Home › Product Recognition
Product recognition, without the barcode
Instant identification of fresh produce and unpackaged items on the scales, scanners and self-checkouts already in your stores.

Fresh is where grocery makes its margin and loses its throughput. Every loose item needs a human to identify it, find the code and get it right. Edgify removes that step entirely.
What it does
- Identifies loose produce automatically, including through plastic bags
- Removes the PLU lookup, the code sheet and the scrolling picker
- Cuts mislabelling and false selections that quietly erode fresh margin
- Runs on existing self-service scales, scanners, SCO and staffed lanes
Results in production
| Measure | Result |
|---|---|
| Checkout speed on fresh items | 5x faster |
| Mislabelling | 69% reduction |
| Recognition accuracy | 98.9% |
| Real-world training samples | 400M+ |
Why it works where others haven't
Produce recognition demos well and deploys badly. A model trained on clean catalogue imagery meets a real store and fails on the things that make stores real: condensation on the bag, a half-empty display, an unfamiliar local variety, overhead lighting that changes through the day.
Edgify trains on the device, in that store, on that store's items. The model that runs in a Tel Aviv branch is not the model that runs in a Manchester one, and neither of them needed an annotation contract to get there.
Beyond the checkout

The same recognition layer runs in back-of-house for waste tracking, in fulfilment for weighing and packing accuracy, and across food service for operational verification. One model estate, several P&L lines.
Frequently asked questions
Does recognition work through plastic bags?
Yes. Loose produce is identified through a plastic bag on an existing self-service scale, with no barcode, manual lookup or code sheet.
What hardware does product recognition run on?
Existing self-service scales, scanners, self-checkouts and staffed lanes. Where a device is too constrained to train locally it still runs inference and receives model updates.
How accurate is it, and does accuracy degrade over time?
98.9% recognition accuracy in production. Because each store's model keeps training on that store's own items, accuracy improves after go-live rather than decaying.
Does it need a data annotation project?
No. Every customer and staff confirmation at the checkout is a label. The model learns from it locally, so there is no annotation vendor and no annotation budget line.
What results have retailers seen?
5x faster checkout on fresh items and a 69% reduction in mislabelling, with no new hardware purchased.
See your store through Edgify's eyes
A 12-week pilot on the hardware you already own. Measurable ROI, from day one.