Self-checkout 2.0: Revolutionizing retail security and the customer experience
04th June 2026
This content is updated periodically to ensure information and statistics are current. Refreshed June 2026.
Modernising self-checkout security: How Vision AI Detects Fraud and Drives Smarter Retail
Self-checkout (SCO) isn’t going away – it’s evolving. In fact, projections indicate that by 2030, over 24,000 stores will offer self-checkout. And the gap between first-generation systems built around security scales and what Vision AI can now deliver has never been wider.
Early SCO was designed for a controlled, linear journey: one shopper, scanning items in order across a scanning plate, placing them neatly in a bagging area, waiting for the scale to confirm. The infamous “unexpected item in bagging area” led many retailers to tune security scale sensitivity down – or off entirely – raising the question: what’s the point of the scale anyway?
Vision AI proved itself in that highly structured shopping environment, enabling shoppers to self-correct with video-based alerts, reducing employee interventions and delivering new insight. The technology is now evolving to secure self-checkouts regardless of configuration – with or without scale, with or without designated bagging areas – empowering shoppers and securing inventory for retailers.
This guide explores the reality of SCO loss, why legacy systems fall short, how self-checkout is evolving, and how vision AI bridges the gap across all checkout formats – positioning intelligent loss prevention as core retail infrastructure rather than a security add-on.
Why traditional self-checkout security measures are not enough
Self-checkout entered retail in 1986 as a transaction efficiency tool – a way to process more customers with fewer staff. For four decades it delivered on that promise, expanding steadily into one of the most consequential shifts in how retailers operate at the front end of their stores. The global market now stands at $7.88 billion and is projected to reach $13.5 billion by 2030.
For much of that period, the relationship between shoppers and the technology worked well. Adoption grew consistently – driven by retailer investment in operational efficiency and a generational shift in shopper preference toward speed and autonomy. Each year, more SCO lanes were installed, more transactions processed, more shoppers who had never used self-checkout became shoppers who used little else. COVID-19 reinforced what was already in motion: with contactless and staff-light shopping suddenly a priority, the hesitation some shoppers still had about self-service largely disappeared.
Today, 53% of younger shoppers select SCO purely for speed, and nearly half of all demographics say shorter queues are non-negotiable. The checkout is no longer a place people are willing to wait. It is a place they expect to move through on their own terms – scanning in any order, bagging as they go, using whatever device is in their hand.
That change in how shoppers behave is what first-generation self-checkout security was never designed to accommodate. The format scaled to meet the new shopper. The security model built around the old one did not.
Here are traditional security measures and what they miss:
The evolution of checkout types and why it’s important for security
The original security-scale based self-checkout (SCO) model was built around a simple assumption: one customer scanning items sequentially across a scan plate into a bagging area, with a security scale validating each action.
Vision AI builds on this by adding a critical layer of oversight. It can detect missed scans, concealed items, or walkaways-even when the sensitivity of the security scale is reduced. Critically, it doesn’t just alert to an issue -it delivers a short video clip of the event with the relevant item already highlighted, giving colleagues the visual evidence they need to act with confidence in seconds.
This visibility changes the experience. Shoppers can immediately see and correct mistakes, while attendants are better equipped to guide and coach in real time. It also provides analytics to highlight operational gaps, identify training needs, and surface patterns such as which products are most often miss-scanned (Hint: It’s often not what you may think it is).
Discover how Edeka delivered frictionless checkout with Vision AI
This evolution gave rise to two new and very different configurations. Self-checkouts with shelves for larger shops and compact, sleek self-checkouts with minimal bagging area – both without the security scale.
Guided | Traditional SCO
Items scanned left to right or right to left, into a designated bagging area with a security scale. Scales are frequently tuned down as they over-report errors and frustrate shoppers removing the primary security mechanism.Semi-guided | Dynamic bagging area
No fixed scanning structure, but the AI learns the shopper's bagging area from their behaviour. The system nudges based on observed deviations. Newly developed, well-received, and rapidly adopted.Unguided | Full reconciliation
No rules at all. Shoppers may scan in any order, from trolleys, from bags or their arms. The AI performs a full reconciliation at payment comparing every item observed by camera against every item scanned.How common is fraud and theft at self-checkout?
