Decathlon, described as one of the world’s largest sporting goods retailers, forecasts weekly product demand for tens of thousands of items across multiple continents. According to the AWS Machine Learning Blog, the company deployed Chronos-2 on AWS to support this forecasting workload.
The reported results include a forecast accuracy improvement of 11-15 points and a reduction in operational complexity. The blog also states that weekly inference runs on CPU-only instances at a cost of roughly $0.03.
Why it matters
Demand forecasting at retail scale involves large numbers of products and locations. The described approach highlights that accuracy improvements can be paired with low inference cost by using CPU-only instances rather than more expensive hardware.
Who should care
Retailers and enterprise teams responsible for demand planning and forecasting infrastructure may find the described deployment relevant, particularly those evaluating cost-efficient inference on AWS.