LO
DeploymentCognitive automation2025-12-22
In a field deployment at JD.com, an inventory model cut turnover by 5.27 days, raised in-stock rates by 2.29% and cut holding costs by 29.95%, on items chosen for high automation rates
Supply chain planneroccupation page →
Event date / reported
2025-12-22
Evidence stage
DeploymentAn employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Tasks this bears on
Replenishment and inventory
Setting safety stock and reorder quantities, deciding what to replenish where, and moving excess stock to where it is needed.
Automating✓ Evidence-backed
Where this applies
China, JD.com. A replenishment model validated through a field deployment with a difference-in-differences analysis, reporting a 5.27-day reduction in turnover, a 2.29% increase in in-stock rates and a 29.95% decrease in holding costs against incumbent practice. The paper defines the automation rate as the percentage of replenishment decisions that are output by models and adopted for execution, and says treatment items were chosen by JD's managers for high automation rates. Several authors work for JD.com, which has a stake in the result.
What this means
A large retailer measures how many replenishment decisions go out untouched by people, and builds its trials on the items where that share is high. For routine replenishment, the decision itself is moving to the model.
What it does not yet show
A company-co-authored study of one retailer; it does not report planner headcount.
What you can check
Open arXiv 2512.19001 and find "a 5.27-day reduction in turnover" in the abstract.
Does it change the assessment?
No. The impact index is never moved by a single event. Nor did this record change a layer: all 1 linked judgement above already rested on earlier evidence. This one adds to them.
Source
Zhao, Yu, Qi et al. (Tsinghua University, JD.com Supply Chain Tech, University of Hong Kong) — ORPR: An OR-Guided Pretrain-then-Reinforce Learning Model for Inventory Management, arXiv 2512.19001 (submitted 22 Dec 2025; v2 6 Jan 2026) · verified 2026-09-29 · Claude (VOLO agent) · interpreted 2026-09-29 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.
This record is cited in