(Rachel Silverman) Employers want to know who has one foot out the door. As turnover becomes a bigger worry—and expense—in a tightening labor market, companies including Wal-Mart Stores Inc.,Credit Suisse Group AG and Box Inc. are analyzing a vast array of data points to determine who is likely to leave a post.
The idea, say people who run analytics teams, is to give managers early warning so they can take action before employees jump ship.
Corporate data crunchers play with dozens of factors, which may include job tenure, geography, performance reviews, employee surveys, communication patterns and even personality tests to identify flight risks, a term human-resources departments sometimes use for people likely to leave.
The data often reveal a complex picture of what motivates workers to stay—and what causes them to look elsewhere.
At Box, for example, a worker’s pay or relationship with his boss matters far less than how connected the worker feels to his team, according to an analysis from human-resources analytics firm Culture Amp. At Credit Suisse, managers’ performance and team size turn out to be surprisingly powerful influences, with a spike in attrition among employees working on large teams with low-rated managers.
Human-resources software company Ultimate Software Group Inc. assigns clients’ employees, and even its own workers, individual “retention predictor” numbers, similar to a credit score, to indicate the likelihood that a worker will leave.
As the employment picture improves, companies are focusing more on retaining workers, largely because replacing them is costly. The median cost of turnover for most jobs is about 21% of an employee’s annual salary, according to the Center for American Progress, a liberal-leaning think tank.And it can cost, on average, some $3,341 to hire a new employee, according to the Society for Human Resource Management.
William Wolf, Credit Suisse’s global head of talent acquisition and development, says a one-point reduction in unwanted attrition rates saves the bank $75 million to $100 million a year.
No single piece of data predicts whether an employee will stay or go, though many employers wish it were so. Data scientists create models to predict which workers might leave a company in the near future, combining a range of variables and testing the predictions over time. They might refine the calculations depending on which variables are most predictive for a given company or group of employees.
“One of the things that people want to find is that one nugget, that key thing that correlates with someone leaving, but it is never that simple,” says Thomas Daglis, a data scientist at Ultimate Software.
Employers may not mind that some employees are at risk of leaving, though companies stress that they are using the data to find ways to improve retention, and not nudge people out.
Those caveats aside, data scientists who study retention say they have found some meaningful correlations.
VoloMetrix Inc., which examines HR data as well as anonymized employee email and calendar data, found that it could predict flight risk up to a year in advance for employees who were spending less time interacting with certain colleagues or attending events beyond required meetings. And Ultimate Software found a correlation between a client’s employees who waived their benefits coverage and those who left the company.
The big challenge for employers is what, exactly, to do with the information. Some aren’t sure how to approach employees at risk of leaving.
“Our goal is to never say the only reason we are coming to talk to you is because an algorithm told us to do so,” says John Callery, director of people analytics at AOL Inc.,which recently started working with workforce analytics firm Visier Inc. on a program to help predict attrition down to the individual employee. Mr. Callery says it is too early to tell whether AOL’s retention figures will improve, or by how much, since it takes at least a year to test a predictive model.
For the past three years, Credit Suisse has studied what happens to employees over time, including raises, promotions and life transitions, to predict whether they will choose to stay or leave the bank in the subsequent year. Changing jobs makes people “sticky,” or likely to stay on, says Mr. Wolf, who oversees the bank’s people analytics team. Yet as recently as five years ago, fewer than half of open jobs at the bank were posted, and most went to outsiders.
About a year and a half ago, the bank launched a global effort allowing its workers to raise their hands for internal moves. Credit Suisse recruiters now post 80% of open jobs, and cold-call employees when jobs open up.
‘Our goal is to never say the only reason we are coming to talk to you is because an algorithm told us to do so.’
After observing that some who volunteered to be considered for internal moves ended up leaving for jobs elsewhere, bank recruiters began using attrition probability estimates in deciding which employees to target when positions opened up.
Some 300 people have been promoted through the internal program; many of those people, Mr. Wolf says, might have left otherwise. “We believe we’ve saved a number of them from taking jobs at other banks.”
Credit Suisse also used analytics to investigate why women with certain job titles left the company at higher rates than men, and determined that transitions—whether a promotion or a personal milestone, such as a maternity leave—increased the probability of a woman deciding to leave.
To kindle closeness among team members, and help forestall attrition, Box has encouraged managers to throw more social events, recognize team-based work and hold more mentoring meetings between senior leaders and newer employees. Since workers were more likely to leave if they didn’t see clear career opportunities at the company, it has sought to improve at pointing out career possibilities for individual workers and encouraging “stretch” assignments.
Semiconductor maker Micron Technology Inc. is using data in its efforts to reduce turnover among first-year employees, who have about a 20% world-wide attrition rate, largely driven by manufacturing personnel.
Among its early findings, Micron discovered that workers were more likely to leave if they felt their job hadn’t been accurately described when they were hired, so the company is trying to create clearer job descriptions. Micron also found that people who relocated for a job were more likely to leave, but it isn’t sure why.
“It’s very delicate how you approach things,” says Timothy Long, the company’s director of workforce analytics and systems. “The idea is to determine, what can we do to get [people] to stay?”
Companies also are trying to predict when workers might leave their positions, but not necessarily the company. Wal-Mart is trying to determine in advance which employees are likely to get promoted so that it can line up replacements more quickly. The company says it promotes some 160,000 to 170,000 people a year.
“If we can tell three months in advance [that a position is going to be open], we can start hiring and training people. You don’t want the jobs vacant for that long a time,” saysElpida Ormanidou, Wal-Mart’s vice president of global people analytics.