Using machine learning, NLP, and deep domain knowledge, Auchan Retail International achieved an impressive 18% reduction in out-of-stock items and overstock across national operations in just one year. Their dual-model strategy, extensive feature engineering, and close collaboration with stakeholders led to substantial operational improvements and efficiency in retail forecasting.
Case Study: AI-Powered Demand Forecasting for Retail Giant
Introduction
In the fast-paced world of retail, Auchan, a global leader, grappled with the challenge of accurately forecasting demand for promotional items. Leveraging machine learning, NLP, and deep domain expertise, we achieved an impressive 18% reduction in stockouts and overstock across national operations within just one year.
The Challenge at Hand
Forecasting demand for promotional items in every store was a critical challenge for Auchan. The goal was to align supply with ever-changing customer demand to avoid surplus inventory and ensure customer satisfaction.
Practical Solutions
Our approach involved crafting a forecasting model adaptable across diverse countries with minimal changes. Leveraging available promotional pricing, display and dates, we designed features specifically for promotions to capture every aspect.
Crafting the Solution: A Dual-Model Strategy
We opted for a machine learning algorithm due to its ability to handle a wide range of products and external factors such as promotions. Utilizing LightGBM, we employed a dual-model strategy to effectively manage erratic promotion patterns and maintain stability.
The Path to Precision: Selecting the Right Metrics
Working closely with demand planners, we defined the scope of evaluation and used metrics such as WMAPE, bias, and accurate operations at a certain threshold. These were computed through a thorough backtest to verify the algorithm’s robustness.
Overcoming Obstacles: Data Preprocessing
Data preparation proved essential, including reconstructing promotions from catalog datasets and refining promo mechanics. Through this process, we addressed pricing anomalies and factual shortages, significantly improving forecast accuracy.
Measurable Success: The Impact
Our AI-powered model not only outperformed traditional forecasting methods but also led to a 15% improvement over previous demand planner forecasts. This resulted in over 30,000 hours saved annually and an 18% reduction in overstock and shortages, generating a profit of $100,000.
Conclusion
This case study demonstrates the transformative power of AI in retail forecasting. By embracing data-driven strategies, substantial operational improvements and efficiency gains can be achieved.
For more insights into leveraging AI for retail optimization, feel free to reach out to us at hello@itinai.com or visit our website.
