Service Time Window Design in Last-Mile Delivery

TL;DR

This paper develops models for designing reliable service time windows in last-mile delivery, balancing risk, route efficiency, and customer satisfaction, using stochastic and distributionally robust optimization techniques.

Contribution

It introduces two novel optimization frameworks for time window design that incorporate risk preferences and integrate routing, with closed-form solutions and practical implementation insights.

Findings

Distributionally robust model maintains service guarantees within risk tolerance.

Stochastic model occasionally violates in out-of-sample tests.

Frameworks are compatible with existing routing solutions.

Abstract

Our study focuses on designing reliable service time windows for customers in a last-mile delivery system to boost dependability and enhance customer satisfaction. To construct time windows for a pre-determined route (e.g., provided by commercial routing software), we introduce two criteria that balance window length and the risk of violation. The service provider can allocate different penalties reflecting risk tolerances to each criterion, resulting in various time windows with varying levels of service guarantee. Depending on the degree of information available about the travel time distribution, we develop two modeling frameworks based on stochastic and distributionally robust optimization. In each setting, we derive closed-form solutions for the optimal time windows, which are functions of risk preferences and the sequence of visits. We further investigate fixed-width time windows, which standardize service intervals, and the use of a policy that allows vehicles arriving before the lower bounds to wait rather than incur a penalty. Next, we integrate service time window design with routing optimization into a unified framework that simultaneously determines optimal routing and time window allocations. We demonstrate the efficacy of our models on a rich collection of instances from well-known datasets. While a small portion of the time windows designed by the stochastic model was violated in out-of-sample tests, the distributionally robust model consistently delivered routes and time windows within the service provider's risk tolerance. Our proposed frameworks are readily compatible with existing routing solutions, enabling service providers to design time windows aligned with their risk preferences. It can also be leveraged to produce the most efficient routes with narrow time windows that meet operational constraints at controlled levels of service guarantee.