Smart self-checkout security: Integrating AI video and POS to reduce retail shrink

Retail customers increasingly prefer self-checkout. This has been shown by numerous studies, including the latest by Diebold Nixdorf. Retailers reciprocate this preference, as self-checkout counters help reduce wait times, optimize inventory accuracy, improve overall customer experience and alleviate staffing shortages through streamlined processes.
 
One issue remains, however. With a self-checkout solution in place, customers leave the store without performing the symbolic handover of their purchase from the cashier. Ensuring that customers pay for what they take home—without feeling unduly scrutinized—requires the deployment of multiple technologies that make self-checkout beneficial for all.
 

Customers want:

• Fast, reliable checkouts
• An intuitive user experience that reduces the fear of making mistakes—and potentially being held liable for them
• Feeling valued by the retailer instead of being scrutinized

 

Retailers want:

• Streamlined processes that free up the attention of staff to other customer-facing tasks
• Reduce shrinkage, at self-checkout counters and elsewhere
• Create a customer experience that makes full use of modern technology and reduces potential areas of friction

 
When deploying technology—from cameras for situational awareness to fraud-proof cash point-of-sales (POS) terminals—integrators of security (and in-store IT) systems face the challenge of putting an integrated system in place that serves the needs of all stakeholders equally.
 

What are scenarios retailers need to prepare for?

Customers who see a self-checkout counter as an invitation to defraud the retailer might be few and far between, but they can spoil the experience for everyone. The most important rationale is therefore to reduce shrinkage at those counters to a minimum and ensure they can be operated economically.
 

Retailers need to be prepared that a small minority of shoppers may try to:

• Disguise expensive items as cheaper ones (barcode switching)
• Hiding smaller items in bigger ones (item stacking)
• Misrepresenting non-barcode items that are being weighed during self-checkout, for example labeling pricier cherries as cheaper potatoes

 
A more important factor, however, are accidental “mis-scans” that lead to customers paying for too many, too few, or the wrong items, leading to a disappointing shopping experience on their part and a mismatch in the stock at the retailer.
 
Self-checkout security requires integrators of security systems and in-store IT to work hand-in-glove. As manufacturers typically focus on either this or that realm, integrators need to have a firm understanding of both. It is inevitable, however, that the design of a security deployment in a self-checkout area approaches the task from either this or that end.
 

Security cameras: How to place and integrate them to deter fraud

Let’s first focus on security systems and, most importantly, video security, which can play an important supporting role in securing self-checkout counters. Several manufacturers offer systems tailored to the needs of retailers, with solutions that go beyond creating visibility from the security control room to the checkout area.
 
Verkada, which offers integrated security solutions centered on its cloud-based platform Command, has recently introduced new features that retailers can benefit from.
 
“By combining AI analytics with cloud infrastructure, retailers can set up real-time alerts for unusual activity—like loitering near self-checkout areas,” said Oscar Popravka, Head of Integrations at Verkada. “Beyond alerts, webhook-based activity logging turns video events into structured data that can be integrated with other systems.”
 
Verkada solutions are tailored to the needs of self-checkout security, addressing multiple pain points in the workflow, from automated real-time responses to streamlining forensic investigations and preventive measures against repeat offenders.
 
“Verkada’s cameras can use facial recognition to create Persons of Interest (POIs) alerts,” Popravka said. “This enables retailers to be notified if the same individual enters a store again.”
 
Integration is key to creating a tight-knit safety net against fraud attempts.
 
“Verkada Helix makes it possible to link point-of-sale transaction data with video footage at self-checkout counters, essentially creating a customized video search engine for all transactions,” Popravka  explained. “This has been a game-changer for retailers trying to catch more subtle forms of fraud.”
 
“Once a fraudulent transaction is identified, teams can quickly piece together what happened,” he added. “Verkada’s History Player Search makes it easy to trace where an individual appeared across cameras before, during, and after a self-checkout transaction, eliminating the need to jump between feeds or scrubbing through footage. From there, the AI-Powered Unified Timeline organizes those moments into a single, chronological view, creating a clear picture of a person’s full path through the store.”
 
Verkada’s approach to retail security is comprehensive as it also encompasses the creation of business intelligence that can help store managers optimize stock, store layout and many other areas.
 
“The result is a well-monitored, unobtrusive environment,” Popravka concluded. “With modern safety technology, retailers gain deeper insights into the behavior of their customers throughout a given store and checkout area – keeping wait times low and optimizing store layouts.”
 

