Talent Management

How to Predict Which Employees Will Leave

Increasingly advanced predictive algorithms take into account everything from performance evaluations to conflict avoidance to timely task completion.
David McCannApril 17, 2019

Almost 3.5 million U.S. workers quit their jobs in January, the highest monthly total ever, according to Bureau of Labor Statistics data. With the economy essentially at full employment, retaining good employees is as crucial as finding new ones.

It’s not an issue only for human resources managers to worry about. It matters greatly to all business and functional leaders.

How Startup CFO Grew Food Company 50% YoY

How Startup CFO Grew Food Company 50% YoY

This case study of JonnyPops’ success highlights the unusual financial and operational strategies that enabled rapid expansion into a crowded and highly competitive frozen treat market. 

Today, a fair amount of science can be brought to efforts to identify which employees are at a high risk of leaving, notes data scientist Jon Christiansen.

“I’ve run countless predictive models on employee retention and customer churn across 14 industry verticals,” says Christiansen, who teaches data mining courses at Charleston Southern University and is chief intelligence officer at both business intelligence firm Sparks Research and data mining firm Ins & Outs. “The one common theme I’ve seen is that it’s easy to predict who will stay. Predicting who will leave is not easy.”

He likens that difference to the famous Leo Tolstoy quote from his novel Anna Karenina: “All happy families are alike; each unhappy family is unhappy in its own way.”

A key predictor is whether an employee’s performance is fairly evaluated. Christiansen offered the following anecdote.

A bank in California set up a bonus structure around sales of a new credit card. When results started rolling it, all of the top salespeople were in Orange County. The bank’s initial assumption was that all of its best sales talent was located in that area.

But after Christiansen constructed a model to predict salespeople’s performance, it became clear that was not the case. He factored in income levels and credit ratings in the bank’s various locations, foot-traffic levels at branches, the percentage of customer interactions that were for servicing customers vs. sales opportunities, and activity data from the salespeople’s computers, as well as demographic information on the sellers’ experience.

What best predicted sales success in this case turned out to be the average credit ratings in each sales territory. Where they were higher, more credit applications were approved. The model found that a salesperson in Orange County would, if moved to a different territory, sell fewer credit cards, and vice versa.

If the bank had followed through on its initial assumption about why sales were higher in Orange County, it likely would have lost a number of good salespeople, according to Christiansen. “If they’re told that their performance is low, why would they stay?” he says.

Another key factor in employee departures that plays into predictive algorithms is whether people feel they have control of their workday. Information for making that evaluation can be gleaned from exit interviews, but also from employees’ Outlook or Google calendars.

“If someone gets an assignment on Monday to build a new marketing plan on Friday, and their calendar shows five hours of meetings on Tuesday, seven hours of meetings on Wednesday, and five more hours of meetings on Thursday, when are they supposed to do the real work?” Christiansen says. “People see that it affects their performance and ultimately their ability to move up in the company.”

There are a number of other, perhaps surprising criteria that inform the algorithms:

Is an employee avoiding conflict, or offering thoughts and ideas? A moderate level of conflict is healthy because it allows for differing views, which often leads to growth. “Someone who doesn’t feel safe to engage in conflict, or used to do so and no longer does, is showing signs of disengagement, suggesting that they don’t feel like they fit,” says Christiansen.

Is an employee subject to a moderate amount of pressure? Conventional wisdom might suggest that employees don’t do well with pressure. That’s not accurate, according to Christiansen.

“Studies show that a moderate level of pressure is healthy,” he says. “The key is moderation. Too much pressure, and an employee might feel like they’re in the Marine Corps. No pressure at all, and they might wonder if anything they do matters.”

Is an employee focused more on rewards or the organization’s common goal? Employees that target rewards over the common goal don’t do well in the long run, says Christiansen. Research shows that most people value relationships and experiences over material outcomes. “I’m not suggesting incentives don’t work, but this is another consideration for moderation.”

Is an employee in a silo or trying to fight their way out of a silo? Companies can put employees in silos, but employees can also put themselves in one. That signals disengagement and a sheltering of information from others. “One study found that the number-two thing that makes us question our well-being in the workplace is being alone,” Christiansen says.

It’s “probably worth noting,” he adds, that the number-one item on that list is “the immediate presence of a superior.”

Does the employee take an inordinate amount of time to perform a task? When this happens, it’s likely because they have nothing else to do once the task is complete, says Christiansen.

“This is the idea of emotional suppression,” he says. “The emotion is that they’re bored. But they suppress it because they have to look busy. That’s actually more emotionally and physically exhausting than doing a lot.”