Understanding order cancellation behavior in on-demand delivery services

The rise of digital platforms has transformed urban mobility by offering a wide range of on-demand transportation and delivery services, such as ride-hailing, grocery delivery, and food delivery. Consumers using these services typically experience two key waiting stages: waiting online for matching and waiting physically for driver pick-up. A common consequence of waiting is order cancellation, which not only disrupts platform operations but also generates inefficient vehicle movements and puts strain on urban road networks. This paper examines the dynamics of order cancellations in the two stages and their interactions by using a two-stage survival analysis combined with a Heckman correction model. Based on a dataset of delivered and cancelled orders from an on-demand food delivery platform in Asia, we reveal several key findings. First, we identify an asymmetric effect of accumulated waiting time on cancellations before and after the expected waiting time, which provides evidence to the existence of reference-dependence preferences in on-demand services. Second, while higher delivery fees reduce cancellations in the matching stage, they increase the risk of orders being cancelled shortly after they are matched to drivers. Third, the risk of cancellation is significantly reduced when the order is delivered by a familiar driver. Lastly, we reveal how the risk of order cancellation varies with waiting time in both the matching and pick-up stages. These findings provide valuable insights for optimizing pricing and matching strategies to mitigate order cancellations, enhance the efficiency of on-demand services, and reduce inefficient trips caused by cancellations.