University of RochesterGuanling Yang

RESEARCH

Technology changes.
Decisions matter.

[1] Information Disclosure Policy in Omnichannel Services with Invisible Customers

with Ricky Roet-Green

  • Under Revision at Production & Operations Management
  • Finalist, MIF Summer Webinar Competition, 2026
  • Presentations: INFORMS MSOM Conference 2026, 2025, 2024; INFORMS Applied Probability Society Conference 2025; POMS Annual Conference 2025; INFORMS RMP 2025

The queue you see may not be the queue you join. When online customers are invisible to walk-in customers, what should firms reveal about demand across their service channels?

Illustration of in-person and online customers and uncertainty about the queue.
Invisible customers across service channels
Restaurant brands positioned by walk-in order percentage and customers’ normalized value from service.
Omnichannel service across industries

[2] On the design of an AI-Human Service System with Strategic Customers: Should AI be Mandated?

with Ricky Roet-Green

  • Under Revision at Manufacturing & Service Operations Management
  • Presentations: INFORMS Annual Meeting 2026

Should every customer have to try AI before reaching a human? We examine when an AI mandate improves throughput and social welfare—and when giving customers a choice works better.

Throughput comparison by AI proficiency and demand level. Regions identify Choice, Mandate, or Equivalent performance.
Throughput
Social welfare comparison by AI proficiency and demand level. Regions identify Choice, Mandate, or Equivalent performance.
Social welfare

[3] When Better AI Hurts: Verification, Compliance, and AI-Human Handoffs in Services Systems

with Cuihong Li, Ricky Roet-Green, Fasheng Xu

  • Under Review
  • Accepted for 2026 MSOM TIE SIG Conference
  • First place, NYON Student Best Paper Award Competition, 2026
  • Presentation: NYC Operations Day 2026, INFORMS MSOM SIG Conference 2026, AI in Business Conference 2026, INFORMS Annual Meeting 2026

Can more accurate AI make a service less profitable? We uncover an AI improvement trap: when customers can skip human verification, reducing hallucination risk can paradoxically lower a firm’s optimal profit.

Customers join AI service, then exit unverified or decide whether to comply with human verification.
AI–human service workflow
Optimal operating profit falls then rises with hallucination risk level, following the curve supplied by the author.
Hallucination risk and operating profit

[4] Power Without the Grid? Data Centers and Business Models

with Ricky Roet-Green, Nur Sunar, Lin Zang

  • Working paper

How will data centers secure the electricity needed to power the AI boom? This project explores the business models emerging as AI drives growing demand for computing capacity and electricity.

[5] On Social Loss From Balking: Counting the Unserved Customers

with Ricky Roet-Green

  • Working paper

When someone gives up on a queue, who bears the cost? In services such as voting, healthcare, and humanitarian aid, we examine how the social cost of lost participation changes the design of a welfare-maximizing queue.

[6] Pro-worker AI? The Pricing and Staffing Consequences of Augmentation and Automation AI

  • Working paper

Does AI help workers—or replace them? We compare AI that augments people with autonomous AI agents, tracing their consequences for pricing, staffing, profits, customer welfare, and investment.