Decathlon Selects Chronos-2 Model for Global Supply Chain
One of the world's leading sportswear retailers has deployed the Chronos-2 system on AWS infrastructure to enhance its demand forecasting capabilities.
Global retail giant Decathlon has chosen Chronos-2 as its primary forecasting component following evaluations among numerous time-series models. Thanks to the new architecture built on AWS, the company has begun executing large-scale demand forecasts for thousands of products and operations across multiple continents.
Details of the Forecasting System
The company's new forecasting system predicts weekly sales quantities across a 12-week replenishment window and a 52-week strategic horizon. This structure plays a critical role in overcoming supply chain challenges spanning tens of thousands of products and multiple continents.
AWS and Modern Technologies
The production deployment utilizes Amazon EC2 instances for batch inference, Databricks jobs, and Airflow orchestration. Additionally, LoRA-backed AutoGluon is preferred for fine-tuning every six months, and MLflow is used for model registry.
Achieved Benefits
Following comparisons with previous methods such as Amazon SageMaker AI DeepAR and Temporal Fusion Transformer, the fine-tuned Chronos-2 was determined to deliver significantly superior performance. This integration has driven major improvements in forecast accuracy, inventory savings, product availability, and operational efficiency.