How Decathlon runs demand forecasting at scale with Chronos-2
Decathlon cut forecast costs to $0.03/week per product line using Chronos-2 on CPU. 11-15 point accuracy gain without the GPU tax.
Why it matters
A real deployment case study showing how practitioners can ship time-series ML at scale using open models and commodity hardware—practical ROI for operations teams considering similar workloads.
The key facts
12 to knowDecathlon: tens of thousands of products, multi-continent demand forecasting
Model: Chronos-2 (time-series foundation model)
Accuracy improvement: 11-15 percentage points
Infrastructure: AWS CPU-only instances
Cost: ~$0.03 per week for inference
Frequency: weekly inference cycles
Benefit: reduced operational complexity vs. legacy forecasting
Decathlon: tens of thousands of SKUs, weekly forecasts, multiple continents
Chronos-2 model deployed on AWS
Weekly inference cost: ~$0.03 on CPU-only instances
Operational complexity reduced
Time-series forecasting use case (demand planning)
Go to the source
AWS Machine Learning Blogaws.amazon.com
Publisher excerpt: Decathlon, one of the world's largest sporting goods retailers, forecasts weekly demand for tens of thousands of products across multiple continents. Learn how they deployed Chronos-2 on AWS to improve forecast accuracy by 11-15 points while cutting operational complexity and running weekly…