Abstract Food delivery platforms increasingly employ hybrid workforces combining hourly-paid in-house couriers with per-delivery crowdsourced couriers. A key operational challenge is that crowdsourced couriers may drop out of the platform if they remain idle for too long, so that dispatch decisions affect not only immediate service quality but also future delivery capacity. We formalize this as a sequential decision-making problem and propose a rolling-horizon mixed-integer programming (MIP) dispatch policy with scenario-based lookahead. The MIP incorporates a drop-out constraint that links assignment timing to crowdsourced courier retention, creating a structural incentive to dispatch crowd couriers before they leave the platform. Simulation experiments calibrated with operational data from a Tokyo food delivery platform show that the proposed policy substantially reduces lateness relative to myopic bipartite matching, particularly when the platform relies heavily on crowdsourced capacity. Analysis of the results indicates that the primary mechanism is the early activation of crowdsourced couriers to maintain the available pool during peak demand, and that the policy is robust to misspecification of the drop-out parameter and does not require highly accurate demand forecasts.
Takazawa et al. (Thu,) studied this question.