Why Predicting Customer Behavior Is Problematic
BY Erik Sherman
Using data to predict your customers' moves is a tricky art and science--and it's not for the uninitiated.
A new start-up, Grokr, is in the beta stage of a mobile search service that it hopes will eventually help make mobile search "autonomous, predictive, and personalized in order to provide people with the information they need to know, just when they need it." By letting user behavior fuel predictive analytics, the company aims to alleviate much of the data entry required to do a Web search to free people up for more useful things. (Like playing the next level of Angry Birds?)
Predictive analytics are all the rage. They can do amazing things: improve weather forecasts, let retailers better find fraud attempts before theft actually occurs, identify potential new markets for a product, and help stop crime by sending officers to where it is likely to happen.
And yet, like anything, you can be harmed by too much reliance on the technology. Take the Grokr example, for a moment. Apps that find things for you can be a time saver. Google does it on a much larger scale. Facebook uses predictive analytics to try to determine what ads might most appeal to a given user.
But prediction is still a difficult science and art, and gets even harder when you try to narrow the field to what one given person would want, versus a likely outcome of a majority of a large group of people. The more focused on the individual, the more individual quirks can complicate matters. The statistical methods start to break down. Plus, depending too much on predictive analytics might mean your customers miss out on the benefits of accidents that can't be predicted.
Making analytics work well is difficult even for experts in the field. It can be foolhardy for the barely initiated. We're talking about a playground for PhDs in mathematics and data science. Many entrepreneurs will want to use predictive analytics to improve their businesses and add a cutting edge to their products and services. But getting the right help can be difficult at best.
As someone in the industry was telling me the other day, there aren't enough experts in the field to come close to satisfying all the demand. Yes, there are increasingly expert system tools that can do much of the work, but using them is like trying to do statistics with an Excel spreadsheet. You can get things to work, but getting them to work out right means really knowing what you're doing.
Make no mistake, this is powerful technology that every entrepreneur should learn. Just be careful not to underestimate how badly things can go wrong and how much experience and knowledge you need to steer them the right way.