How AI is trying to influence consumer behaviour in supermarkets...
24 February 2022

Interview with 

Kanta Dihal, University of Cambridge


A supermarket aisle empty of people.


AI is having more and more influence on the way the world operates each year, conveniencing our day to day lives in ways we could never have imagined. It can also be used, however, to encourage our behaviour for the benefit of companies after our cash. Kanta Dihal lifts the veil for Chris Smith, on how supermarkets are attempting to alter our shopping habits using patterns identified by artificial intelligence...

Kanta - I wonder how many people are aware, for example, of the ways in which AI related software and algorithms work in supermarkets. My favourite example in this comes from Hannah Fry and her book on algorithms. TESCO, which for my American fellow panelists is one of the biggest supermarkets in the UK, has been collecting data on what people buy in the supermarket for decades and has been using that information to identify trends. For example, people who buy 'xyz' are also more likely to buy 'abc'. Therefore if you put one on offer, but raise the price of the other one then you might end up selling more and making more of a profit. Or maybe we should put one specific product closer to another product on the shelves, because people always buy them together. My favourite example, which Hannah Fry described, was when they started combining this with offering mortgages and loans. They discovered that people who buy fresh fennel are more likely not to default on their loans.

Chris - Does she offer any kind of reasoning why or is it just that very middle class people who are probably better off buying fennel and they've probably got a bigger bank balance?

Kanta - That's probably a very large part of it. There's lots of things coming together here about people being middle class and being traditionally middle class, having grown up knowing what fennel is and how you use it along with having enough time to cook it and enough money to purchase it. Lots of assumptions about class background and ethnicity come together here that influence those correlations. These are kinds of discoveries that these algorithms make that can reveal biases or assumptions that we didn't even have, or that we didn't know we had, but that become extremely stark when you just run the numbers on them.


Add a comment