A Long-Term Demand Forecasting Model Implementation Case Study with a Major Retailer

Author(s): Alexandre Warembourg

Originally published on Towards AI.

Explore how I developed a core demand forecasting algorithm for ten countries, dealing with an average product sales history of 11 months and 30% new products in each batch.

source : Dall-E generated picture from Author Prompt

Auchan Retail International’s World Wide Product Direction (DPW) presented us with a challenging task: Forecasts were required for all Own Brand Products in several countries, including Spain, Romania, France, Portugal, Luxembourg, Poland, Taiwan, Russia, China, and Hungary. The forecasts needed to cover each product code (EAN) and store format, extending 24 months into the future. This long-term forecast was crucial for planning next year’s product assortment and managing supplier orders.

Our goal was to assist planners with a wide array of products, improving forecast accuracy beyond their existing method. Previously, they relied on historical Weekly Mean per Product (VMH) and similar product codes (mirror EANs) for insufficient historical data.

Forecasting thousands of products using Excel formulas has limitations due to the complexity of the data and the vast number of products and countries involved. Therefore, machine learning was the right choice.

Our journey led us to a pivotal decision — forecasting at the individual store level, a stark departure from the broader store format level. Why? Aggregated data… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI