In my day job as a marketer, I have recently undertaken a particularly daring project. I am working on a model to predict customer actions, specifically which customers will defect in the future. However, after reading up on the science of prediction, I’m starting to wonder if I’ve gone on a fool’s errand.

Prediction, as it turns out, is an extremely elusive art and the path to accurate forecasting is strewn with broken promises, false starts, disheartened scientists, and dead hedge funds. It has been said that a person could predict with absolute certainty (i.e. knowing) what number would be rolled in a craps game, if he or she knew all every detail of the die roll, such as the size and shape of the die, the material and lay of the table, the direction and velocity of the throw, and so on. The same could be said for virtually any physical action, from predicting home run swings to forecasting traffic accidents. Unfortunately, we neither have access to the entire universe of relevant data concerning the die roll/baseball swing/vehicle motion nor a computer powerful enough to process all the information and spew out an answer. And, in case we are not disheartened already, all this just applies to physical actions, not the infinitely more complex analysis of emotionally driven human actions. Customer loyalty most definitely falls in the human action category.

All is not lost. While we cannot predict things with certainty, we can ascertain the probabilities of things happening. In fact, there is rich history of calculating probabilities and, with a little searching, one quickly finds success stories from fields as diverse as quantum mechanics and behavioral finance. In fact, good statisticians have been doing this kind of probability modeling for a long time. Its interesting that predictive analytics is absent from the traditional toolset of marketers. I can certainly imagine the reason for this oversight — marketing people are notoriously shy around numbers. Now, I believe that marketers who can’t handle spreadsheets and math equations are a dying breed, but that discussion is best left for future blog posts. The point is marketing can benefit from statistics in many ways. What marketer would not want to know which customers are likely to abandon the company’s services next month?

My simple mission of predicating customer defections doesn’t seem so impossible anymore, and, furthermore, I won’t even need access to a supercomputer to do the job. I never expected to predict outcomes with absolute certainty, just to know more about future customer behavior than I know now, which is very little. I’m looking forward to plugging in a few variables culled from my web analytics reports and seeing if there is link to specific customer behavior. I’ll be bouncing off the walls with any positive results and won’t even worry about the whole correlation versus causation issue until later.