Faster Deliveries and Smarter Order Assignments for an On-Demand Meal Delivery Platform

Abstract

The rapid growth of on-demand meal delivery platforms has heightened competition, making customer retention a critical priority. While prior research on order dispatch algorithms has largely focused on minimizing delivery time or delay, the direct impact of delivery performance on repeat purchases remains underexplored. Using transactional data from an online meal delivery platform in China, we empirically investigate the asymmetric effects of early and late deliveries on customer repurchasing behavior. To address potential endogeneity, we introduce driver experience and local knowledge, two previously overlooked factors in platform algorithms, as novel instrumental variables. The survival analysis shows that late deliveries significantly reduce future orders, while early deliveries provide only limited benefits. Guided by these empirical insights, we develop a simulation-based evaluation of different order dispatch algorithms, revealing that maximizing future orders, rather than minimizing delivery time or delays, yields the highest future orders. These insights offer actionable recommendations for platform managers, stressing the importance of strategic adjustments in dispatch algorithms and integrating heterogeneous treatment effects into algorithmic design. By merging operational delivery performance with consumer behavior insights through causal inference and optimization, this study provides a novel end-to-end framework for creating data-driven dispatch algorithms that enhance both service efficiency and customer retention.