[UTMD-115] Exposure Design for Two-Sided Platforms (by Kei Ikegami)

Kei Ikegami

Many high-stakes matching platforms still rely on human intermediaries, resulting in inconsistent decisions, high operating costs, and a bias toward high-probability matches. This paper replaces such heuristics with data-driven exposure design. I develop a two-sided sequential-search model in which the platform controls pairwise meeting propensities. I show that maximizing short-run flow surplus is dynamically inefficient: prioritizing top pairs too early causes dynamic cannibalization, which reduces future search options for remaining users. As an alternative objective for expanding and sustaining the user base, I consider long-run user value, defined as the aggregate continuation value of search. I characterize and compute the optimal exposure rule under this objective via entropic regularization and Bregman–Dykstra projections. In a doctor–spot-job platform, counterfactual simulations that replace current policies with the computed optimum reveal that existing rules over-penalize distance. Welfare gains arise primarily from correcting under-exposure of viable matches and expanding users’ effective option sets, not from mere reshuffling within a fixed exposure volume.