Self-checkout systems have transformed the retail experience, but they have also created new opportunities for theft. As Dr Matt Hopkins, Associate Professor in Criminology at University of Leicester, explains: “Self-checkouts create opportunities for people to steal who would not otherwise consider shoplifting.” This highlights a growing concern for retailers as the convenience of self-service can sometimes come at the cost of increased fraud and theft.
A LendingTree survey of 2,050 respondents found that 27% of US consumers admitted to purposefully stealing at self-checkout, with 15% saying they had done so frequently. In the UK, research for The Grocer found that 37% of shoppers had failed to scan an item, though not all intentionally.
Dr Matt Hopkins, explains that, “The self-serve format reduces the social friction that normally accompanies dishonest behaviour in a staffed environment. The machine does not judge. The colleague is three lanes away. For a subset of shoppers, that is enough.”
Pre-deployment shrink baselines run by SeeChange have shown that fraud can represent 1% to 8% of all transactions at self-checkout – a figure that combines intentional and unintentional fraud. Yet, even in environments with a low fraud rate, improvements in employee efficiency from technology deployments offer the payback.
Laurent Hugou, Director of Intermarché La Farlède, gives a retailer’s perspective on vision AI:
“We can now clearly differentiate between intentional and unintentional fraud. There are often items that are not scanned due to handling errors. All of this causes losses for the store of up to 3% of all transactions. Since these AI-enabled cameras were installed, this figure has been halved and our goal is to get it below 1%.”
What are the common types of self-checkout theft and fraud?
To effectively reduce shrink, retailers must first understand how it occurs. Not all SCO loss is deliberate. Customers forget items at the bottom of trolleys, misidentify produce, or scan incorrectly. These errors are equally costly and deserve the same response.
While you may hear about 30+ behaviour patterns, they typically fall into one of the below categories. The key is choosing an AI loss prevention system that continually learns, so you can defend against new fraud patterns identified across other retailers, as well as within your own stores, and across checkout configurations.
Below are the five most common categories of loss:
You’ll also hear about the banana trick, so named because shoppers were often placing expensive items (like a bottle of alcohol) on the produce scale, and then selecting a low-cost fruit like bananas from the list of non-barcoded items on the SCO’s menu, paying only for the weight of the cheaper item. Without visual cross-referencing, the discrepancy is undetectable by weight alone.
What is a computer vision powered self-checkout?
AI-powered, computer vision self-checkouts use cameras, sensors, and artificial intelligence to observe, interpret, and act on checkout behaviour in real time. Unlike traditional systems, which rely on manual scanning and weight verification, AI self-checkout can detect products visually, cross-reference what it sees against what the POS records, and intervene before the transaction completes.
Terms you’ll hear – nudge vs block
Whether the system nudges or blocks is entirely down to the retailer. Every intervention should be configured through an AI Rules Engine, giving operators full control over how each use case responds. Rules can also flex automatically based on wider conditions: reducing intervention thresholds during quieter periods, tightening them during peak hours, or applying different logic entirely depending on the day of the week or store format. The retailer sets the parameters. The system operates within them.
- Soft nudge: The default for most fraud events with a designated or validated bagging area. A neutral, non-accusatory prompt with video evidence on the SCO screen invites the customer to check their last item. Between 50–80% of customers self-correct without any staff involvement.
- Hard block: Where repeated nudges have failed to course correct a transaction or where there is evidence of an activity a retailer considers high risk (e.g. product switching). The transaction is paused and the colleague is alerted with video-evidence alert on their handheld device or via the self-checkout.
