Self-Checkout Fraud: the tills where shrinkage concentrates

What it does

The cameras over the self-checkout bank already record everything; this check reads the behaviour in frame and flags what looks like scan avoidance or a switch-out. Items that move straight from basket to bag without passing the scanner. The trolley base that never gets unloaded. A barcode covered by a thumb, or swapped for a cheaper one. Because an alerting check judges a window of recent frames, the whole move reads as a sequence rather than one blurry still, and for the fastest sleight of hand there's the Precise watch level, a frame every second. Every flag is exactly that, a flag for review, never a verdict.

What an alert looks like

An alert lands on screen and by email with a plain-English description (‘customer at till 3 bagged two items without scanning’) and a link to the exact moment. There's a JSON webhook too, which shops typically point at the POS back office or a supervisor's device. The colleague reviews the clip, then does the normal polite intervention, the ‘shall I help you with that?’ that resolves most of these without anyone being accused of anything.

Why the self-checkout bank needs its own check

One colleague supervising six tills can't watch twelve hands, and the machine's own scales only catch the tricks that involve weight. The item that never approached the scanner doesn't register anywhere, except on camera. That's the gap this check covers: live prompts while the customer is still at the till, an end-of-shift review of the flagged moments, and over time a sense of which tills collect the most trouble.

Where it runs

Enabled per camera, so it can run on the self-checkout bank and nowhere else if that's all you want. It works on RTSP feeds, uploads and a propped phone camera, and recorded footage goes through the same pipeline as live, so you can run it over last week's export and see what it would have flagged before committing a camera to it. A fully on-prem appliance mode exists for footage that can't leave the building. Its shop-floor sibling is Theft Watch: that one covers concealment in the aisles and movement towards exits, this one covers the till bank, and plenty of shops run both.

What it costs

The Store shows Self-Checkout Fraud with a live monthly price computed from your own usage before you buy, as ‘+$X/mo at your usage’: your camera count, hours a day and watch level, nothing else. That sits on a subscription priced from your own usage, floored at $9 a month, with the derivation shown on screen at signup. The pricing section has the detail.

How is this different from the scales and cameras built into the till?

The till knows weights and scans. It can't see the item that never came near the scanner, or the trolley that never got unloaded. This watches the behaviour around the till, which is where the losses the scales miss actually happen.

What does staff intervention look like?

An alert is a prompt to review a clip. If it holds up, a colleague has the normal ‘shall I help you with that?’ conversation while the customer is still at the till, which is quieter and works better than anything at the door.

Does it accuse customers automatically?

No. Every flag goes to a human with the footage attached, and nothing happens to anyone unless a person looks at the clip and decides it should.