Flight risk at IBM :

A similar analysis was done at IBM, where turnover was high for certain business-critical roles. Using IBM’s Watson machine learning capabilities, the workforce analytics team build an algorithm that included sources like recruitment data, tenure, promotion history, performance, role, salary, location, job role, and more.

The company also included employee sentiment, measured through their Social Pulse. The hypothesis here was that engagement with social media might fall when employees are thinking about leaving.

The investment yielded $ 300,000,000 over four years and turnover for critical roles has fallen by 25%. According to the report, productivity has also improved while recruitment cost have fallen. IBM CEO Ginni Rometty recently made a splash when she said that the technology pioneer can now predict with 95 percent accuracy which employees are likely to leave their jobs within six months.

She said that IBM’s predictive attrition tool, developed with its Watson artificial intelligence (AI) technology, analyzes thousands of pieces of data to predict employee flight risk and prescribe actions for managers to take to address the underlying issues. The new tech is one of the more high-profile examples of the way traditionally low-tech HR has been investing in data science to assist its decision-making.

Diane Gherson, IBM’s chief human resources officer, said that the company had previously tested hypotheses about who might leave but that “the value you get from AI is it doesn’t rely on hypotheses being developed in advance—it actually finds the patterns.”

“Predictive analytics can be sensitive to things that management may not easily apprehend just by talking with employees or looking at their employee records,” said Jason McPherson, chief scientist at employee feedback platform Culture Amp in Melbourne, Australia. “We know that waiting for people to resign and chasing them out the door with a better offer doesn’t work, at least not for long. Imagine the money you could save, and the goodwill, productivity and engagement you could retain, if you could see into the future and understand which employees are going to leave and why. Knowing what’s driving turnover arms HR professionals and people leaders with the keys to turning turnover around,” he said.

[SHRM members-only toolkit: Managing for Employee Retention]

How to Build a Flight Risk Model

A flight risk model takes about a month to develop, according to Mondore, and should result in HR understanding:

• The attitudes that drive turnover.
• What leaders do that can lead to high turnover, and what they do that can lead to low turnover.
• Specific investments needed to drive retention.

The quality of the predictions HR can make ultimately depends on the quality of the data it feeds into the algorithms and machine learning models, explained Toby Roger, lead product manager at Culture Amp.

“Since our founding in 2011, we’ve collected insights from over 2.5 million employees globally,” he said. The data includes behavioral indicators gleaned from employee engagement surveys, onboarding and new-hire feedback, and soon performance management reviews.

Mondore recommended including demographic data like age, gender, marital status, education and tenure; performance metrics and quarterly or annual reviews; engagement survey data; workload; paid-time-off usage; absenteeism; and salary and career growth. Compensation data should be worked into the model to be able to explain to leadership the amount of money at risk if turnover isn’t prevented. He added that the problem with collecting information from exit surveys is that it’s too far downstream—coming from people too late to save.

Some companies are alerting managers about individuals at risk for attrition, Mondore said, but that’s a bad idea, experts agreed. He suggested reporting aggregated data up to the manager or department level.

“We aggregate the data, typically from groups of 15 people or more,” Roger said. “Confidentiality is the primary reason for that. It’s important for trust and ensuring that we are collecting quality data. When someone feels the information they provide will be disclosed to management, they tend to not respond or give pat responses. And as soon as a manager sees a little red flag on someone’s record, they can’t get it out of their mind.”

Mondore said unwanted manager reactions could be showing positive or negative bias toward the person, being resentful toward him or her, trying to fire the person, reducing professional development, or going out of the way to induce the person to stay with increased compensation and perks.

Experts agreed that it’s also not helpful to analyze or monitor employees’ e-mail or external social media accounts.

“Predictive attrition technology works better when the aim is to improve the workplace rather than target individuals,” Roger said.

Tips for Success

Flight risk models are a wasted exercise when nothing is done to address the findings, Mondore said. “The data is interesting at a high level, but you must get it in the hands of your frontline leaders and make it actionable. Go to the areas with turnover risk and start setting up stay interviews, talking to your managers and training those managers on retention practices.”

Roger added that Culture Amp provides a list of drivers for each group at risk for turnover. “We also provide actions that the organization can take to address or improve those areas. Just telling someone they have a problem is not particularly helpful.”

Gherson said that IBM’s program urges managers to intervene with high-potential workers or those with hard-to-find skills but not necessarily with workers across the organization.

“The ones who are in high demand today and high demand tomorrow are going to be the ones we treat with a very high-touch” response, she said.

Happy Reading