reduced throughput. We characterize equilibrium and socially optimal policies in a tractable continuous-time model, establish a sharp comparison between them, and show that decentralization of the social optimum requires pricing both entry and match acceptance.
Bio ▶
Saif Benjaafar is the Seth Bonder Collegiate Professor of Industrial and Operations Engineering and Professor of Electrical Engineering and Computer Science at the University of Michigan, where he also serves as the Goff Smith Co-Director of the Tauber Institute for Global Operations, a joint institute of the Ross School of Business and the College of Engineering. Prior to joining the University of Michigan, he was McKnight Presidential Endowed Professor and Distinguished McKnight University Professor at the University of Minnesota. He is a founding member of the Singapore University of Technology and Design and served as Pillar Head of Engineering Systems and Design. He is a Fellow of INFORMS, MSOM, and IISE and a former Editor-in-Chief of the INFORMS journal Service Science.
12:00 – 14:00
Lunch
14:00 – 14:40
Feng Tian
The University of Hong Kong
Managing Human and AI in Innovation: A Dynamic Principal–Agent Framework
Discussant: Wenbin Wang (Shanghai University of Finance and Economics)
Abstract ▶
This paper studies how a firm should dynamically combine human researchers and artificial intelligence in innovation when human effort is privately chosen, but AI can be directly controlled. We develop a continuous-time principal-agent model in which a principal seeks a Poisson breakthrough from a human agent, an AI agent, or both operating in parallel. The human agent can shirk and must be incentivized through a dynamic contract, while AI entails an operating cost but no moral hazard. The principal commits to deployment rules, success payments, and stopping times. We characterize the optimal contract. Access to AI benefits the principal but hurts the human agent relative to a no-AI benchmark. Compared with hiring a second human, AI adoption can create either preference conflict or preference alignment depending on AI productivity: weak AI is preferred by the human agent, strong AI is preferred by the principal, and intermediate AI may generate joint preference for either Human-AI or Human-Human. Extensions show that imperfect AI breakthroughs can induce time-varying verification, while mandatory background AI may reduce the principal’s payoff by weakening AI’s role as an incentive threat.
14:40 – 15:20
Minje Park
The University of Hong Kong
Do Surgeons Observe the Queue? Surgeons’ Strategic Behavior and Its Consequences on Patients
Discussant: Hailiang Chen (The University of Hong Kong)
Abstract ▶
We study surgeons’ strategic behavior in adjusting service rates based on upcoming scheduled surgeries. Using an extensive dataset from a large U.S. healthcare system, we find that surgeons accelerate their pace when facing both their own subsequent cases (self-interest) and those of other lead surgeons (altruism). Notably, this speed-up effect is absent when no queue exists, suggesting that surgeon productivity is a strategic response to system workload.
15:20 – 15:40
Coffee Break
15:40 – 16:40
Jeannette Song
Duke University
Innovator’s Edge in Supply Chain Transparency and Food Waste Reduction
Abstract ▶
New digital technologies allow retailers to track product freshness in real time and, in some cases, disclose that information directly to consumers. In the fresh produce retail industry, these tools can improve inventory decisions, reduce food waste, and influence consumer demand. Yet adoption is costly, and early movers often face higher implementation costs than later adopters. This raises a central question: when does moving first actually pay off? To address this question, we develop a dynamic model of competing retailers facing uncertain supply freshness and heterogeneous consumers. In our setting, transparency creates value both operationally, by improving inventory decisions, and strategically, by reducing consumers’ information frictions. We show that the value of transparency depends not only on the technology itself but also on how retailers deploy it, including whether they use it solely for internal decision-making, disclose it to consumers, or combine it with pricing actions. Our analysis yields new insights into when early adoption is advantageous, how transparency reshapes competition, and how digital adoption affects both profitability and food waste. (Joint work with Bora Keskin and Chenghuai Li.)
Bio ▶
Jeannette Song is the R. David Thomas Professor of Business Administration and a Professor of Operations Management at the Fuqua School of Business of Duke University. She studies supply chain management and operations strategy. Her recent research covers supply chain digitization, data-driven operational decision-making, resilient supply chain strategies, and responsible operations. She has published extensively in leading academic journals and has edited the “Research Handbook on Inventory Management” (Edward Elgar Publishing, 2023). Professor Song is an INFORMS Fellow, an MSOM Fellow, and a former President of MSOM; she is also a Department Editor for Management Science.
16:50 – 17:30
Wei Zhang
The University of Hong Kong
Post-Experiment Decisions: Predict, Adjust, then Rollout and Optimize
Discussant: Peng Wu (Sichuan University)
Abstract ▶
Firms increasingly use randomized experiments to decide whether to scale up an intervention and, if so, how to re-optimize related operational choices such as inventory, capacity, or pricing. In many settings, experiments are performed on small samples, so the estimated effect of the intervention is uncertain. A common practice is to plug a “significant” estimate of the effect into both (i) the rollout rule and (ii) the downstream optimization. However, this can lead to avoidable losses because the costs of over- versus under-estimating the effect are often asymmetric. The technically ideal approach is to obtain a data-dependent decision rule that minimizes the Bayes risk, but this lacks transparency and requires more computations. We propose Predict–Adjust–Then–Rollout–Optimize (PATRO), a plug-in approach that keeps the standard estimate, but makes data-independent adjustments, respectively, for the two types of decision. We show that the two adjustments can be substitutes or complements and provide an alternating-iteration method to compute the pair. Surprisingly, PATRO performs both in theory and numerically close or equivalent to the Bayes-optimal benchmark, making it a simple, effective way to convert noisy experimental results into better rollout and operational decisions without changing a firm’s estimation pipeline or decision models.
17:40 – 18:20
Huiyin Ouyang
The University of Hong Kong
Autonomous Delivery Vehicle Integration in Instant Retail: A Multi-Level Analysis
Discussant: Zhixi Wan (The University of Hong Kong)
Abstract ▶
Integrating autonomous delivery vehicles (ADVs) with human riders promises efficiency gains in instant retail, yet operational consequences remain poorly understood. Using 1.1 million orders from a leading Chinese front-warehouse instant retail platform (March 2025–January 2026), we exploit non-monotonic ADV rollout (0→1→0→1) to identify causal effects via difference-in-differences with treatment reversal, stacking matching, and coarsened exact matching. We document a multi-level efficiency paradox. At the order level, hybrid ADV-rider delivery increases completion time by 240–311 seconds and raises delay probability by 3–5 percentage points—a “coordination tax” from batching and handoff friction. Yet at the rider–AOI–hour level, throughput rises by 0.35–0.42 orders per hour as ADVs offload long-distance transit, enabling spatial specialization. These system gains come at a rider-level cost: idle time increases by 3.9–5.3 minutes per hour. Our findings reveal that automation transforms rather than eliminates operational friction; managing coordination costs, not merely deploying ADVs, is the critical lever for hybrid delivery performance.
18:20 – 18:30
Closing Remarks
Prof. Zhixi Wan
18:30 ~
Dinner