Performance Evaluation and Explainability of Last-Mile Delivery

Abstract

The demand for last-mile delivery (LMD) services worldwide increased following online sales growth, so better methods to assess efficiency issues are paramount. This work explores a data-driven approach to evaluate LMD services and inform logistics service providers about possible improvement directions. It uses multi-directional efficiency analysis to benchmark LMD services based on process variables, such as delivery time and service cost. Then, by fitting machine learning models and using explainability algorithms with new metrics, characterizes factors that influence LMD performance. Early discussions with experts show that the approach produces understandable and integrable results that generate valuable insights, e.g., regarding the impact of each variable on service quality informing the direction for further improvement action.