Demand Forecasting for Omnichannel Retail

The Challenge

A fast-growing omnichannel retailer with 200+ stores and a rapidly expanding e-commerce presence was struggling with demand forecasting accuracy. Traditional forecasting methods based on historical averages could not account for channel-specific patterns, promotional effects, and regional variations. This led to chronic overstock in some categories and stockouts in others — eroding margins and customer satisfaction.

Our Approach

DataLumin partnered with the retailer's supply chain and merchandising teams to build an ML-driven demand forecasting system. We started with an extensive feature engineering phase, incorporating historical sales, promotional calendars, weather patterns, regional events, and channel-specific demand signals. The model was trained on three years of SKU-level transaction data across all channels.

The Solution

The final solution uses an ensemble of gradient boosting and time-series models, deployed on Azure ML and integrated directly into the retailer's replenishment system. Forecasts are generated daily at the SKU-store level for a 14-day horizon, with weekly forecasts extending to 90 days for procurement planning. A Power BI dashboard allows merchandising teams to review forecast accuracy, override predictions when needed, and track forecast-vs-actual performance.

Results & Impact