Decathlon, one of the world’s largest sporting goods retailers, has implemented Chronos-2 to enhance its demand forecasting capabilities. The company, which serves over 400 million users worldwide, relies on accurate forecasting to ensure product availability in its stores. After evaluating several time series foundation models, Decathlon selected Chronos-2 as a core component of its forecasting system. This decision was driven by the model’s ability to deliver higher accuracy with reduced operational complexity, addressing the challenges of forecasting demand for tens of thousands of products across multiple regions.
Decathlon’s demand forecasting system predicts weekly sales for products on two critical time horizons: a 12-week replenishment window and a 52-week strategic horizon. These forecasts are essential for inventory management and capacity planning. The system runs weekly and is deployed across multiple supply zones, including Europe, India, China, South East Asia, and Latin America. Each zone covers up to 25,000 products, with plans to expand to the Middle East and Africa. The forecasting system’s scalability and efficiency are critical to supporting Decathlon’s global operations.
Decathlon’s previous forecasting approach used a hybrid model combining Amazon SageMaker AI DeepAR for short-term forecasts and Holt-Winters exponential smoothing for long-term projections. While effective, this method required weekly retraining and posed challenges in scaling to new regions. The introduction of Temporal Fusion Transformer (TFT) improved long-term accuracy but still had operational limitations. These challenges prompted Decathlon to explore more efficient and scalable solutions, leading to the adoption of Chronos-2.
Source: awsml
