When it comes to building predictive analytic models, the mainstay for many businesses has long been “general linear models.” But those models, while valuable for a large number of problems, are only marginally predictive for many others. It’s not uncommon that false positives dramatically outweigh true positives, undermining the usefulness of the models.
More recently, machine-learning techniques, which in many cases aren’t particularly new, have become dramatically more practical due to the ready availability of computing power and a much wider variety of data sources.
As a result, machine-learning models commonly have fundamentally better predictive power. Here are examples from some of our recent work with clients:
- A global retailer used an advanced machine learning technique to forecast customer demand and found that the new method cut its forecasting error in half.
- A bank used a machine-learning method to analyze its collection activities and learned it could eliminate more than 40% of calls to customers because it knew much more precisely who was likely to repay, who needed a nudge, and who needed financial help.
- A telecom company found that state-of-the-art machine-learning methods yielded a 75-fold reduction in “false alarms” for customer churn. As a result, the firm could lavish far more attention and resources on customers truly at risk of leaving.
Better predictive power is valuable in and of itself. But, in a number of cases, machine-learning techniques make it possible to build action programs with good ROI that would have been impossible with old, more weakly predictive, techniques. For instance, consider the telecom example above: Weakly predictive traditional models would have wasted more money on false positives than they would have gained by reducing churn. With dramatically fewer false alarms, machine learning reverses that pattern.
Beyond the direct value for a company applying predictive analytics in a particular decision, a lot of companies operate in a zero-sum environment—when you win, your competitor loses. Models that allow better targeting, acquisition, and retention of high lifetime-value customers starve competitors of that business. That simultaneously increases your capacity to invest in your customer value propositions, and diminishes your competitors’ ability to do the same, levering open an expanding performance and capability gap that is hard to close. The reverse is also true: If your competitors establish a lead in applying these new approaches to modeling, your long-term pain can be substantial.
Given the power of machine-learning techniques to significantly influence competitive dynamics within an industry, sectors will see players split into two divergent camps—those that master modern machine-language methods and apply them to their most pressing business challenges, and those that fail to do so. Over time, the gap between the two camps will only widen. In our view, adopting machine learning is both a necessity and a race. Bringing these approaches to your business before your competitors do is a strategic imperative.