How to police self-checkout without sacrifice or surveillance

The first, public self-checkout machine debuted in July 1986.  By 2027 the global self-checkout market will reach $7.8bn.

To put that in context, it is the same growth as the UK video games market - and the UK has been an enthusiastic adopter.  

There are around 80,000 self-checkout tills in the UK, and larger retailers now dedicate up to 50% of their tills to these machines.

The appeal is obvious: lower labour costs, greater throughput from faster transactions, and a smaller check-out footprint.

But self-checkout is far from perfect. The processes inherently increase exposure to risk, and the possibility of substantial loss.

The anatomy of self-checkout loss

A shopper waves a cereal box past the scanner. There’s no beep or scan, and the event goes unnoticed. The busy shopper misses it while processing a basket full of items, and the one associate keeping an eye on multiple lanes does too. A single unscanned item in a full basket, three lanes over, is almost impossible to see by eye.

When repeated across hundreds of thousands of checkout transactions, every day plus other loss patterns such as walkaways, across multiple stores, this passing moment becomes one of the costliest and hardest problems to solve in retail today.

According to the ECR, self-checkout loss now makes up 28.5% of all retail loss, up from under 10% in 2018.

This puts retailers between a rock and a hard place. On one hand, they need to enable customers to quickly, frictionlessly, and conveniently complete transactions at the self-checkout stations.

On the other hand, they need to protect their revenue and mitigate against these losses in real-time, without introducing friction.

The problems with solutions

In the UK, retailers have built up layers of protection: staff on checkout lanes, security scales in the bagging area, tagged boxes for high-value goods, and exit gates that ask shoppers to scan a receipt.

Security scales are the most common of these, and for good reason. They are installed across most large UK estates, shoppers know them, and they stop the simplest cases of loss. But scales only measure weight. They cannot see what was placed in the bag. That leaves clear gaps: a cheap item scanned while a heavier, more expensive item goes into the bag, a barcode swapped from a low-price product, items that never leave the basket or trolley, and shoppers who walk away before paying.

Scales also cause friction of their own. “Unexpected item in the bagging area” is now a well-known phrase in the UK, and most of these alerts are false. Each one stops the shopper and needs an associate to clear it.

Elsewhere, retailers have sought to evaluate camera-based systems at self-checkout but have found themselves falling foul of GDPR's strict rules on biometric data.

And of course, selecting the wrong technology is not only a substantial compliance risk, but also a major threat to customer trust in the brand - built over years, but destroyed in a single news cycle.

The ECR Retail Loss 2026 study names biometrics, customer data protection, and return on investment as the central open questions facing AI-based self-checkout technology.

In essence, the challenge is not just about detecting loss but doing so responsibly.

Just look down

To do this, one approach is to focus on the transaction, not the shopper. Instead of facial recognition or biometric templates, a kiosk camera, powered by vision AI, watches the scan session and correlates what it sees with the live scanner and point of sale (PoS) data stream. This does not replace the security scale. It adds a second layer. The scale checks the weight, and the camera checks what actually happened in the scan zone, the bagging area and the basket.

Because the system can see the item, it can also help confirm what the scale is reporting. Over time, this can mean fewer false weight alerts and fewer interruptions for honest shoppers.

If an item leaves the scan zone, bagging area, or basket without a valid scan, there’s an instant system notification and an on-screen prompt asks the shopper to scan the flagged item or issue.

This soft, self-correction resolves 80-97% of flagged cases with zero associate involvement. Escalation is the exception and depending on the type of incident detected and the retailer’s policy, an alert with a short clip can also go to a supervisor

There’s always a human making the decision and the technology never needs to know, or store, the identity of the shopper. Neither biometric data nor facial recognition is required to prevent losses at self checkout.

This is critical because self-checkout systems and retail loss prevention technology cannot recognise motivation.

No AI can distinguish a shopper who genuinely forgot to scan an item from an intentional theft. And there is no need for the system to make such a distinction.

This limitation is inherent to the technology.  There is actually data that shows items like oranges, water, and paper towels are among the most commonly missed items because their barcodes are awkward to find or scan, but even here it is impossible for the system to recognise that this context is different from someone who had no intention of paying.

And if most missed scans are honest, unintentional mistakes, protecting privacy and revenue shouldn’t be a binary choice.

Beyond this, the implementation of technology to secure revenue and maintain a premium customer experience needs to be handled sensitively.

Looking down at the transaction and not up at the face of the person at the checkout, is a necessary first step.

‍