About

My main interest is in service operations management; see [this explanation](http://review.chicagobooth.edu/strategy/2019/article/amy-r-ward-says-customers-are-getting-impatient) in the Chicago Booth Review. Services encompass a broad and diverse range of industries including airlines, hospitals, call centers, and online marketplaces. The importance of promoting efficient operations in service firms is largely due to the fact that service firms provide most of the GDP and employment in post-industrial economies such as the U.S. Service firms cannot predict either when customers will arrive or how long the processing of each customer will take. Even when customers schedule appointments (such as to visit a physician), the service provider must deal with early and late arrivals, cancellations, and no-shows. In contrast to firms that produce products, service firms cannot build up inventory in order to buffer themselves from unexpected bursts in customer arrivals. Hence service firms prior

Research topics

• Computer Science
• Marketing
• Machine Learning
• Economics
• Data Mining
• Engineering
• Business
• Operations research
• Artificial Intelligence
• Mathematical optimization
• Microeconomics
• Econometrics
• Mathematics
• Labour economics
• Industrial organization

Selected publications

• Enhancing Predictive Model Learning via Domain-Knowledge Augmented Latent Feature Mining

  Proceedings of the AAAI Conference on Artificial Intelligence · 2026-03-14

  articleOpen accessSenior author

  Predictive modeling in high-stakes domains often suffers from limited observed features due to ethical and practical constraints. To address this challenge, we propose a novel approach that formulates latent feature mining as a text-to-text propositional logic reasoning task, facilitating domain knowledge integration and improving the interpretability of latent features. We design FLAME, a domain knowledge-augmented reasoning framework for latent feature mining, offering an efficient training paradigm to strengthen the domain-specific reasoning capabilities of large language models (LLMs) for latent feature extraction. The goal of our framework is to augment observed features with inferred latent features, enhancing the performance of predictive models in downstream machine learning tasks. We validate our approach through two case studies: (1) the criminal justice system, where data collection is ethically challenging and inherently limited, and (2) the healthcare domain, where patient privacy concerns and the complexity of medical data restrict comprehensive feature collection. Experimental results demonstrate that the inferred latent features significantly enhance the performance of downstream classifiers by over 10%.

  PublisherOA PDFDOI

• Admission Decisions under Imperfect Classification: An Application in Criminal Justice

  SSRN Electronic Journal · 2025-01-01

  preprintOpen accessSenior author
  PublisherDOI

• Data-Driven Matching for Impatient and Heterogeneous Demand and Supply

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access
  PublisherDOI

• When Strategic Customers Meet Strategic Servers: Individual and Social Optimization in Many-Server Queueing Systems

  SSRN Electronic Journal · 2025-01-01

  preprintOpen accessSenior author
  PublisherDOI

• Learning to Schedule in Multiclass Many-Server Queues with Abandonment

  Operations Research · 2024-11-25 · 4 citations

  articleSenior author

  How to Learn Which Customer Class to Serve Next? In “Learning to Schedule in Multiclass Many-Server Queues with Abandonment”, Zhong, Birge, and Ward tackle the challenge of scheduling (that is, how to choose the customer that a newly available server will serve) in a multiclass many-server queueing system where customers may abandon the queue. The goal is to develop a scheduling policy that performs nearly as well as a benchmark policy under full knowledge of the model primitives despite these primitives being unknown and needing to be learned. They propose a Learn-then-Schedule policy that first estimates the unknown model primitives empirically and then schedules according to the benchmark policy structure using these estimates. Such a policy achieves an optimal regret rate of order logT (where T is the system time), meaning that the performance gap between the proposed policy and the benchmark policy grows logarithmically over time.

  PublisherDOI

• Mental Health Services from 1990 to 2023

  2024-04-18 · 2 citations

  reference-entrySenior author

  Mental health services in the United States have changed dramatically over the past century. At that time, only inpatient services were generally available, principally through state mental hospitals. In the intervening period, a community-based mental health system has been developed that runs in parallel with inpatient services but also allows for a much broader continuum of care. This community-based system includes brief psychotherapy, pharmacology, other forms of ambulatory care, and short-term residential care. The ongoing transition from inpatient to community-based mental health services has not been without internal controversies, particularly the issues surrounding the availability and use of inpatient care for community programs. The purpose of this bibliography is to examine US mental health services and the system that embeds them.

