Services Mini-Conference

Ride-hailing platforms face two important challenges: (i) There are significant spatial demand imbalances that require some supply to reposition; (ii) the control of supply is partially decentralized in that drivers strategically decide whether to join the network, and if so, whether and where to reposition when not serving riders. For such ride-hailing networks we study the question: under what conditions can driver wage policies align these decentralized driver decisions with the system-optimal (centralized) allocation of drivers across a spatial network? 

We develop a stationary game-theoretic fluid model of a multi-location ride-hailing network with general network topology, strategic driver participation and forward-looking repositioning, flexible wage policies, and both constant and congestion-sensitive travel times. The platform sets wages to maximize profit, while drivers make participation and repositioning decisions to maximize their long-run profit rate.

Our equilibrium characterization transforms the drivers’ repositioning problem into conditions on node visits, network flows, wages, and waiting times. This characterization identifies two distinct components: wage–waiting-time conditions that shape where drivers seek demand, and network-flow optimality conditions that govern their repositioning decisions. Building on this characterization, we show how the effectiveness of driver wage policies under decentralized repositioning depends on the interplay of wage flexibility, and the congestion-sensitivity and spatial relations of travel times. With constant travel times, decentralized repositioning can implement the centrally optimal repositioning flows, but sufficient wage flexibility is critical: origin-dependent wages eliminate driver idling and attain the centralized capacity and profit benchmark, whereas restricted wages generally require inefficient driver waiting to balance spatial incentives. With congestion-sensitive travel times, the centrally optimal repositioning flows can generally not be implemented under decentralized repositioning, even with full wage flexibility, so that decentralized repositioning leads to higher congestion, capacity levels, and wage rates compared to centralized repositioning.

(Joint work with Andre Cire and Uta Mohring.)

In many service systems, routing decisions must be made before detailed individual-level information (such as clinical assessments or risk profiles) is available. This creates a fundamental tradeoff between committing early based on aggregate information and deferring decisions until more precise information is revealed. The challenge is amplified in settings where routing decisions have downstream effects on congestion, and the benefit of adjustment to early decisions becomes unclear. In this work, we develop a dynamic routing model that explicitly captures delayed information and long-term congestion effects. We show that under full information, the optimal policy admits an exact two-step structure: an aggregate planning step followed by a priority-based assignment. This motivates a two-step approach in the delayed-information setting, where we leverage fluid approximation to capture first-order congestion effects. Our analysis characterizes both the necessity and the magnitude of adjustment, which reveals a three-way tradeoff among the value of information, the cost of adjustment, and the intensity of long-term congestion. We illustrate the framework using a case study of jail diversion programs calibrated with ten years of operational data. The results show that ignoring long-term congestion can lead to overly reactive policies that destabilize the system and perform worse than a simple no-adjustment policy. This is a joint work with Xiaoquan Gao at Singapore Management University.

GenAI-powered shopping agents (e.g., GPT, Claude, and Gemini) automate product evaluations, comparisons, and transactions, releasing consumers from the burdensome shopping task and also improving product matching accuracy. Nonetheless, the new shopping model triggers a new strategy of generative-engine optimization (GEO), through which a seller creates content for GenAI engines to bias AI decisions. This study develops a game-theoretic model to examine how agentic commerce affects consumer buying, seller strategies, and welfare implications, generating insights for proper AI governance.

We obtain three main results.

(1) When consumers delegate to AI shopping agents, the seller adjusts its strategy in two ways—investing in GEO to bias AI decisions and shifting from a demand-oriented to a margin-oriented pricing. Both strategy adjustments enhances seller profit but harm buyer surplus from transactions.

(2) Consumer sophistication about seller GEO and associated AI decision bias curbs a seller’s incentive to implement GEO for moderate- or high-quality products, even when GEO is costless, and also discourages the seller from increasing price. As such, both harms are alleviated. Interestingly, when AI shoppers are sophisticated, an increased AI delegation rate makes the seller reduce, but not enhance GEO efforts. This result highlights the importance of AI literacy for improving consumer welfare.

(3) Sellers offering moderate-quality products benefit more from shoppers’ AI delegation than those offering low- or high-quality products, revealing a “mid-quality advantage.” Correspondingly, total buyer surplus drops the most with mid-quality products, showing a “mid-quality surplus trap.” This result suggests that policy oversight and intervention should be selectively allocated.