AI in Retail Loss Prevention

Retailers use the word shrink for inventory that disappears between delivery to the store and sale. It includes shoplifting and organized theft. It also includes employee theft, vendor fraud and plain administrative error, so not all shrink is customer theft. AI tools target several of these. At self-checkout, overhead cameras run models that recognize products and hand movements. The system compares what it sees with what the point-of-sale system recorded. Common patterns include skipped scans, moving an item around the scanner, and entering the produce code for a cheaper item. Most deployments don't call security. They show a replay on the screen or alert an attendant, which also catches honest mistakes. Vendors in this space include Everseen, and several point-of-sale providers build similar features into their checkout lanes. Behind the scenes, exception-based reporting scans transaction logs for unusual refunds, voids and discounts, and for sweethearting, where a cashier lets a friend skip paying for items. These are statistical outlier models. A flag means look closer, not proof. The most controversial tool is facial recognition against watchlists. In December 2023 the US Federal Trade Commission banned Rite Aid from using facial recognition for surveillance for five years. It found that the system had produced many false matches and that the company had not taken reasonable steps to protect consumers. The FTC also noted that women and people of color faced particular risk of misidentification. In the UK, retailers' use of Facewatch has drawn complaints from privacy groups. A key misconception is that an AI alert is evidence of theft. Cameras miss context and models make errors. Theft is also rare compared with honest transactions, so even an accurate model can produce many false alarms. Responsible deployments keep humans in the loop, don't confront anyone based on an alert alone, and limit how long data is kept.