UPS company deploys AI to spot fakes amid surge in holiday returns

LOS ANGELES, Dec 18 : Nearly one in every 10 retail items that are returned for a refund in the U.S. is fraudulent, according to a UPS-owned reverse logistics company that says it plans to combat the $76.5 billion problem for the country’s retailers with artificial intelligence. This holiday season, UPS-owned boxless returns processing company Happy Returns is testing its AI fraud detection tool with a handful of clients, including apparel sellers Everlane, Revolve, and Under Armour, Happy Returns CEO David Sobie told Reuters during a recent tour of the company's Los Angeles-area hub. These retailers are among the many U.S. companies that have been hit by return fraud, a method of stealing in which a customer applies for a refund on a retail item and ships back something else of less value, like a cheap knock-off that cannot be resold. Happy Returns' AI tool called Return Vision helps to find fraudulent returns by flagging suspicious packages, analyzing their contents, and sending them for final audit by humans who can verify the fraud and withhold the refund, Sobie said. Happy Returns specializes in no-box and no-label returns. Shoppers take unwanted items to any of the nearly 8,000 so-called "returns bars" inside Ulta Beauty, Staples or UPS stores, where workers scan, bag, and label items that are then combined in large boxes and sent daily to processing hubs, saving retailers time and money. Shoppers - and fraudsters - like the service because it is easy and often offers immediate refunds. Return Vision testing started in early November and additional retailers will start trying out the tool later this month during the holiday season's surge in returns processing, Sobie said. The effort aims to tackle a problem that compounds costs for retailers, according to Jim Green, director of logistics and fulfillment at Everlane, which sells cashmere sweaters and other clothing, primarily online. Returned items are already a drag on profits because costs for shipping the packages, preparing the goods for resale, and restocking shelves add up, said Green, who added 85 per cent of Everlane's U.S. domestic online returns are handled by Happy Returns' in-person drop-off and consolidation network. "Not getting back the real items is a double whammy. It's hundreds of thousands of dollars for us alone per year," he said. Representatives from Under Armour declined comment and Revolve did not respond. Around $849.9 billion worth of retail goods will be returned in 2025, amounting to around 15.8 per cent of sales, according to a study released by Happy Returns and the National Retail Federation, which relied on estimates provided by nearly 360 e-commerce professionals at large U.S. retailers. About 9 per cent of those returns will be fraudulent, the report said. UPS competitors Amazon.com and FedEx also offer boxless returns, while the United States Postal Service is beginning a rollout. Amazon said its returns service also uses automated tools to flag potentially risky returns and physical inspections. FedEx did not immediately comment. Executives polled by a variety of large consulting firms say they believe generative AI will eventually transform their businesses, but are reconsidering how quickly that will happen within their organizations, Reuters recently reported. To that end, 85 per cent of merchants responding to the Happy Returns/NRF survey said they were using AI or machine learning to identify and combat fraud, and that results have been mixed. Happy Returns executives say their AI program only helps identify when incorrect items are being returned. It does not address other issues, such as "wardrobing" - when customers return items that they have worn or damaged. HOW IT WORKS Happy Returns said its AI fraud detection tool starts working the moment a shopper initiates a return online. It flags things like returns initiated before or shortly after an item is delivered, entries from shoppers that have multiple linked email addresses, and returns from individuals involved in previous suspicious activity. Workers at drop-off points have access to photos of the items that should be returned when they scan the unwanted goods into the Happy Returns system. They can reject obvious mismatches. "Sometimes humans don't always catch small differentiation points between an item that's been returned and the item that was purchased," CEO Sobie said. Once the returns reach the Happy Returns hubs in California, Pennsylvania, and Mississippi, human auditors open the flagged packages. They take photos, which are then fed back into the AI tool, which compares them to images and other information about the goods expected in the return, the company said. Human teams then review the AI assessment and make the final call. "If you're returning a pair of $300 boots and you show up with a pair of dirty old sneakers, that should be caught immediately. What Return Vision does is add an extra layer of protection for some of the not-so-obvious cases," said Green. Less than 1 per cent of returns in the Happy Returns network are flagged by the tool as high probability of being fraud, and about 10 per cent of those flagged items are ultimately confirmed as fraud, the company said. The average value of each fraud is around $261. Happy Returns Chief Operating Officer Juan Hernandez-Campos said the tool was becoming increasingly important as fraudsters become more sophisticated. "Bad actors adapt. We need to adapt too," he said.