This post originally appeared on LinkedIn 

Long ago, one of my statistics professors in college cautioned me that statisticians tend to obsess about creating the highest R-squared. Inflating one’s R-squared might be just the ticket for getting your results published in an academic journal, but the resulting model, my professor told us, is not necessarily the most useful.

The most useful model is one that helps people make good decisions and take action. The more complicated you make your model in search of superior R-squared—lots of variables, logarithmic transformations, and so on—the harder that model is to understand intuitively and the less often people will have the confidence to use it and learn from it. Statisticians' hearts may start to palpitate when their model’s R-squared creeps upward toward 1.0, but nobody else is really moved by the square of the correlation coefficient.

This simple piece of advice stuck with me, and years later it gave me the confidence to develop a system that asks just one question to predict promoter, passive or detractor categories. That may not satisfy some hard-core number crunchers, but it has made a huge difference—both in my career and for the thousands of companies that have adopted Net Promoter®.

In fact, we initially chose the question “How likely would you be to recommend my company to a friend” because it most strongly predicted what customers were likely to do based on their answers. Over time, we also learned that employees take their reputation seriously, so the customer’s response motivated action. Part of the power of the Net Promoter System® is that it is based on a simple model that requires one simple question, yet it drives real, actionable feedback that helps companies grow.

Most humans aren’t fired with a passion to act when they see the results of a complex, multivariate regression formula. But they do take it personally when a good customer tells them that they would not recommend them to a friend.