It ain’t guessing anymore, folks
An interesting article, “Guessing the Online Customer’s Next Want” by Eric Taub, appeared last week in The New York Times. Taub starts out on the right track when he says, “Marketers have always tried to predict what people want, and then get them to buy it.” Many marketers do try to predict. Unfortunately many more don’t bother trying. As a result, you and I are inundated with uninteresting direct mail and email. Happily, Taub focuses on a retailer that does try, and on the methodology being used, collaborative filtering. Calling attention to this effort is worthwhile, and I’m pleased to see a spotlight on predictive analytics.
However the headline and the article itself miss a larger point. They suggest that only in the online world is mathematical marketing supplying customer insight. That’s simply not true. Advanced companies have been using all of their transaction data to predict their customers’ purchasing behavior for some time. Valuable insights are mined from in-store sales and catalog sales, even more than from ecommerce. When a customer pulls cash from her wallet to make an over-the-counter purchase, transaction information is collected and can be analyzed to predict future purchases. Savvy companies, using loyalty programs, can identify over 80% of their retail customers at point of purchase, even if the customer doesn’t use a credit card. And it’s easy to track mail and phone orders since the goods have to be shipped somewhere.
Customer analytics makes accurate predictions
Another problem is the implication that we’re ‘guessing’ when we predict what someone is likely to buy. Reporting the past is more accurate than predicting the future, but the accuracy of predictive analytics is a long way ahead of guessing. Today we can accurately predict which customers are likely to buy in the near future and what products they will likely select. On a higher level we regularly predict the response rate and the revenue from a marketing campaign before the campaign is launched. Further, we can quantitatively measure the accuracy of these predictions, and that accuracy is very good.
Take risk assessment, for example, where we predict the likelihood that a particular customer will defect to a competitor. We routinely review our predictions and measure their accuracy using a correlation coefficient that ranges from minus one to plus one. Minus one means we are totally wrong. A value of zero means our predictions are no better than chance. Plus one means we were right every time. We typically get values of 0.5, which is very good indeed. If you had that kind of accuracy when betting on horse racing or football games, you would get very rich very quickly.
So why not do the same with your marketing? The kind of mathematical marketing I’m describing predicts a customer’s next move largely based on his or her own behavior. Making a targeted offer of a specific product that is attractive to your customer is one of the most reliable ways to improve marketing campaigns. If mathematical marketing can help you do that, how smart are the marketers who ignore it? Or the ones who just “guess”?