How machine learning can stop fraud at self-checkout kiosks

Self-checkout is a great asset to any store, reducing labor cost and improving checkout options, especially with retailers continually facing labor shortages.

As of June 2023, the U.S. Chamber of Commerce determined that the quit rate for the retail industry was higher than the national average and 30% of retail jobs were unfilled.

Despite the cost savings and labor efficiencies, unmanned checkout lanes are at a greater risk of theft and fraud than the traditional checkout alternative. In fact, one in five shoppers have intentionally committed fraud at self-checkout kiosks.

To mitigate risk, retailers have added cameras near their self-checkout kiosks or integrated rule-based or scale-based technology that initiates rescans or alerts employees to irregularities. These solutions are expensive, both in terms of financial costs and customer loyalty.

For example, an honest customer’s checkout experience is delayed and negatively impacted by unnecessary rescans or employee check-ins.

Let’s discuss how retailers can instead rely on artificial intelligence (AI) and machine learning (ML) to avoid unnecessary loss at the self-checkout kiosk.  

Why is self-checkout fraud higher than other points-of-sale?

Not only is self-checkout fraud common, but 58% of consumers said it is “easy” or “very easy” to commit this type of fraud. One reason for this is that shoppers may feel less remorse sneaking an item into their bag at an unmanned kiosk than they would in front of a cashier.

Shoppers may be more persuaded to scan a cheaper item twice, like a banana, instead of a more expensive product like a steak. Additionally, other shoppers may mistakenly commit self-checkout fraud by assuming an item was scanned when the barcode may have been faced away from the scanner. 

To identify these instances of risk, retailers should implement systems based on machine learning. These tools incorporate a configurable set of rules that automatically learn and improve as shoppers invent new ways to steal from the self-checkout kiosk.

What current self-checkout kiosks are missing?

Today, many retailers are relying on static, rule-based fraud detection systems that require constant monitoring and management. But these retailers are wasting time configuring new data into their systems.

For example, if a CPG company announces new, more eco-friendly packaging, their products will soon become a different weight. Retailers that used manual fraud detection tools will have to re-configure the weights in their system.

What’s more, these details may differ across regions if the CPG brand rolls out the new packaging in California before introducing it to the rest of the country. When these changes are handled manually, mistakes can occur that will decrease the accuracy of the retailer’s fraud detection methods.

If a fraud detection solution is working with inaccuracies, even your best, most reliable customers will be impacted. For example, if a shopper is trying to scan the product with the new packaging and the system doesn’t recognize it, the kiosk will likely prompt the shopper for several rescans or charge them the wrong price.

This will negatively impact the customer experience by slowing them down and forcing them to locate a store employee. For this reason, is it critical that retailers invest in ML-driven fraud detection that can keep theft down without impeding on the efficient self-checkout experience for reliable shoppers.

How retailers can use ML

Instead of using static rules, retailers can rely on ML-based fraud detection technology that considers a wide breadth of data points before requiring a rescan. These tools evaluate the retailer’s ever-changing assortment, the shopper’s basket, items commonly purchased together, as well as current and historic data before calculating the likelihood of there being an irregularity in the basket.

If the probability of fraud is high, the system will prompt the kiosk to request a rescan. If the rescan was “successful” and the irregularity is no longer of concern, the system will learn from this mistake and the same scenario will not happen in the future.

Not only do these types of ML-based systems adapt to product assortment changes and customer reliability, they also learn new fraud strategies and automatically