Optimising centralisation in distribution networks for perishable products through mathematical modelling, parametric analysis, and machine learning

The success of distribution companies for perishable products is enabled by optimally configuring distribution networks, which allows for reducing total logistic costs while ensuring reduced product spoilage and high service levels. Since customer demand for perishable products varies over time, the network configuration should not be optimised once, but periodically reviewed. Among the decisions to be reviewed, determining whether to centralise or decentralise inventory (i.e. stock allocation in distribution centres) is crucial. However, the literature overlooks stock allocation decisions, and existing methodologies to compare the economic performance of centralised, decentralised, and hybrid policies neglect important cost items, also requiring advanced computational technologies and skills to be applied. This paper addresses these gaps by providing two contributions. First, a novel mathematical model is offered to compare five stock allocation policies (ranging from centralisation to decentralisation, crossing through three hybrid policies), identifying the most cost-effective one under a comprehensive economic analysis. Next, a parametric analysis is accomplished and a machine learning algorithm is trained to obtain a quick and easy-to-use Decision Support System (DSS). The DSS results in a decision tree, which was tested in a case study and provided managerial insights on how to review the stock allocation of perishable products.