The Complete Guide to Self-Checkout Fraud Prevention in 2025

What actually gets reported — and what doesn't

Self-checkout losses are easy to underestimate because most retail theft never becomes a police statistic. In the UK, police recorded 509,566 shoplifting offences in the year ending December 2025, while retailers logged 20.4 million theft incidents over a comparable year. The gap is not a rounding error — it is the majority of the problem, and it is why store-level detection data usually tells you more than crime reports do.

Sources: ONS — Crime in England and Wales, year ending December 2025; BRC Retail Crime Survey 2025. Full figures and provenance: UK retail theft statistics.

The size of that gap varies by country, and so does whether it can be measured at all. In Israel, for example, the police offence classification has no shoplifting category, so no equivalent reported-vs-actual comparison can be constructed — see what official Israeli data does and does not measure.

Where automated detection helps — and where it doesn't

Detection narrows what a human has to review; it does not decide guilt. Treat every alert as an exception to check, not a conclusion. Three limits are worth stating plainly:

• Intent is not observable. A missed scan and a deliberate skip look similar on video. The system surfaces the event and the matching till exception; a person judges it.
• Camera coverage bounds accuracy. If the lane, the bagging area and the basket are not all visible, some events cannot be reconstructed regardless of the model.
• Volume shifts the cost. A sensitive threshold finds more real incidents and also more false alarms; the right setting depends on how much review time a store can actually staff.

De Flow AI runs on the cameras a store already has, pairs each flagged event with its till exception so the reviewer sees both, and does not use facial recognition.

Frequently asked questions

What counts as self-checkout fraud?

The common patterns are non-scanning (passing an item over the scanner without a read), label or barcode swapping to charge a cheaper item, understating quantity on loose goods, and mis-selecting a cheap produce code for an expensive item. Each leaves a different trace: non-scans show as a weight/scan mismatch, swaps show as a price anomaly against the item seen.

Do we need to install new cameras or hardware?

No. De Flow AI is designed to work with existing in-store CCTV, so the deployment question is usually about camera angles and coverage rather than new equipment.

Does self-checkout monitoring use facial recognition?

De Flow AI does not use facial recognition. Detection is based on actions at the lane — items, scans and till events — not on identifying who a shopper is.

How do you avoid accusing honest customers?

By treating detection as evidence-gathering rather than enforcement. An alert pairs the video moment with the corresponding till exception so a trained reviewer can see what happened before anyone speaks to a customer. Most flagged events turn out to be genuine mistakes, and handling them that way is both fairer and lower-risk.

Is self-checkout theft mostly customers or staff?

Both occur, and they need different controls. Customer-side losses cluster around non-scans and label swaps; staff-side losses cluster around voids, refunds and discount misuse, which is why till-exception data matters as much as video.

How should we measure whether prevention is working?

Compare shrink for the same stores over comparable trading periods, and track review workload alongside it: incidents surfaced, incidents confirmed, and time spent reviewing. A programme that raises alerts without reducing confirmed loss is generating work, not results.