Brief
An intelligent warehouse management system combining demand forecasting, OCR-powered receiving and real-time stock tracking, deployed across 14 fulfilment centres.
−47%
Stockouts
14
Warehouses
<3s
Invoice → Stock
Problem
Receiving depended on someone typing supplier invoices into a system, which meant stock levels were always slightly wrong and always slightly late. Reordering was based on those numbers, so stockouts were structural rather than occasional.
Solution
- An OCR pipeline that ingests supplier invoices and updates stock in under three seconds, removing the manual keying step entirely.
- Demand forecasting that reorders against predicted movement rather than last month’s figures.
- A mobile picker app with barcode and voice-pick, so floor staff are not walking back to a terminal.
- An operations dashboard with anomaly detection on stock velocity and shrinkage, surfacing problems while they are still small.
Result
Stockouts fell 47% across 14 warehouses, and invoice-to-stock time dropped to under three seconds from what had been a manual data-entry queue.
Under the hood
| Forecasting | Prophet demand models |
|---|---|
| Document AI | OCR invoice ingestion pipeline |
| Backend | Python, Django, PostgreSQL, Celery |
| Operations | Anomaly detection on velocity and shrinkage |
