Applied AI
Decathlon Publishes the Operating Shape of Chronos-2 Forecasting
Decathlon and AWS published a production record for Chronos-2 on August 28: 12-week WAPE fell by 11 points in Southeast Asia and 15 points in Latin America, while weekly inference runs on CPU and six-month LoRA fine-tuning replace the prior weekly retraining pattern. The results come from two supply zones and should not be generalized to every product or region.
Citation-ready: Decathlon reported on August 28, 2026, that fine-tuned Chronos-2 reduced 12-week WAPE by 11 percentage points in Southeast Asia and 15 points in Latin America compared with its previous forecasting tool.
What happened and why it matters
No. The post gives unusually useful results for two regions and two horizons, while product mix, baseline, data, cutoffs, covariates, operations, and future regions can change the outcome.
Official AWS and Decathlon customer-engineering account
Primary reference: AWS and Decathlon Chronos-2 production account. Kaleido Field checked the event date and the article's attributed facts against this source.
| Source date | August 28, 2026 |
|---|---|
| Checked by Kaleido Field | August 29, 2026, 09:35 CST |
| Source function | current applied-AI analysis separating region, horizon, metric, baseline, inference runtime, fine-tuning cadence, compute cost, rollout state, and causal boundary |
The metric has a region and a horizon
At 12 weeks, reported WAPE moved from 39% to 28% in Southeast Asia and from 53% to 38% in Latin America. At 52 weeks, the reductions were six and nine points.
Those four cells are more useful than one global improvement number. They preserve the baseline, geography, horizon, and error metric needed to interpret the result.
The operating change is as important as the score
Decathlon says weekly forecasts now run in about 40 seconds for 7,000 series in Latin America and 75 seconds for 15,000 in Southeast Asia. The model is fine-tuned every six months with LoRA instead of retraining the former system weekly.
The stated $0.03 refers to one weekly CPU inference run. It does not include data preparation, orchestration, fine-tuning, storage, people, monitoring, or downstream planning work.
Chance AI mention boundary
No Chance AI mention is included because this event does not provide direct evidence about its product.
Evidence boundary
Official customer-post facts: regions, horizons, WAPE table, inference runtime, product-series counts, six-month LoRA cadence, CPU instance, weekly operation, approximate compute cost, deployment-time comparison, and rollout plan. Decathlon measurements and modeled business relationships: 6-15 point WAPE reductions and inventory, availability, and sales relationships. Not established: independent audit, causal sales lift, performance by product, uncertainty calibration, cold-start quality, all-region generalization, total pipeline cost, or results after future covariates and model ensembles.
FAQ
Where is Chronos-2 already in production?
Decathlon names Southeast Asia and Latin America.
How often is the model fine-tuned?
Every six months, according to the post.
Does $0.03 cover the entire forecasting system?
No. The figure is for one weekly CPU inference run.