UC-047
Demand Forecasting & Replenishment
Forecasts demand at 15-minute granularity across thousands of stores and hundreds of SKUs, then optimizes store replenishment.
TBDBuild Duration
XXXXIndicative ROI
XXXXBuild Cost
XXXXValue
The Challenge
Forecast error around 30% MAPE drives stockouts and perishable waste. Inventory turns still have to be balanced against stockout risk across thousands of stores.
How It Works
- Uses graph neural nets for store-region similarity and transformers for SKU time series.
- Optimizes replenishment allocation with a mixed-integer solver.
- Models pay-cycle peaks, seasonal uplift, cannibalization and weather shocks.
What It Removes
- Ordering against a forecast that misses by ~30%
- Perishable stock written off before it sells
- Working capital tied up in buffer inventory
Input Data RequirementsStore and SKU sales history, distribution centre and store inventory, weather, pay-cycle and seasonal calendars
Output Format15-minute demand forecasts, distribution-centre-to-store replenishment allocations
“Shelves hold what actually sells that day, and far less of it ends up thrown away.”