In 2019, IBM noted that its artificial intelligence can predict with 95% accuracy which workers are about to quit their jobs. It is a remarkable achievement as cost of employee replacement is estimated to be between 30 and 400% of that person’s annual salary. IBM is assumed to use a lot of dependent variables to predict attrition; the variables include length of time between promotions, overall job tenure, promotion of peers, commute time and overtime. IBM’s CEO said this algorithm has saved IBM $300 mn in retention costs.
I strongly support IBM’s use of employee data to predict employee attrition because of three reasons: cost savings, talent management and ability to address employee turnover risks more effectively. First, when job markets are heated, employees can easily change jobs. New positions become available and hefty signing bonuses can be distributed. In those times, an ability to identify employees at flight risks directly contributes to financials, creating observable impact on the P&L. IBM is a clear example of that with $300 mn savings. Second, such an algorithm also helps talent management. No organization in the world would like to lose its best talent because of some issues that can be solved. By keeping the best talent in the company and advancing them career-wise would create an enormous value for the company. Third, if used well, this algorithm can inform decisions about promotion schedule or give insights on employee burnout.
I also believe that IBM’s or other companies’ algorithms for employee flight risks can be furthered by two additional items: predicting employee flight risk at the time of hiring, and addressing flight risks more effectively by relevant HR tools.
First, even though not inclusive, the variables mentioned in the first part are all the variables related to worktime. I think this algorithm can get better if it can use some other variables – e.g. personality and other companies’ potential interest in employees. Using other variables would also help company reshape its recruiting policies based on employee flight risk. Let’s assume a case IBM hires an extravert person to a role that has minimal people contact. As this person would probably lack job satisfaction, this situation would inherently increase turnover risk for this employee. Even though predicting flight risk at the time of hiring would not be 95% accurate, I still believe it can inform hiring decisions.
Second, this algorithm is built for addressing flight risks. However, if company lacks appropriate tools to address these risks, the algorithm itself would be useless. If a top performing talent is only interested in increasing his or her salary and the company would not be able to provide him or her desired salary levels because of HR policies, then the algorithm would not be useful. In another case, in which a person would desire less travel, but there are no rotation opportunities for this person, then the algorithm would again be less useful. The algorithm would only help company target people with more risk, but HR policies should be adjusted to address the problems that cause people to leave in the first place.
All in all, I believe such uses of people analytics can be beneficial for the employers and employees. Use of people analytics in employee retention can be further advanced by adding more variables, tying it to hiring decisions and having relevant tools to address the problems.