Anticipative Dynamic Slotting for Attended Home Deliveries

Abstract

For attended home deliveries, customers expect narrow delivery time slots that fit their personal schedules. This severely limits the solution space for delivery planning and thereby makes attended home deliveries costly and complex for retailers. As one remedy, dynamic slotting lets the firm control the offered time slots per arriving customer. To that end, solution approaches have to quickly compare the immediate reward from accepting an order to the opportunity cost of thereby reducing the delivery capacity in the selected time slot. As the opportunity cost depend on a complex vehicle routing and scheduling problem under demand uncertainty, they are notoriously difficult to quantify. To quickly compute good solutions, we present approaches that let an extensive, simulation-based preparation phase inform online decision making. These approaches estimate opportunity cost via approximate value function models and rely on delivery schedule anticipation. Specifically, we propose to aid the process of dynamic slotting by generating spatio-temporal delivery patterns via team orienteering on sample arrival streams. In an extensive computational study, we compare approaches on both synthetic and empirically-validated demand scenarios. Rather than suggesting a one-size-fits-all view, we propose a differentiated set of recommendations for selecting methods dependent on the specific problem scenario. In particular, we find that benefits from relying on anticipated patterns highly depend on demand segments' basket value and location distributions.