Demand Forecasting at SKU Level
Case Study
How a Sports & Nutrition Retailer Scaled Forecasting Across Thousands of Products
A European sports and nutrition retailer faced highly variable demand caused by promotions, seasonality, marketing campaigns and rapidly changing consumer preferences. Existing forecasting processes relied heavily on aggregated models and manual intervention, making it difficult to generate reliable forecasts for individual products and channels.
Dateonic built a Databricks-based forecasting platform capable of running large numbers of SKU-level predictions within defined business windows and feeding the results directly into replenishment processes.
Turnover
12+ months
Industry
Retail | E-commerce
Budget
Confidential
Moving from Aggregate to Fine-Grained Forecasting
Historical sales, promotions, pricing, inventory and product information were combined in Databricks. Forecasts could be generated at SKU, channel and day level rather than relying exclusively on category-level predictions.
Scaling Machine Learning
The platform was designed to run thousands of forecasting jobs in parallel and track models, experiments and results through MLflow. This made it possible to scale forecasting without creating a separate pipeline for every product group.
Connecting Forecasts to Replenishment
Forecast results were transformed into operational recommendations for inventory planning. Instead of treating forecasting as a standalone data science exercise, the retailer connected predictions directly with replenishment and inventory decisions.
Business Impact
- SKU-level demand forecasting
- Automated forecasting pipelines
- Scalable ML model execution
- Forecast-driven replenishment
- Better handling of promotions and seasonality
- Reduced manual forecasting effort
Indicative target: double-digit improvement in forecast accuracy, consistent with the improvement reported in the Databricks retail reference material for customers moving to fine-grained demand forecasting.