Data-Driven Production and Operations Management Decision-Making

The talk begins with an overview of the challenges in extracting real value from large‑scale operational data and introduces a three‑layer framework—data collection, analytical modeling (including statistics, machine learning, and econometrics), and decision optimization (using operations research and game theory).

It then examines promotion analysis, highlighting the need to choose appropriate metrics, the pitfalls of naïve sales growth interpretation, and the importance of accounting for substitution effects, reference price theory, and Simpson’s paradox when evaluating conversion rates.

Next, the speaker discusses the “small data in big data” problem of long‑tail product pricing, describing a semi‑automated pricing system that groups low‑volume SKUs, applies robust optimization, and incorporates user perception differences between online and offline channels.

Finally, the session presents a case study of an unmanned warehouse management system for JD.com, detailing how a three‑dimensional matching problem (AGV, shelves, workstations) is decomposed via Lagrangian relaxation into tractable sub‑problems, achieving 3‑40% solution quality improvement over CPLEX and enabling real‑time decision making.

Throughout, the speaker emphasizes the integration of big‑data analytics, machine‑learning models, and optimization algorithms to enhance decision quality while maintaining operational speed.