Five Myths about Predicting Workforce Behaviors

The science of predictive analytics, commonly used in corporate financial planning, is missing from HR at most companies. Here’s why.
David McCannMarch 22, 2013

CFOs may want more analytical muscle from human resources, but at most companies that’s a far-off dream.

PwC Saratoga, a PricewaterhouseCoopers unit focused on workforce measurement, defines four levels of HR analytical capability. Level 1 is producing ad hoc metrics and reports that tell “what happened.” Next up the ladder are descriptive benchmarking and dashboards, which get at “what happened, and how do we compare with others on a defined set of metrics?” Then comes advanced survey analytics, which is about “why did it happen, and how/where can we improve?”

Level 4 is creating predictive solutions, or “what is likely to happen, and how can we be better prepared?” That’s the kind of work financial planning and analysis teams routinely engage in. But the same is rare in HR, even though the department is sitting on a mountain of data that could help make all kinds of useful predictions: How will job applicants with certain profiles perform in various roles? Which high performers are most likely to leave the company? Who has leadership potential? How effective will training curricula be? How often will people be absent from work? When are various types of people likely to retire?

The 7 Habits of Highly Effective CFOs

The 7 Habits of Highly Effective CFOs

Download our whitepaper to discover the technical and behavioral skills needed to lead your business forward.

Yet in a 2012 study of 383 organizations by PwC Saratoga, just 12% said they had reached Level 3. So few are at Level 4 that the survey didn’t bother to ask about that. Ranjan Dutta, a director in Saratoga’s human-capital metrics and benchmarking practice, says he expects less than 1% were that advanced.

It’s not only finance that out-analyzes HR. For example, “Look at how the marketing function uses predictive analytics all the time to predict customer behavior,” says Dutta. “But marketing has far less information about customers than HR has about employees, which, after all, are a captive audience for employers.”

There is employee demographic data, benefits information, information from employee surveys on engagement and other workplace topics, performance reviews, productivity measurements, compensation information, training results, exit surveys, and even hours worked and vacation days used. Such information can be merged in a data warehouse and, through such advanced analytics techniques as multivariate regression, decision trees, and time-series analysis, be used to answer questions such as those listed above.

What prevents companies from using predictive analytics for purposes of managing human capital? In a webcast on Wednesday and in a follow-up interview with CFO, Dutta argued that the big hang-ups are around five interrelated myths about predictive analytics.

Myth 1: We (HR) have not matured enough to do predictive analytics. Although there is a maturity scale, workforce analytics is not sequential in nature, and it’s not necessary to be fully proficient in reporting and dashboards before progressing into predictive analytics, says Dutta.

Reports and dashboards are the products of descriptive analytics, which is about slicing and dicing data and recognizing patterns. Predictive analytics is about testing hypotheses. “They are parallel processes,” Dutta says. “Even if you become very proficient at descriptive analytics, there is no guarantee you’ll ever be able to do predictive modeling.” But predictive analytics does not require organizational maturity; it requires finding people with predictive-analytics skills (see Myth 5 below).

Myth 2: We don’t capture enough data to do predictive modeling. Predictive modeling is not about having the most data, it’s about testing relevant hypotheses with the relevant data. Says Dutta: “Many people think, ‘Our transactional systems — our ERP [enterprise resource planning] and our HRIS [human-resources information systems] — are really poor at measuring things. So we can’t do predictive analytics.’ But those systems are not designed for that purpose.”

The answer, he says, is to look at business problems you’re facing, create some hypotheses for how to attack the problems, decide what data elements you need to test the hypotheses, and see whether that data is available. “At the end of that exercise, you might discover you really do not have the relevant data pieces,” Dutta says. “But just blindly assuming you are not ready? That’s no good.”

Myth 3: We need to make big investments in data technology to do predictive analytics. HR organizations can start building advanced analytics capabilities without relying on advanced technology solutions. Using an example similar to that in Myth 2, HR leaders may believe they need to achieve full, end-to-end integration of their ERP and HRIS in order to tap the data they will need. That could well be a very expensive process, but you don’t have to do it for purposes of predictive-HR analytics, says Dutta. Again, it’s not necessary to get all the information stowed in those transactional systems.

“If you do a predictive-analytics pilot, you might get in a much better position to decide what fields from which databases you actually need to aggregate into a data warehouse,” he says. That may actually turn out to be quite affordable.

Myth 4: We can simply buy a predictive-modeling capability by investing in advanced HR business-intelligence solutions. This is a close cousin of Myth 3, or perhaps a subset. Says Dutta: “Some BI vendors will say, ‘Buy this expensive application and it will automatically spew out predictive solutions.’ Or ‘Spend $15 million to build your data warehouse, and you’ll have predictive modeling.’ No. If you have the right people with the right skill sets, you don’t need a huge investment.”

But won’t companies below a certain size threshold be unable to afford even a small data warehouse and some predictive analysts? To some extent, yes. But, referring to midsize companies, Dutta says, “You don’t need to spend $15 million. Maybe you need to spend $1 million.” And the cost for the people to do the modeling will be a fraction of that.

Myth 5: We need to hire a group of statisticians before we can do predictive modeling. This is only a partial myth. Ultimately, statisticians will be needed to build predictive models. But they probably won’t think like businesspeople. The most important skill set for deriving value from predictive analytics involves interpreting the results to arrive at business insights and linking them to actionable workforce decisions.

Most companies won’t find people with those skills in HR, Dutta says, but rather in finance or in business units. Some organizations are bringing in people from those areas to play this role for HR. “Those people are used to working with data and complicated models and converting them into business language. HR is not traditionally a data-intensive function.”

Meanwhile, Dutta stresses that falling short of Level 4 does not necessarily mean companies aren’t pursuing predictive analytics. “Those at Level 3 may be doing so in some form or other but are simply not ready to build, deploy, and maintain predictive solutions on the fly,” he says.

Understanding Which ERP Modules Your Business Needs – And When