Supporting warehouse management through Machine Learning

Warehouse Management System (WMS) software and integrated Enterprise Resource Planning (ERP) are widespread tools for modern operations management. Alongside ensuring smoother warehouse management, these solutions collect data regarding products inflows and outflows, too. These data can help daily management and be used to enhance supply chain performance for more strategic management if explored systematically. In this context, this manuscript suggests a Machine Learning (ML) methodology based on historic warehouse WMS and ERP transactions to identify hidden patterns from the combination of different events. Historic data are used to feed two ML solutions: (i) a saliency-based anomaly detection algorithm; (ii) a hierarchical clustering algorithm. The former spots anomalous operative time frames by analyzing a warehouse Key Performance Indicator (KPI). The latter is used to group sets of similar operations. Their joint usage allows calculating an error propension metric, which gives hints on the expected tendency of a set of transactions to contribute to lowering warehouse management performance. The methodology has been designed as a decision support tool for inventory managers to help them prioritize critical resources, organize properly warehouse movements, and set diverse layout solutions for logistic operations. The analytical steps of the methodology are exemplified in a demonstrative use case in an industrial warehouse.