Alibaba aligned forecasting, stock and pricing across every store channel

Alibaba's business spans dozens of retail models, from mobile apps to brick-and-mortar grocery, and each has different demand patterns, shelf lives and inventory rules.

Traditional forecasting and inventory tools miss the interactions that matter: pork price moves beef sales, and chips move salsa. Alibaba used deep learning to capture those interactions, simulation optimization to model how customers pick stock, and optimization to set prices and recommendation ranks in real time.

The result was higher forecast accuracy, less shrinkage and fewer out-of-stock items, translating to more than $100 million a year in lower costs and better service across its retail businesses. The work was a finalist for the INFORMS Franz Edelman Award in 2022.

Forecast, stock and price each channel as one system

A shared demand model lets buying, pricing and placement respond to the same forecast instead of three guesses.

Simulation captures the FIFO and LIFO mixes that break closed-form inventory rules.

Real-time price and recommendation optimization turns short shelf life into a lever rather than a loss.

Better forecasts and fewer stockouts raise revenue and cut waste at the same time.

Retail is one decision problem, so a shared demand model beats disconnected forecasts, ordering and markdown tools.

The suite was rolled out across Alibaba's omnichannel retail subsidiaries, where the company said both shrinkage and out-of-stock rates fell steadily. It was a 2022 Franz Edelman finalist and the underlying methods were published in the INFORMS Journal on Applied Analytics.