Zooming in on the self-checkout terminal

Video security systems play only a supporting role, though, as the key component to ensure everything is scanned, weighed and paid for correctly is the self-checkout counter itself. This is where the POS data is being generated that enable security systems to detect fraud. It is also the first line of defense to prevent them from happening in the first place.
 
Diebold Nixdorf is a leading manufacturer of self-service transaction systems and POS terminals. Its self-checkout counter solutions integrate various technologies to nib fraud in the bud, going beyond the ubiquitous sensors that ensure shoppers place items in the packaging area after scanning them.
 
“Our AI-powered solution Vynamic Smart Vision I Shrink Reduction is designed to prevent and address the most common sources of loss at self-service checkouts,” said Stefano Lai, AI Solutions Expert at Diebold Nixdorf “It uses cameras on top of checkout devices and Smart Vision technology to analyze behavior and activities in real time.”
 
Aside from the AI-powered analysis of visual cues of fraud attempts, the company’s solutions also addresses  potential events that are harder to detect.
 
“Combining this solution with Vynamic Smart Vision | Fresh Produce Recognition for automated identification of non-barcoded items such as fresh fruit and vegetables results in probably the most comprehensive toolbox currently available for reducing the most common sources of loss and friction at self-service checkouts,” Lai explained.
 
“The objectives included making the checkout process smoother and simpler for both customers and employees and reducing the interaction rate for employees at the self-service checkouts so that they have more time for customer service and other value-add tasks in the stores,” he added. “The solutions also help shoppers reduce scanning errors and retailers avoid alienating those who don’t intend to steal anything, especially as we are experiencing that up to 80% of the anomalies detected by the system include unmalicious transaction errors like missed scans.”
 
Another area specific to self-checkout security is escalating alerts. Simply put, not every mis-scan is a fraud attempt that needs to be handled as such.
 
An underlying rules engine applies retailer-defined rules to determine the appropriate action for each detection,” Lai explained. “For example, if the system detects a failed scan or an incorrect operation, shoppers receive a short message on the device display indicating that an item was not scanned correctly. This ‘nudging’ approach allows them to self-correct the transaction, which means the scanning process can be repeated without major interruption. The store associates are only informed if the system displays another error message, so they can provide support to complete the checkout process correctly.”
 
“Should the system detect any malicious attempts to manipulate the transaction (for example, when shoppers try to stack items on top of each other or switch product barcodes), it can immediately block the operation and alert store attendants on their mobile terminal or device,” Lai added. “In case of a necessary employee intervention, short video sequences of the identified anomaly help them better assess the situation awaiting them at the checkout so they can react appropriately.”
 
Automation is another area where fraud attempts and convenience can be addressed simultaneously, thanks to AI-powered object classification.
 
“Additionally, AI-powered fresh produce recognition solutions enable shoppers to easily and correctly scan non-barcoded products such as fresh fruit and vegetables,” Lai stated. “Another camera placed on top of the scale, combined with sophisticated algorithms, identifies items and their quantities which are then displayed at the checkout. This eliminates the often-frustrating manual selection of fresh produce for shoppers or weighing and affixing a price label in the fresh produce section. In parallel, it helps to avoid additional use cases of malicious attempts to manipulate the system, for example when shoppers should try to label an expensive bottle of wine as bananas.”
 

How to bridge the gap with APIs and webhooks

A key challenge for integrators is ensuring that the video security and and POS systems speak the same language. While these systems often operate in separate silos, the modern security stack is moving toward a "plug-and-play" ecosystem.
 
The bridge between these two realms is built on APIs and webhooks. In well-integrated systems, it might play out like this: When the POS terminal identifies a high-risk event, it can “notify” the video security system by triggering a webhook, based on which the VMS or cloud platform automatically tags the corresponding video footage with the transaction ID.
 
Setting up a comprehensive of alerts and triggers might be complex, but modern systems—on both sides—are set up for such integrations.
 
“The open design [of Vynamic Smart Vision] also allows integration into third-party software environments,” Lai said.
 

Final thoughts

Integration transforms retail security from a reactive task to a proactive one. The interplay between video security system and POS terminal is one example of a link that helps transcend security solutions.
 
Full integration can create full, AI-based situational awareness, following items from the shelf, to the shopping basket, the self-checkout counter and eventually out of the front door. If any irregularities are being detected along the way, the system can automate responses and, if necessary, alert security personnel, giving them the context they need to respond to each situation in the best possible way.