Self-checkout has always required a strange little suspension of disbelief. A retailer hands the customer the scanner, the bagging area and most of the cashier’s job, then surrounds the transaction with cameras, weight sensors, alerts and occasional employee interventions because it cannot entirely trust the person it just deputized.
The shopper is simultaneously the customer, the cashier and, under the wrong circumstances, the suspect. Most of us have learned to live inside that contradiction. We scan the cereal. We search for the produce code. We hold up a reusable bag for inspection by a machine that has decided its weight is suspicious. Then we wait for somebody in a vest to clear an error created by a system that was supposed to reduce the need for somebody in a vest.
Scandit’s latest self-checkout technology adds another role to the arrangement: the polite observer.
The company announced Self-Checkout Loss Prevention on Sept. 15, a vision-AI system designed to detect common loss patterns as they happen. Cameras watch the transaction rather than identify the person, comparing item movement against scanner and point-of-sale data. If something passes through the checkout without a valid scan, the system can flag it before payment is completed.
The interesting part is what happens next. Instead of treating every discrepancy as an accusation, Scandit says the system can issue what it calls a soft on-screen nudge, giving the shopper an opportunity to rescan the item and continue without an employee becoming involved. Associates are brought in when the situation is not resolved, when a shopper walks away, or when a retailer’s policy calls for escalation. Scandit says the majority of flagged incidents can be resolved without employee intervention.
That is a technical feature, but it is also a social decision. The machine has seen the same thing either way: an item appears to have moved through the transaction without being scanned. What changes is the interpretation offered to the human standing in front of it.
One interface effectively says: We caught you. The other says: Looks like you missed this. Those two sentences can emerge from exactly the same evidence and produce completely different experiences.
The distinction matters because self-checkout systems operate in a place where intent is unusually difficult to infer. A missed scan can be theft. It can also be distraction, a barcode that did not register, an item hidden beneath another item, a shopper moving too quickly, a child adding something to the bagging area, or the ordinary clumsiness created when a person who does not work for the store is temporarily asked to perform part of the store’s job.
Retailers still have a real loss problem to solve. Scandit cites 2026 ECR Retail Loss research saying stores with self-checkout experience higher losses than comparable stores without it, and the company says its system can recover or deter more than 75 percent of self-checkout loss. Those are vendor-presented claims and measurements, but the broader design problem is easy to understand: retailers want the labor and convenience advantages of self-service without accepting every cost that comes with transferring the transaction to the customer.
Historically, many of the tools used to close that gap have added friction. Scales stop the lane because the bagging area disagrees with what just happened. Gates and receipt checks turn the exit into another checkpoint. Random inspections interrupt people who may have done nothing wrong. Even when the security logic is defensible, the experience can make an honest customer feel as though completing a transaction requires repeatedly proving innocence.
Scandit’s approach is interesting because it does not eliminate observation. It changes the etiquette of observation.
The cameras are still there. The transaction is still being monitored. The system still notices behavior that falls outside the expected pattern. But the first intervention can be designed around the possibility that the person made a mistake rather than the assumption that the person intended to steal. That is surveillance learning manners.
There is a larger lesson hiding in the checkout lane. Human systems constantly have to decide whether an irregularity represents incompetence, confusion, negligence, bad faith or something more deliberate. Managers do it when work is incomplete. Teachers do it when assignments look suspicious. Moderators do it when behavior trips an automated rule. Security systems do it when somebody moves in an unexpected way. The evidence usually arrives before the intent does.
Humans are not particularly good at that distinction, especially when institutions are designed to prioritize risk. Once a system is built around preventing loss, fraud, cheating or abuse, every anomaly can begin to look like a threat. The safest operational assumption is often the harshest social assumption.
The soft nudge offers a different sequence: detect first, permit correction second, escalate third.
That order is important. It creates room between noticing that something went wrong and deciding what the wrong thing means.
It is also an unusually appropriate design for self-checkout because the retailer created the ambiguity in the first place. A staffed register separates the roles cleanly. The customer shops. The cashier scans. If an item is not scanned, responsibility belongs largely to the employee performing the transaction. Self-checkout collapses those roles into one person and then asks technology to monitor the resulting uncertainty.
Put differently: stores outsourced part of the labor to shoppers and then discovered that shoppers are not professional cashiers.
That sounds obvious when stated plainly, but much of self-checkout security has been built around treating deviations from professional cashier behavior as potential loss events. The customer is expected to perform the task correctly while also being monitored because the store knows the customer may not perform the task correctly. It is a wonderfully circular piece of modern retail design.
Scandit says its system tries to resolve that contradiction without identifying shoppers. Its product materials describe item-based detection rather than facial recognition or biometric identification, with video processing occurring locally at the lane. That does not make the interaction invisible or remove every privacy question surrounding camera-equipped retail spaces, but it narrows what the system claims to be trying to know. The central question is not who are you? It is did this item get scanned?
That limitation may be as important as the nudge itself. A system that only needs enough information to correct the transaction does not necessarily need enough information to construct a theory about the person.
There is something almost humane in that restraint, even if the underlying purpose remains loss prevention. The technology does not need to decide whether the shopper is honest. It needs to notice that the transaction is inconsistent and offer the easiest available path back to consistency. That is a much smaller judgment, and smaller judgments are often safer ones.
The irony is that self-checkout began as an automation story. Remove the cashier, speed up the lane, let customers handle simple transactions themselves. But each layer of automation has produced another layer designed to manage the complications created by the previous one: cameras to watch the scanning, AI to interpret the cameras, alerts to manage the AI, and now carefully calibrated language so the automated observer can intervene without making the customer feel accused.
We automated the cashier and then had to teach the surveillance system customer service.
That may be where the idea becomes bigger than retail. As automated systems increasingly watch for errors, risk and noncompliance, the quality of the experience will depend not only on what those systems can detect, but on what they assume after detection. A machine does not have to possess empathy to be designed around a more generous sequence of possibilities.
Sometimes the most consequential feature is not whether a system can catch you doing something wrong. It is whether the system gives you a chance to demonstrate that wrong was simply a mistake.
And in a self-checkout lane, where we have somehow agreed that the customer should do the work while the store watches to make sure the customer does the work correctly, a little mechanical benefit of the doubt may be the most frictionless feature of all.
SOURCE NOTES
• Scandit: Scandit Enters Loss Prevention Market: Self-Checkout Solution Reduces Loss While Keeping Shopping Frictionless (Sept. 15, 2026)
• Scandit: Inside Self-Checkout Loss Prevention: How Vision AI Helps Stop Loss as It Happens (Sept. 15, 2026)
• Scandit: Self-Checkout Loss Prevention Solution
• PR Newswire release (Sept. 15, 2026)
Product claims attributed to Scandit / PR Newswire materials cited in SOURCE NOTES. Cultural framing is RMN's.