Sainsbury’s has discovered an innovative solution to retail theft: eject paying customers so aggressively that they never return. It is a bold strategy, and it is working exactly as unintended.
The chain’s AI anti-shoplifting system recently flagged Matt Arnold as a potential thief. Arnold was not stealing. He was shopping. The system, apparently trained on a dataset that treats all humans as suspicious, recommended his immediate removal from the premises. Sainsbury’s complied. The branch has now paused the program, which is corporate speak for “we got caught.”
Here is the beautiful economic logic at play: retailers invest millions in AI to prevent £500 worth of stolen goods per store annually. The system then ejects a legitimate customer who was about to spend £40 on groceries. That customer tells ten friends. Those ten friends tell their friends. Suddenly, your loss prevention technology has become your loss acceleration technology.
The real genius is that Sainsbury’s has inadvertently created a new market: wrongful ejection settlements and apology vouchers. Matt Arnold now has leverage. Other falsely flagged shoppers will follow. Soon, retailers will be paying out more in compensation than they ever saved on theft.
The lesson here is not that AI is bad at detecting shoplifters—though it clearly is. The lesson is that when you choose between trusting your customers or trusting an algorithm, the algorithm will always choose maximum caution. It has no skin in the game. It does not care if you shop there again. It only cares about being right, even when being right means being catastrophically wrong.
Sainsbury’s has paused the program. Smart. Some problems are best solved by humans who actually want your business.