The Challenge
Manual shelf checks were slow and inconsistent, so out-of-stock and misplaced products went unnoticed for hours, costing sales and frustrating customers.
Our Approach
We trained object-detection models on shelf imagery and deployed them on edge devices for low-latency inference, generating real-time alerts the moment a shelf needs attention.
Results
- 99.2% of restocking issues detected
- Sub-100ms inference latency at the edge
- Real-time alerts instead of periodic manual checks
- More consistent on-shelf availability