Districting in last-mile delivery with stochastic customers

Due to the peculiarities of city streets, last-mile logistics is typically organized using territory-based routing approaches, which divide the city into a set of districts and assign drivers to deliver in one or more of them. This allows drivers to develop a deep understanding of the characteristics of each district, and clients benefit from consistent service. However, these advantages must be carefully weighed against the flexibility of daily customer assignments, which enable planners to maximize driver utilization and minimize routing costs. In this paper, we propose a new holistic framework for defining districts, considering the impact on the quality of logistics decisions and fleet capacity usage. Specifically, we address how districting decisions affect the demand distribution within each district. Computational experiments on both simulated and realistic instances demonstrated a significant reduction in costs compared to benchmark techniques.