The classic approach of many companies to talent analytics is what Pasha Roberts, chief data scientist at Talent Analytics, Corp., calls the “fishing expedition.” In it, someone at the company is asked to fish among HR records on hires, terminations, payroll, demographics, performance reviews, and so forth, and come back with “something interesting, rather than any specific objective or business outcome” writes Roberts. “Inevitably, with enough data, enough time, and an expensive visualization package, something will be found. Let it be noted that you can almost always find something of interest in any dataset. Whether it is actionable or not is another story.”

HR data are what data analysts call “input variables”—metadata that don’t give insight until they are combined with a business outcome, such as revenue or customer retention, for the purpose of seeing how they affect that outcome. However, HR doesn’t track these outcomes; the appropriate line of business does—that could mean sales, finance, customer service, marketing, operations, and so forth.

“Without measurement of actual business performance from the line-of-business itself, the analysis is fated to discover trivial relationships between input variables,” writes Roberts. That could mean you’ll discover that employees from an affluent part of town tend to get more promotions, or that employees from certain schools sell more, or that younger workers use the internal chat more than older workers. But what can you do with this information? As Roberts points out, you wouldn’t limit your hiring of salespeople to those graduating from a handful of colleges. You wouldn’t spend resources only developing workers from certain neighborhoods, just because the data shows they’re more likely to get promotions anyway. These descriptive analytics—or “glorified reporting”—won’t tell you what to do to make your business more successful.

Instead, you need to go for the “targeted win.” In this approach, HR or a line of business (or both) identify a specific problem—for example, the sales reps aren’t making their quota, there’s high turnover among customer service specialists, or technicians aren’t passing their training course. Be sure to have basic employee cost measurements in place—the costs of acquiring, training, and managing staff in each role, in addition to compensation. This information will form the basis for your predictive talent analytics model, allow you to place a value on employee turnover, and help you prioritize your targets.

Then, go diving for the pearl. “It seems obvious,” writes Roberts, “but only by considering outcomes in the analysis…can one predict outcomes.”