Sponsored by Appriss Retail
Customer self-checkout, in its many evolving forms, is plainly the future of retail for many store types. Millennials have a clear preference for it, and it’s nearly impossible to imagine that an even more tech-savvy generation to follow will yearn for a return to the good old days of queues and cashiers.
Yet, while it’s certainly the future, self-checkout is also a mature technology with persistent, unaddressed problems in need of a fresh fraud prevention approach if retailers are to capitalize on the efficiency gains it offers.
In the most recent effort to take consumers’ temperature on the subject, 73 percent of 526 shoppers surveyed by SOTI said they prefer retail self-service technologies, such as self-checkout, over engaging with store associates, an increase of 11 percent from a year ago.
Stores in the United States are increasingly responding to that demand, noted Nathan Smith, senior vice president of products at Appriss Retail, a leader in data and analytics solutions for retail organizations. “The adoption rate is increasing worldwide, not as fast in the United States as in Europe, but it is accelerating,” he explained. “There is a strategic business driver to meet consumers’ desire to get in and out quickly, to check people out on different devices or even their own device, and it really represents a win-win for embattled retailers in brick and mortar, because it is an opportunity to reduce headcount and for labor cost-savings in addition to giving consumers what they want.”
Though not a deal breaker, there is a significant drag on this ‘everybody wins’ proposition: namely, theft at the point of sale. Losses from theft at self-checkout are, simply put, staggering.
In a study of theft associated with fixed self-checkout, stores with 55 to 60 percent of transactions going through those systems experienced 31 percent higher losses on average. Grocery store case studies showed losses can be as much as 147 percent higher (Self-Checkout in Retail: Measuring the Loss, ECR Research Paper, Oct. 2018). “It’s still a net positive for them, so retailers are rolling ahead with self-checkout, but they’ve come to view the increase in shrink as an expense associated with implementing the technology, as a cost that they just have to bear,” said Smith. “What we’re looking at now is that maybe we don’t have to assume that those losses are a given.”
Appriss Retail is aiming to apply a new loss prevention approach to self-checkout iterations, augmenting or replacing cameras that try to observe fraudulent behavior with artificial intelligence that can spot anomalies in the transaction data, flag suspicious transactions, and alert staff to audit suspect shoppers.
“From our perspective, the use of computer vision to see when items are not being scanned or the switching of bar codes seems difficult to scale across an entire retail chain,” said Smith. “We think there is a smarter way to look at this as data.”
There are clear advantages to a data-driven approach over human or video surveillance. It can work however shoppers do checkout, in aisles, at fixed machines, or any other variant of ‘cashier not present’ transactions. It operates seamlessly in the cloud and requires no hardware in the store, but can integrate with other technologies, including cameras, to make even smarter decisions about which transactions to audit. Finally, analysis can be inverted so as to reduce consumer friction during transactions that are almost certainly legitimate, such as eliminating weight checks or void restrictions, thus rewarding loyal, honest customers with fewer frustrations during self-checkout.
A simplified example highlights how this approach would work: A thief with a bar code for a can of pears taped to his watchband scans his grocery cart full of filet mignon, infused olive oil, and other pricey items. As he passes each item over the scanner, he deftly rings up the $1.84 pear price instead. It’s nothing nefarious to the eyes of an associate or a camera, especially if weight checks are turned off in pursuit of lower friction, but to an intelligent audit mechanism using historical and real-time data, it warrants an alert to a store associate to audit the transaction: Nobody buys 21 cans of pears.
Of course, not all schemes are so obvious. But the constant feedback loop of data and audit results make the challenge of identifying suspect transactions a perfect target for artificial intelligence, according to David Speights, PhD, chief data scientist for Appriss Retail. He and his team of 10 PhDs have been doing research on the problem for more than a year. He’s encouraged by their progress and, now, cautiously optimistic. “We don’t chase things we think we can’t solve, and we’re pretty bullish we’re going to be able to help retailers with this problem.”
Naturally, even with AI, not every fraudulent transaction will raise a red flag—”to say that a specific transaction is definitely a fraud is terrifically hard to do,” says Speights—but he says it now seems a solvable challenge to identify pools of transactions to audit that will allow retailers to substantially improve fraud detection. Random auditing might allow a retailer to find theft in, say, 5 percent of transactions. Intelligent real-time self-service checkout auditing could bump that up to 20, 30, or 40 percent. Even at 10 percent, for a large retailer—given the huge amount of self-checkout theft—it could mean millions in savings.
Loss prevention surely looks forward to this research coming to fruition. Driven by increasing labor costs and consumer demand, cashierless store checkout systems are expected to grow significantly. “In this transition to cashierless technology, the focus of LP will need to move from the traditional approaches of looking at employee cashiers to the analyzing of customer cashiers,” said Smith. “That’s why we’re developing the next generation of tools, so that retailers can better manage those risks.”