Yufei Zhu

Yufei Zhu

Ph.D. Candidate
Institute of Operations Research and Analytics (IORA)
National University of Singapore (NUS)

Email:

Hi! I am a Ph.D. candidate in the Institute of Operations Research and Analytics at the National University of Singapore, advised by Zhenyu Hu and Joel Goh.

I study the design of dynamic operational processes, particularly searches for talent, ideas, and breakthroughs. In these settings, participants respond strategically to one another and, increasingly, to AI agents. My work asks what information to reveal, how to structure incentives, when to introduce competition, and when to delegate decisions to participants. A recurring theme is that policies that look beneficial can backfire through the behavior they induce. I examine these questions in contexts including labor-platform hiring, innovation sourcing, and human–AI contracting.

More recently, I have been studying healthcare operations in collaboration with practitioners at the National University Health System. I develop routing and scheduling models, together with an LLM-assisted planning tool, for their hospital-at-home program.

I received my B.Eng. in Industrial Systems Engineering and Management from NUS, with a Minor in Financial Mathematics.

Research Interests

Research Topics: Information and Incentive Design, Human–AI Operations, Healthcare Operations

Methodologies: Game Theory, Dynamic Programming, Dynamic Contracting

© 2026 by Yufei Zhu

Research

Working Papers

  1. 1.
    The Strategic Secretary Problem (Job Market Paper)
    with Joel Goh.
    Under review at Management Science.
    Abstract

    In the classical secretary problem, applicants have no agency and arrive in a random order. Motivated by online labor platforms, we study a variant in which secretaries are strategic agents who choose their arrival times after privately observing their own quality scores, and we ask whether this agency improves or reduces the firm’s ability to hire the best candidate. When the firm can commit to a hiring policy and observe quality scores, strategic behavior raises the probability of hiring the best secretary to approximately 0.71 as the applicant pool grows, compared with 0.58 in the classical problem. When the firm observes only rank-based information, the probability is the same as in the classical problem; when it observes quality scores but cannot commit, no symmetric pure-strategy equilibrium exists. Granting applicants agency thus improves hiring only when firms can commit to hiring policies and access accurate quality signals.

  2. 2.
    Delegating Innovation: Sequential and Parallel Search under Competition
    with Zhi Chen and Zhenyu Hu.
    Under review at Production and Operations Management.
    Abstract

    Concept generation is critical to new product development, and firms must search broadly across possible concepts to find a winning idea. Sequential search develops and evaluates one concept at a time, stopping once a promising idea is found; parallel search explores multiple concepts simultaneously before selecting the best. When a buying firm delegates this search to competing suppliers, which strategy should it prefer? Embedding a classical search model into a contest framework, we show that the buyer prefers sequential search when the unit search cost is high or the maximum number of evaluation opportunities is large: it lets suppliers adapt using performance feedback (a “quality effect”) and encourages them to search more extensively (a “quantity effect”). Otherwise, parallel search is preferred. The optimal supplier base always features restricted entry, with as few as two suppliers.

  3. 3.
    Racing Against Automation: Dynamic Incentives When AI Can Preempt the Human
    with Feng Tian, Yangge Xiao and Ying-Ju Chen.
    Major revision at Manufacturing & Service Operations Management.
    Abstract

    AI agents are reshaping innovation by enabling autonomous search, yet human researchers remain essential in novel and poorly structured settings. A human agent may be more productive but privately chooses effort, creating a moral hazard problem; an AI agent requires no incentives but is costly to operate and can preempt the human’s success. When, if ever, should a firm race the AI agent against the human rather than rely on either one alone? Using a continuous-time principal–agent model, we show that keeping a cost-effective AI agent active as a persistent rival may lower both the human agent’s and the principal’s payoffs relative to human-alone search. Instead, the optimal contract may begin with human-alone search, transition to a human–AI race, and finally switch to autonomous AI search if no breakthrough occurs. AI rivalry is a double-edged sword: it can accelerate discovery and discipline human effort, but deploying it too early crowds out valuable human-alone search.

Work in Progress

  1. 4.
    Timely Care at Home: LLM-Assisted Routing and Scheduling for Hospital-at-Home Programs
    with Yvonne Huijun Zhu, Joel Goh and Stephanie Ko.
  2. 5.
    Intertemporal Reference Design Through Performance Transparency
    with Lin Chen and Guillaume Roels.

Selected Talks

The Strategic Secretary Problem

Delegating Innovation: Sequential and Parallel Search under Competition

Racing Against Automation: Dynamic Incentives When AI Can Preempt the Human

© 2026 by Yufei Zhu

Teaching

Guest Instructor

Teaching Assistant

Tutorial Instructor

© 2026 by Yufei Zhu

Open CV ↗

© 2026 by Yufei Zhu