The Challenge
A retail chain was losing money at both ends of its inventory: perishable stock expiring on shelves, and popular items selling out before replenishment arrived. Ordering decisions lived in spreadsheets and gut feel, and every branch did it differently.
What We Built
- Demand forecasting — ML models trained on historical sales predict demand per product, per branch
- Predictive restocking — the system recommends order quantities and timing before shelves run dry
- Automated supplier ordering — approved recommendations become purchase orders automatically
- Chain-wide visibility — one dashboard for stock health across every branch
Technology
TensorFlow models served through a FastAPI layer, with MongoDB storing sales history and stock movements across the chain.
The Outcome
- 40% reduction in stock wastage in the first months of operation
- Stock-outs on high-velocity items dropped sharply
- Reordering is now consistent and automatic across all branches