The encyclopedia · Engineering & Operations · Operational decision · 2020–2024
JD.com optimized what each front warehouse stocks and how it is refilled
JD's two-tier network faced a tradeoff of what to stock at small front centers and how to refill them; a joint algorithm lifted local fulfillment.
JD.com
the move
JD.com promises fast delivery, so most items sit close to the customer in small front distribution centers, but each FDC can store only a limited set of SKUs.
The tension is real: stock the wrong SKUs and you fulfill from the far regional center, which is slow, costly and prone to lost sales; and the refill decision repeats every day.
JD's teams built a joint framework: assortment algorithms (ML-Top-K and Reverse-Exclude) pick each FDC's SKUs, and an end-to-end forecasting-plus-optimization-plus-simulation framework sets replenishment and target inventory.
Deployed across JD's China network it improved local order fulfillment and satisfaction; the work won the 2024 INFORMS Daniel H. Wagner Prize, with the Interfaces paper reporting fulfillment rate up 0.54% and demand satisfaction up 1.05%.
why it works
- An FDC's tiny assortment is the single biggest lever on whether an order is served locally.
- Assortment and refill are coupled, so separate decisions create stockouts and waste.
- Forecast uncertainty and long-tail demand make heuristic stocking unreliable.
- A joint, self-learning model is explainable and fast enough for millions of SKUs.
what transfers
In a two-tier network the expensive mistake is optimizing the warehouse and the replenishment separately. Decide what is on the shelf and how it is refilled together, against forecast and uncertainty.
what came after
The algorithms went live across JD's network, improving on-shelf availability and cutting stockout loss, and won the 2024 INFORMS Daniel H. Wagner Prize.
references
- JD.com Improves Fulfillment Efficiency with Data-driven Integrated Assortment Planning and Inventory Allocation
- 京东供应链创新与实践：应用数据驱动的库存选品和调拨算法提升履约效率
spotted an error? The archive wants to know.