Six examples of how vision AI reduces retail loss
Fraud type
Missed scan / Skip scanningHow AI detects it
The camera tracks every product entering the scanning zone. If an item reaches the bagging area without a corresponding POS scan event within a defined window, the discrepancy is flagged. The system detects the absence of a scan event — it does not distinguish between an intentional pass-around and a genuine handling error. The detection mechanism and the response are identical in both cases.What could happen next*
For example: A soft nudge appears on the SCO screen - 'It looks like an item may not have been scanned. Would you like to check?' Between 50–80% of customers self-correct without any staff involvement. Transaction continues. No queue impact, no confrontation — whether the miss was deliberate or accidental.Fraud type
Product stackingHow AI detects it
Item-level visual counting tracks discrete products entering the bagging area. If the count does not match the number of scan events for items placed, one scanned the quantity discrepancy is flagged.What could happen next*
Nudge for low-confidence events. Hard block and colleague alert for high-confidence quantity mismatches.Fraud type
Barcode / product switchingHow AI detects it
The reconciliation engine compares the product the camera identifies against the barcode registered in the POS. A visual mismatch triggers a flag. This requires both vision and live POS integration neither alone is sufficient.What could happen next*
Hard block: transaction paused. Colleague receives a video-evidence alert on their handheld within 500ms. One tap to review and resolve.Fraud type
WalkawayHow AI detects it
Session-level tracking monitors transaction lifecycle from initiation to payment. If products are in the bagging area and the session is abandoned, the system flags the incomplete transaction before the lane resets.What could happen next*
Colleague alert before the lane clears. Alert includes what was in the bagging area at abandonment. Store or SOC escalation available for repeat patterns.Fraud type
Start and end anomaliesHow AI detects it
Start: Detects items in bagging zones and raises an alert. End: Reconciles unscanned items and flags any remaining anomalies-real time at the checkout.What could happen next*
Soft nudge. Customer self-corrects. Followed by hard block if not corrected.Fraud type
MultipackHow AI detects it
When a barcode representing an individual item is scanned but the appearance resembles a multipack, the system sends a multipack alert in real-time to the self-checkout.What could happen next*
Soft nudge. Customer self-corrects. Followed by hard block if not corrected.* In reality, business priorities fluctuate based on changing circumstances like seasonality, footfall and even time of day. That’s why a rules engine must support dynamic configurability to enable precise, real-time tuning of sensitivity to what matters most, automatically or manually.
Self-checkout security ROI: What retailers actually see
Loss prevention investment needs to justify itself at board level, and AI self-checkout security is no exception. The good news is that the ROI is measurable, the timeline is short, and the metrics are straightforward to track. Most deployments demonstrate material impact within 60-90 days of live operation.
Metric
Shrink reductionWhat good likes like
50%+ reduction in SCO-attributable shrink. Measured against a matched baseline period or a control store group running without the system.What drives it
Real-time reconciliation catches missed scans and barcode switching mid-transaction before loss occurs. Self-correction handles the remainder.Metric
Self-correction rateWhat good likes like
50 to 80% of customers nudged by the system self-correct without any staff involvement.What drives it
Soft nudge intervention. Neutral language. Customer corrects, transaction continues. No queue impact, no staff required.Metric
Employee intervention reductionWhat good likes like
15%+ reduction in call-outs to SCO lanes.What drives it
The system aids shoppers to self-correct with staff being alerted when risk has increased with evidence videos already available reducing false alarms and wasted time.Metric
Throughput impactWhat good likes like
4x transaction speed improvement by removing gallery look-up for non-barcoded items.What drives it
Fewer colleague interventions, faster transaction times as the system guides the shopper. Lowers frustration fraud as shoppers are presented with shortlist.For retailers, the checkout is no longer just a place for shoppers to pay. It is a data node – one that generates structured intelligence about every product handled, every transaction completed or abandoned, and every anomaly across the full front end. Retailers who unlock that intelligence first will hold a structural advantage that compounds with every store added to the estate.
Beyond loss prevention: self-checkout part of a store intelligence ecosystem
The future of Vision AI at self-checkout is not isolated or standalone. It is connected, intelligent, and generating value at every data point in the transaction.
When vision AI is deployed at self-checkout, something more than fraud detection is switched on. Every transaction generates a structured record of what was physically present at the lane, when, in what sequence, and how it was handled. Aggregated across thousands of transactions per day, per lane, per store, this data reveals patterns in shopper behaviour, training requirements and evolving trends in fraud.
Loss prevention is the immediate, measurable, board-level business case and it remains a primary driver for initially deploying Vision AI, but it is the first layer of a much larger capability.
At checkout (self or staffed) vision AI opens the box to new insights. When linked to losses that occur in the aisle using overhead CCTV, a more complete picture evolves connecting events from shelf to checkout and exit. Loss prevention may be the entry point, but new use cases can then be added taking advantage of the same infrastructure to determine on-shelf availability, dwell time, slip and trip hazards, queue time and more.
When it comes to vision AI, the self-checkout lane is not the end of the transaction. It is the beginning of something much larger.
Ready to move beyond first-generation SCO security?
