Optimizing warehouse product allocation at JD.com: A community detection-driven approach

In e-commerce platforms, multi-item orders often require split shipments due to inventory distribution across geographically dispersed warehouses. This order splitting results in increased shipping costs and excessive packaging material consumption. The associated order splitting optimization problem presents significant complexity arising from the combinatorial nature of the order structure. Through collaboration with JD.com, China’s largest online retailer, we analyze the real historical orders and inventory data from one distribution center. To minimize order splits, we develop a novel K-Community Swap Algorithm (KCSA) incorporating community detection techniques. Comparative benchmarks demonstrate that KCSA achieves solutions within 5% of optimality against Gurobi’s exact solutions for small-scale instances while delivering superior solution quality in significantly less computational time than existing methods for medium-to-large-scale instances. The implications in JD.com reduce 440,000 order splittings in five distribution centers and save 2.6 million RMB packing material annually. The annual carbon dioxide equivalent emissions saved are approximately 132 tons.