The knockoff handbag was a near-perfect replica of the real thing, but at $175, it was about $1,500 cheaper. So the ol’ switcheroo was made. The designer handbag was purchased online, and the replica was sent back for a refund.
“Fraud is the dark side of convenience,” said David Sobie, co-founder and CEO of Happy Returns, a UPS returns management company that processes returns at a warehouse in Valencia, California. The company has started using AI to help weed out fraud, employing cameras, computer vision and software to identify suspicious merchandise.
The problem becomes particularly pronounced during the holidays, when extended return windows and a surge in online purchases create more opportunities for deception. Fraud prevention company Signifyd reported that fraudulent and abusive returns accounted for 12% of returns during the first 26 days of January 2026.
While human inspectors remain part of the process, some replicas are difficult to distinguish from genuine products, even by the best-trained eye. That is where AI comes in, identifying subtle discrepancies that human inspectors might miss.
“Using Return Vision software, each product is photographed and compared against photos from the merchant’s online catalog,” Happy Returns states, referring to products selected for additional inspection. “Our AI model is trained to catch what the human eye might miss: incorrect or missing logos, a strap that’s too wide, a mislabeled tag, or fabric that doesn’t match. These subtle signals are often the difference between a genuine return and a costly decoy.”
According to the National Retail Federation’s 2025 retail returns report, roughly 9% of all returns in the United States are fraudulent, representing more than $75 billion in merchandise. And the schemes are becoming increasingly sophisticated.
“It’s not just the scale of fraud that is growing, it’s the sophistication,” Happy Returns states. “From straightforward empty boxes to elaborate schemes involving decoy items, price switching, and false tracking IDs, fraudulent activity is growing in prevalence.”
False tracking IDs can involve manipulating shipping labels or tracking information to make it appear that merchandise has been returned when it has not.
Those losses can ultimately affect consumers through higher prices, tighter return policies and additional fees as retailers attempt to recover their costs.
The schemes range from relatively simple violations of return policies to deliberate theft. One practice, known as “wardrobing,” involves purchasing clothing, wearing it to an event and returning it as unused for a refund. Other schemes involve sending back older electronics in place of newly purchased models, substituting counterfeit merchandise or returning empty packages while claiming the original products were enclosed.
At the Happy Returns warehouse in Valencia, merchandise is unloaded and directed to sorting stations, where employees inspect products and automated equipment helps route them back to retailers. But alongside those human inspections, the company’s AI-powered Return Vision system examines photographs of returned merchandise flagged for additional review, searching for subtle discrepancies that could indicate a substitution or counterfeit.
A sweater, for example, might have the wrong knit pattern, an altered neckline or a zipper made from a different material. A counterfeit garment might feature an inaccurate logo, missing buttons or fabric that does not match the original product.
By comparing those visual characteristics with images from a retailer’s product catalog, the software can identify mismatches that might escape an employee examining hundreds of returns during a shift.
The technology does not independently determine whether a customer has committed fraud. Instead, it alerts employees to suspicious merchandise that warrants closer inspection. And while the system can detect substituted or mismatched products, Happy Returns has not established that it can reliably distinguish clothing worn to a party from clothing merely tried on at home.
Another layer of protection begins before merchandise reaches the warehouse. Happy Returns uses behavioral risk scoring to examine a customer’s return frequency, timing, location and transaction history. Unusual patterns can trigger additional scrutiny, with refunds temporarily withheld until the merchandise is inspected.
The company says fewer than 1% of returns in its network are flagged for additional review, allowing most legitimate shoppers to receive prompt refunds while concentrating inspections on higher-risk transactions.
A 2026 report from Happy Returns and fraud intelligence company Wyllo found that mail-in returns experienced six times the fraud rate of merchandise returned in person without packaging or labels. The companies also reported that verifying merchandise at drop-off reduced fraud by at least 85% compared with mail-in returns.
For retailers, the appeal of AI is straightforward: reducing losses without subjecting every customer to an exhaustive inspection. For consumers, the technology could help preserve the convenient return policies that have become central to online shopping.
A suspicious return is not necessarily fraudulent, and a discrepancy might result from a packaging error, an incorrect shipment or an innocent mistake. Human judgment remains essential, particularly when the evidence is inconclusive.
As holiday shopping accelerates, retailers are increasingly turning to AI to help distinguish legitimate returns from deliberate deception, hoping to draw down the billions of dollars in fraud.
