Rory Sutherland

Why does Amazon think my friend is a kidnapper?

Online data reveals all kinds of things about us that we’d never say out loud. In some cases, best hope it’s wrong…

Why does Amazon think my friend is a kidnapper?
Text settings

About four years ago, an irate father in Minneapolis walked into his local Target shop with a complaint. He wanted to know why they were sending his daughter, who was still at school, vouchers for baby clothes and cots. Were they trying to encourage her to get pregnant?

When they telephoned to apologise a few days later he was more diffident. His daughter had fessed up: a child was due in a few months. But if dad hadn’t spotted any telltale signs of pregnancy, the shop had: she’d been rumbled by her recent purchases, in particular unscented lotions and certain dietary supplements. Some algorithm had spotted the significance of a sudden change in her buying habits, and triggered the ‘bombard with new‑baby offers’ subroutine.

But if you think that’s worrying…

Customers with your recent history also purchased…
A friend had been buying a few outdoor items on Amazon, and these three recommendations appeared. Should he expect sudden murderous urges? A visit from the police? Even if the prediction was wrong in his case, it seemed to suggest there were rather more psychopaths and kidnappers in Britain than previously suspected (and also that, as a group, they were inclined to be frugal when buying balaclavas and rope but knew not to skimp when it came to choosing a good duct tape — a handy lesson there for Spectator readers).

Fortunately, there is a more innocent explanation. The incident happened at the height of the paintballing craze and, as someone subsequently explained, all three items are regularly bought by serious paintball aficionados, or by the people who run paintball centres. Phew!

Like it or not, we generate an enormous amount of data as a consequence of our daily actions. This is worrying when it’s our data but, let’s be honest, irresistibly intriguing when it’s everyone else’s: what people do online is just so much more revealing than what they say in public. The wonderful book Dataclysm by Christian Rudder captures this well in its subtitle: ‘Who we are (when we think no one’s looking)’.

Rudder should know: a Harvard mathematician, he is the data supremo of OKCupid, a huge worldwide online dating and matchmaking site. The book analyses what the millions of aggregated clicks reveal about dating preferences — and how these often differ markedly from what people claim to care about.

Buried in the book is a rather interesting fact. Within a percentage point or two, black people on British dating sites attract exactly the same level of interest as other ethnic groups. In the US, it’s a different story, with self-identified liberals showing no less prejudice towards them than anyone else. Even pinko Canada is different from Britain in this respect. Good news for us.

Scots, interestingly, are weirdly prejudiced in a different way. According to analysis published by an adult website, the single most common word used in Scotland when searching for online pornography is ‘Scottish’. It’s not an area I know much about, but this seems unbelievably strange to me. If this finding is robust, I don’t hold much hope for the future of the Union.

Rory Sutherland is vice-chairman of Ogilvy Group UK.