You’ve probably heard about “machine learning” in some context. It encompasses a range of automated techniques for predictive analytics. Engineers train computers on a set of known data, and set it off on an automated learning path. At Oliver Wyman, we’ve found that machine learning techniques are very powerful, and can be applied to a wide range of classic business problems.
For example, we recently applied machine learning to help a telco figure out which customers were likely to “churn”—that is, move to another provider. When telecos think they have identified a potential “churner,” they typically resort to bribery to keep the business—with a phone upgrade, free minutes, or extra data. If you give those gifts to customers who are already satisfied (and not likely to churn), you are throwing money out the window.
How do you predict whether a customer will churn? You hoover up all the call data for a given customer, then look at that customer’s interactions on the website and with the customer call center. With that information — and some automated machine learning models—you can pretty well predict who is likely to churn. In fact, we identified 75x fewer false positives—potentially saving the telecom 75x on wasted spending. 75x!
Take two telcos — one has figured out which customers are likely to churn, the other hasn’t. One is spending heavily for no ROI, the other is spending in a targeted way for positive results. Over time, market share shifts to the telco that’s figured out the churn algorithm. Imagine applying machine language to multiple levers in your business, whatever the problem. Very quickly, you will see exponential returns on efficiency.
