Decathlon has deployed Amazon’s Chronos-2 time-series model to forecast weekly demand for tens of thousands of products. The sporting-goods retailer says the system improved forecast accuracy by 11 to 15 points while reducing operational complexity.
The forecasts cover two planning windows: 12 weeks for replenishment orders and 52 weeks for longer-term stock and capacity decisions. Each regional supply zone can include as many as 25,000 products, ranging from highly seasonal ski equipment to surfboards. The system currently serves several continents and runs every week.
According to the companies, Chronos-2 can perform each weekly inference run for about $0.03 on CPU-only AWS instances. That removes a need for GPU capacity during routine operation, although the post does not provide a full accounting of data preparation, engineering, storage, or model-management costs.
The results are a customer case study co-written by AWS and Decathlon, not an independent benchmark. Performance may differ with another retailer’s data and demand patterns. The deployment is still a useful example of a foundation model moving into a conventional supply-chain task where cost, repeatability, and long forecasting horizons matter more than conversational output.