Scope
- E-commerce platform Forecast at monthly and weekly level including price changes and promotions
- Calculated sell-in forecast for supply planning purposes
- 170+ features identified for Machine Learning
- Approx. 2000 SKU’s sold via 4 e-commerce platforms
Solution
- Generated Demand Forecast on a monthly basis for the e-commerce business using ML techniques
- Forecast Accuracy measured at a Product/Platform level on weekly and monthly basis
- Impact of Launch + Promo/Discounts etc. for rolling 12 months and calculated Primary forecast for supply chain
- Data collected from 10 sources including hourly promotions
- Data trained with 2 years of history and validated with 1 year of historical sales
Results
- Above 90% Forecast accuracy at product / platform level
- ML non-linear forecasts are better than classical statistical techniques at inventory level
- ML forecasts are driven by discount %, promos and 3 months prior sales
