Séminaire hybride au GERAD et Zoom.

Same-day delivery is reshaping e-commerce, but faster, near-free delivery leads customers to split purchases into smaller, more frequent orders, raising fulfillment cost per dollar sold and straining fleet operations. Free-shipping thresholds are a natural lever, since many customers add items to qualify for free shipping. We study how to design these thresholds dynamically in same-day on-demand delivery. Because each accepted order changes the marginal routing cost of serving every other, pricing and routing are tightly coupled. Unlike prior work that prices delivery fees independently of basket-size incentives and fleet state, we let the platform offer a two-tier menu, a free-shipping threshold together with a lower discounted threshold, both adjusted in real time as a function of system state. Customer behavior follows a choice model with heterogeneous types. We formulate the problem as a profit-maximizing Markov decision process, with routing handled by a real-time insertion heuristic whose costs feed back into pricing, and learn the policy via reinforcement learning. Across calibrated instances, the dynamic two-tier menu outperforms static thresholds and revenue-maximization baselines, with the largest gains under peaked demand and tight capacity.

(Joint work with Yossiri Adulyasak and Jean-François Cordeau)