UCL and dunnhumby Unveil Brand-Switch Predictor

Scientists at University College London (UCL) and consumer data giant dunnhumby have developed a model which they claim predicts when people are most likely to try switching to different products.

The team analysed anonymous purchase data from more than 280,000 shoppers who regularly bought products in six categories: beers, breads, coffees, toilet papers, washing detergents and yogurts. They found that the more people buy a product, the more likely they are to continue to do so. However, thousands of participants were then sent coupons, and the new model was used to predict which individuals would use them. According to Professor Brad Love, UCL Psychology, people who had recently switched brands were twice as likely to use the coupons to try a new product.

Love says previous studies using monetary rewards have found the opposite pattern; the longer someone has chosen to receive a reward of known value, the more likely they are to choose an uncertain reward on the next try. The latest research shows that this is different with subjective rewards, where choices drive preferences rather than preferences driving choices. The study also found that shopping behaviour was consistent across the different product categories. For example, people who were more likely to 'explore' new beers were also more likely to try new detergents.

Peter Riefer, who conducted the study for his PhD at UCL and now works at dunnhumby, explains: 'If people show a particular shopping pattern for one product type, they tended to show it for the others as well. This suggests that people have individual 'exploration personalities' when they shop, which is really remarkable given how different the six products were. Over a timescale of many years, the rate at which people explore is remarkably stable - people are always exploring at about the same rate. What we find is that within that stability there are these trends in exploration and exploitation'.

Web sites: www.ucl.ac.uk and www.dunnhumby.com .