Data-Driven Reserve Personnel Placement to Balance Operation Default Risk and Resource Utility

Abstract

In last mile delivery, to manage delivery default risks and ensure delivery completion, reserve personenel are placed. This is due to driver procurement having to be planned and executed about one-month ahead, when delivery demands could only be roughly predicted. Although reserve drivers occasionally work as final defense, it regularly lowers driver utility, and a method to place reserve drivers balancing delivery default risk and driver utility is required. Previous work tackled this problem by stochastic staffing problem approaches, but there existed a limit in feature modelling and result interpretability, which created a gap in algorithms and procurement manager decision making. The proposed method aims to fill this gap, by taking a transdisciplinary approach of traditional scheduling, probability modelling, and explainable AI. In doing so, a flow of first creating a staffing schedule based on fixed staffing number demands, and then determining a fixed number of reserve personnel required for each staffing window, was designed. A probablity distribution of required personnel number per delivery is calculated in doing so, and this distribution is used as a easy to understand decision support tool for delivery managers. Through a case study using delivery demand data of a Japanese EC-logistics company, the proposed method was shown capable of lowering reserve drivers, with having a high potential of no delivery defaults.