Zihao Qu: How I Approach the Science of Decision-Making

When I play video games, I often find myself managing trade-offs that shape the gameplay experience: which character to play, which armor to use, which magic spell to learn, and so on. These choices have always intrigued me, often prompting me to browse online forums to understand the underlying trade-offs more deeply. Maybe I was born with an interest in operational problems, or maybe years of gameplay cultivated my fascination with them. One way or another, this interest now motivates my research on real-time decision making under uncertainty.

Even familiar static decisions can become surprisingly complex at scale, while adding dynamic elements increases the difficulty. For example, sorting a short playlist may be easy, but ranking thousands of songs in a way that fits your mood, preferences, and context is much harder. Business environments add even more complexity: Customers arrive randomly over time, resources are limited, and decisions must often be made before all uncertainty is resolved. My goal is to develop simple, efficient, and near-optimal algorithms for computationally challenging problems that real organizations face.

Online Platforms Weigh Timeliness and Opportunities

One stream of my research studies operational decisions in online platforms. I am drawn to these settings because of their dynamic nature: users arrive, preferences shift, information is incomplete, and the platform has to make decisions in real time. A good decision is rarely just about optimizing one outcome at one moment. It often affects what users experience now, what the platform learns next, and how the system evolves over time.

In one project, my coauthors and I study spatial matching with delays. This work is motivated by settings such as carpooling, food delivery, and online matchmaking, where a platform must decide whether to match requests immediately or wait for better opportunities. Acting quickly reduces waiting time, but waiting can thicken the market and improve match quality. We develop and analyze practical policies that match nearby requests, group requests by region, or batch requests over time. This project highlights a broader theme in my research: simple real-time rules can perform well if they are properly designed.

I explore a related problem in competitive live-service games, a setting that also connects closely to my long-standing interest in gameplay trade-offs. In these platforms, player engagement depends on both fair matchmaking and well-calibrated character strengths. The platform must learn from noisy match outcomes, update character balance over time, and match players in ways that remain fair and engaging. My coauthors and I formulate a dynamic problem that jointly considers balancing, learning, and matchmaking, and we develop a policy that clarifies the complementary roles of these decisions.

Predicting Resource Needs

A second stream of my research studies resource allocation problems under uncertainty. Organizations must commit to resources before knowing exactly what future demand will look like, while still preserving enough flexibility to respond when reality turns out differently.

In post-acute healthcare, for example, providers must decide how many full-time nurses, therapists, and other professionals to employ, and when to rely on more expensive temporary resources to accommodate incoming patient referrals that may require care over many days or weeks. In cloud computing, firms face an analogous problem: they must decide which computing resources to reserve in advance, which to use on demand, and how to assign jobs with different deadlines and processing requirements to different resource types over time. Across both settings, my work asks how organizations can make good real-time allocation decisions when demand, resource needs, and future system conditions are uncertain.

Zihao Qu, PhD, is an assistant professor in the Department of Operations and Information Management at the Isenberg School of Management at UMass Amherst.