The talk warns that overzealous behavioural targeting can follow people around the web as an example of personalisation gone wrong. The talk uses Amazon's recommendations after a purchase as an example of inferring adjacent interests. The talk cites a New York Times case study about Target using purchase patterns to infer a customer's pregnancy. The talk recommends Google Analytics and its site-search data as clues to valuable pages, funnels and products. The talk recommends multivariate tests for straplines, pricing and images when tuning the experience. The talk points to Facebook advertising as an example of using information about users to target marketing.