The Dangers of Using Predictive Analytics to Gauge Employee Flight Risk

​The notion that organizations can predict when top performers are considering moving on to greener pastures and then intervene in ways that convince them to stay is alluring. An entire category of predictive analytics tools has sprung up to make that sooth-saying vision a reality. These flight risk models have grown popular as organizations grapple with a labor market where it's increasingly difficult to find good replacements when employees leave.

Predictive analytics software that's designed to forecast employee flight risk typically measures factors like engagement levels at work, time since last promotion, absenteeism, changes in performance review ratings or even data like commute time and activity on LinkedIn profiles. These models often use algorithms that correlate the patterns of employees who have voluntarily left the organization with those of existing workers to create a flight risk score.

Used in the right manner, such data can help companies identify the top factors causing employee attrition and take preemptive steps to help stem the unwanted outflow. But experts say data that emerges from flight risk models is too often used in suspect or even harmful ways and managers aren't always trained in how to apply these sensitive findings in a productive fashion.

[SHRM members-only platform: SHRM Connect]

Pitfalls of Predictive Tools

One problem arises when companies choose to identify the individual employees deemed at risk of leaving rather than reporting anonymous data to managers or executives.

"A manager can never un-know that information or un-see it," said Helen Poitevin, a Paris-based vice president and analyst who covers HCM technologies for research and advisory firm Gartner. "Will having that information now influence every decision or interaction they have with employees who are deemed flight risks?"

Ben Eubanks, principal analyst for Lighthouse Research and Advisory in Huntsville, Ala., said using analytics to project and head off employee attrition can be a powerful tool but should be used with care.

"Employers need to be careful not to use these analytics as a decision tool but rather as a decision support tool," Eubanks said. For example, if managers have employees who are considered a flight risk, they may not offer them a new training course or development opportunity, even though data says that growth opportunities are a major factor in keeping many employees loyal to an organization.

George LaRocque, founder and principal analyst of HRWins, an HR technology research and advisory firm in New York City, said how effective flight risk assessments are depends both on the quality of the data generated and how that information is used by managers.

 "Managers typically haven't been trained in how to have difficult conversations with employees," he added. "It's a scary proposition if you are identifying or naming individuals who you think are flight risks but not providing a lot of guidance to managers in how to engage with their people about red flags that may or may not indicate a high intention to leave."

LaRocque cited the example of using commute time as one predictor of employee attrition. "Maybe an employee is going through something difficult in their life and has to temporarily move in with a friend or family member, which adds 30 minutes to their commute time," he said. "Factors like those can be false positives. Data collection and analysis needs to be rigorous and managers should be able to have conversations with their people in ways that don't put them on the spot and end up actually forcing attrition when it could have been avoided."

There's also the matter of the accuracy of flight risk tools. Poitevin is skeptical of some vendors' claims that they can predict with 90 percent or better accuracy which employees are likely to depart in the near future.

"For flight risk, a 70 or 80 percent accuracy level is generally what you're looking for, given the many variables involved," she said. "There's a danger in having too many false positives in these predictions. You have to allow for some error when there are factors you don't know or may not have quality data about, because human behavior can be unpredictable."

How to Use Flight Risk Data Productively

Such pitfalls can be avoided if organizations commit only to reporting aggregated flight risk data rather than identifying individuals, Poitevin said. "When companies are very careful about the data sources they're using to make these decisions and when they're using the data to inform better overall strategies about improving employee retention, then the tools can have value," she said.

LaRocque agrees that such anonymity in reporting is key. "It's often more valuable for managers to know they have some issues they need to look at holistically on their teams," he said. "Knowing you have some flight risks on your team in general may make you approach things like your management style, developmental opportunities for individuals or workforce planning differently."

Matthew Stevenson, a partner and co-leader of the workforce strategy and analytics practice with the consulting firm Mercer in Washington, D.C., said flight risk models that rely only on internal and not external data also can be problematic.

"Many of the datasets only contain information about what's happening internally with employees and say nothing about what the competition is doing, what the market pay rates are for similar jobs or external labor market data like the unemployment rate," Stevenson said, factors that can weigh heavily in employees' decisions to stay or move on.

Dave Zielinski is a freelance business writer and editor in Minneapolis.