The challenge
A key retail client with more than 2,000 SKUs needed to give accurate sales forecasts to the vendors who supply them. Inaccurate forecasts led either to lost sales or to overstocking.
What we did
- Built a time-series forecasting model in SAS.
- Accounted for seasonality, holidays, trends, special events and promotions.
- Forecast new SKUs by studying the sales history of similar sister SKUs.
The result
Accurate forecasts, combined with a sound inventory planning model, significantly reduced both stock-outs and overstocking.
Higher sales and lower stock-holding costs increased the company's net margins.
How we would build it today
- Global machine learning forecasting models trained across all SKUs, with hierarchical reconciliation from SKU to category.
- Probabilistic forecasts that give a range, not a single number.
- Automated retraining and a vendor-facing forecast portal.
This describes our current toolkit for the same problem. It is not part of the original engagement.