On-demand restaurant meal delivery platforms, such as DoorDash and Meituan, have recently introduced a multi-order delivery service: Customers may combine delivery requests from different restaurants in a single multi-order with the service promise of a synchronized delivery. However, the platform must not only ensure the synchronization of multi-orders but also improve punctuality and freshness for all customers. This is challenging because, as we show, synchronization, delay, and freshness are conflicting objectives. Uncertainty in the delivery process and unknown future orders further complicate the decision making. This raises several research questions: How does the introduction of a multi-order service affect the overall delivery operations with regard to service quality and operational expenses? How should a multi-order service be strategically rolled out? How can we balance the competing objectives of synchronizing deliveries while minimizing delay and maximizing freshness? To answer the research questions, we propose an effective policy that allows for a careful and controlled balance between the competing objectives and employ it in an extensive computational study. We evaluate the effect of different trade-offs between delay, freshness, and synchronization on delivery operations over varying demand for multi-orders. We observe that enforcing strict synchronization of multi-orders by assigning each multi-order to a single delivery driver is hardly operational feasible. Occasionally using split-deliveries provides the flexibility to better balance all objectives. Our detailed experiments further generate insights on how platforms may roll-out multi-orders as a new service offering without negatively affecting their existing delivery operation while benefiting from reduced operational expenses. • We rewrote the abstract to match the narrative of the introduction. It now focuses on the managerial insights on how to roll-out a multi-order service. • We rewrote the methodology section. In the updated manuscript, we now provide further explanations for the sake of reproducibility as well as an example at which we explain the policies. • We greatly extended our computational study by adding three new experiments. They analyze the policies’ fleet utilization, their operational costs, and their robustness to bad weather or similar negative scenarios.
Florentin D. Hildebrandt (Thu,) studied this question.