On-shelf availability (OSA) is the single biggest, most fixable leak in physical retail. When a product a shopper wants is not on the shelf, the sale is usually lost — and sometimes the shopper too. Out-of-stock detection with computer vision turns ordinary shelf images into real-time alerts, so the gap is closed in minutes rather than days.
What out-of-stock detection actually does
A computer-vision system photographs the shelf — from a fixed camera, a store associate's phone, or an autonomous unit — and a trained model identifies which products are present, which facings are empty, and where stock is running low. It compares what it sees against the planogram (the intended layout) and raises an alert the moment a gap appears.
Widely cited retail research by Gruen and Corsten has long put the average out-of-stock rate near 8%, with retailers losing roughly 4% of sales to stockouts. For most chains that is the largest recoverable revenue line they have.
How the pipeline works
- Capture — images from fixed cameras, associate phones, or autonomous units.
- Detect — an object-detection model (for example a YOLO-family network) locates products and empty facings.
- Compare — detections are matched to the planogram to score availability and share of shelf.
- Alert — gaps become a task for staff, pushed into the tools they already use.
- Learn — outcomes feed back so accuracy improves on your own shelves.
The KPIs it moves
On-shelf availability rate, lost-sales recovery, replenishment response time, planogram compliance, and share of shelf versus competitors. These are board-level numbers, which is why OSA projects tend to pay back quickly.
Getting started without boiling the ocean
The classic mistake is trying to wire every store at once. Start with one category in a handful of stores, prove the lift on real KPIs, then scale. That is exactly how we run a 14-day proof-of-concept with Shelfzar, our retail computer-vision platform — deployed on your own infrastructure so your image and sales data never leave your control. To shape the product early, see the founding cohort.