  PublisherDOI

• Latent Feature Mining for Predictive Model Enhancement with Large Language Models

  arXiv (Cornell University) · 2024-10-06

  preprintOpen accessSenior author

  Predictive modeling often faces challenges due to limited data availability and quality, especially in domains where collected features are weakly correlated with outcomes and where additional feature collection is constrained by ethical or practical difficulties. Traditional machine learning (ML) models struggle to incorporate unobserved yet critical factors. In this work, we introduce an effective approach to formulate latent feature mining as text-to-text propositional logical reasoning. We propose FLAME (Faithful Latent Feature Mining for Predictive Model Enhancement), a framework that leverages large language models (LLMs) to augment observed features with latent features and enhance the predictive power of ML models in downstream tasks. Our framework is generalizable across various domains with necessary domain-specific adaptation, as it is designed to incorporate contextual information unique to each area, ensuring effective transfer to different areas facing similar data availability challenges. We validate our framework with two case studies: (1) the criminal justice system, a domain characterized by limited and ethically challenging data collection; (2) the healthcare domain, where patient privacy concerns and the complexity of medical data limit comprehensive feature collection. Our results show that inferred latent features align well with ground truth labels and significantly enhance the downstream classifier.

  PublisherOA PDFDOI

• Combining Machine Learning and Queueing Theory for Data-Driven Incarceration-Diversion Program Management

  Proceedings of the AAAI Conference on Artificial Intelligence · 2024-03-24 · 3 citations

  articleOpen accessSenior author

  Incarceration-diversion programs have proven effective in reducing recidivism. Accurate prediction of the number of individuals with different characteristics in the program and their program outcomes based on given eligibility criteria is crucial for successful implementation, because this prediction serves as the foundation for determining the appropriate program size and the consequent staffing requirements. However, this task poses challenges due to the complexities arising from varied outcomes and lengths-of-stay for the diverse individuals in incarceration-diversion programs. In collaboration with an Illinois government agency, we develop a framework to address these issues. Our framework combines ML and queueing model simulation, providing accurate predictions for the program census and interpretable insights into program dynamics and the impact of different decisions in counterfactual scenarios. Additionally, we deploy a user-friendly web app beta-version that allows program managers to visualize census data by counties and race groups. We showcase two decision support use cases: Changing program admission criteria and launching similar programs in new counties.

  PublisherOA PDFDOI

• Matching Impatient and Heterogeneous Demand and Supply

  Operations Research · 2024-05-15 · 20 citations

  articleSenior author

  Balancing Speed and Value in On-Demand Matching Platforms In “Matching Impatient and Heterogeneous Demand and Supply,” Aveklouris, DeValve, Stock, and Ward consider a fundamental trade-off faced by many platforms (e.g., Uber/Lyft) that match supply (e.g., drivers) and demand (e.g., riders) dynamically over time: making matches quickly capitalizes on the value of current supply and demand in the system, whereas waiting may enable better matches at the risk of losing impatient customers. They show that this trade-off can be balanced by waiting a short amount of time before making matches: long enough to gather enough agents to make valuable matches but not so long that impatient agents are likely to leave. Intuitively, this balance depends on how long agents are willing to wait, on average, but the authors show that it also depends on the full distribution of the willingness to wait (i.e., not only mean, but also variance and higher moments play a role). Thus, approaches that only take into account the mean willingness to wait may perform quite poorly. Further, the authors develop an algorithm to rank matching priorities in order to achieve an optimized trade-off between speed and value of matches.

  PublisherDOI

• Closing the Service: Contrasting Activity-Based and Time-Based Systematic Closure Policies

  2024-12-15 · 1 citations

  article

  We examine different policies for systematic service closure in messaging service systems. The system is modeled as an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M/UHP/1$</tex> queue, where service times follow a history-based Hawkes cluster process. We propose and examine stopping-time rules that balance between queue length and the probability of prematurely closing conversations. In a simulation study, we compare two families of systematic closure policies: the first relies on predictive information regarding service progress, i.e., the conversation's activity levels, while the second relies on elapsed time without activity. When restricted to static threshold policies, both families provide similar performance. However, when allowing the threshold to vary with the system state, activity-level policies outperform the inactive-time policies. Moreover, a large difference is observed between static and dynamic threshold policies. We therefore conclude that state-dependent (i.e., dynamic) activity-based policy is the most promising candidate to achieve optimal closure rules.

  PublisherDOI

Frequent coauthors

• Harsha Honnappa

  13 shared

• Dongyuan Zhan

  University College London

  12 shared

• Rahul Jain

  University of Southern California

  11 shared

• Chihoon Lee

  Stevens Institute of Technology

  10 shared

• Erica L. Plambeck

  10 shared

• Josh Reed

  New York University

  8 shared

• Yueyang Zhong

  University of Chicago

  8 shared

• Peter W. Glynn

  7 shared

Education

• Ph.D., Management Science and Engineering

  Stanford

  2001

Awards & honors

• Fellow of the INFORMS Manufacturing and Service Operations M…