Target's algorithm knew a teenager was pregnant before her father did

Target statistician Andrew Pole analysed purchase histories from women who had signed up for Target baby registries and identified about 25 products whose combined buying patterns predicted pregnancy with high accuracy. Unscented lotion in the second trimester, calcium and zinc supplements in the first 20 weeks, large quantities of scent-free soap and cotton balls near delivery. The model assigned each shopper a 'pregnancy prediction' score and estimated a due date, letting Target mail coupons timed to specific stages.

The model's accuracy became public through a story that circulated widely after the New York Times Magazine reported it in February 2012. A man walked into a Target outside Minneapolis and demanded to see a manager: his teenage daughter had received mailers with ads for maternity clothing, nursery furniture and smiling infants. He accused Target of encouraging her to get pregnant. The manager apologised. Days later, the father called back. 'She's due in August,' he said. 'I owe you an apology.'

Target realised that demonstrating this knowledge openly could alienate customers. Pole told the Times: 'Even if you're following the law, you can do things where people get queasy.' Target began mixing baby-product coupons with unrelated ads — lawn mowers, wineglasses — so the mailings appeared random. The company had built a system that worked perfectly and then had to disguise the fact that it worked.