After Donald Trump was elected president last month, many pollsters had to eat their words. In fact, one prominent expert, Princeton professor Sam Wang, went on live TV to follow through on a pledge he had made before the election: that he would eat a bug if Trump won. “After all, I was wrong. A lot of people were wrong,” said Wang, before munching on a cricket.
All bets aside, the election results were a jarring reminder of the limitations of polling as a science. Polling is a small numbers game. In order to predict the outcome of an election, pollsters divide up voters into demographic categories (age, geographic location, ethnicity, socioeconomic status, and so on). Based on responses from a small subset of people in each group, they infer what the population at large is going to decide.
In the era of big data, that sounds a little backwards.
If your company is basing its online sales strategies on similar calculations, it’s got it all wrong. The best way to determine the outcome of a consumer is to watch what he or she does across different devices, and across time, and map that onto past observed behaviors. The more data you have on prior customer behaviors and their outcomes, the more accurate your next prediction will be.
This type of demographics-agnostic calculus is what’s called behavioral analytics. With it, you’ll never have to eat a bug on live TV.