Should I Stay or Should I Go? Can AI Predict That?

In the years leading up to 2020, tech companies touted the ability of artificial intelligence and machine learning algorithms to help employers identify which workers were more likely to quit their jobs. 

In 2019, IBM officials reported that their "predictive attrition program," developed with the Watson supercomputer and machine learning tools, could predict with a 95 percent accuracy rate which workers would quit. That was then.

Today, the COVID-19 pandemic has changed the reasons why workers want to stay at or leave their jobs. The appeal of remote and hybrid work has grown, and work/life balance, mental health and wellness are some of the factors that have become more important to workers as they seek greater autonomy over their work lives. Additionally, employee quit rates have hit historical highs.  

Data from the U.S. Bureau of Labor Statistics shows that in September, 4.4 million workers quit their jobs, mainly in the food service, accommodations and retail industries. The September quit rate is the highest on record. Approximately 4 million workers quit in July and August as well.

According to Mike Brennan, chief service officer and co-founder of Leapgen, a Minneapolis-based HR consulting firm, the degree of accuracy in predicting whether a worker will stay at or leave their job varies across industries.

"What's the new variable? In a word, COVID. Fear of the disease has prompted many to quit, and many of those who were already out of work will not return to similar jobs that expose them to the same risks," Brennan said.

He added that the relative strength of variables used in datasets to develop patterns and trends of worker engagement and satisfaction has shifted.

"The most obvious shift has been the influence of location," he said. "This is forcing the hands of employers. Those who have been able to move more aggressively toward work-from-anywhere models—be they completely remote or hybrid—have a leg up in the war for talent."

At the Seattle-based Allen Institute for AI, Tim Mulligan, the organization's chief human resource officer, said while his organization is using AI in its recruiting, onboarding and interview process, leaders aren't planning to use AI to find out which employees want to quit.

"We employ more-traditional means for that with pulse surveys," he said. "Right now, we are conducting an annual diversity, equity and inclusion-focused sentiment and culture survey."

He added that AI experts who are working on people analytics tools are going to be challenged to predict why and when a worker will leave during the pandemic. Mulligan said listening to what Allen Institute workers wanted was enough to change policies and opt for a mix of remote and hybrid work arrangements.

"Employees flat-out said that if they have to come back to Seattle to work in the office, they're leaving," Mulligan said.  

At Visier, which provides people analytics and workplace planning software, executives are betting that their machine learning platform will provide companies with employee insights that will help managers apply appropriate interventions to keep workers.

Visier's model is trained on the real behavior of employees, learned over a three-month window. That data is compared with historical data as changing patterns are identified.

"As the pattern of behavior changes, as long as you retrain your model, that model picks up the changes," said Ian Cook, vice president of people analytics at Visier. "Our machine learning model will have a blend of what was and what is but will adjust to the new world as fast as possible."

Headquartered in Vancouver, British Columbia, Canada, Visier uses machine learning algorithms to answer questions about employee starts and exits, demographics, and diversity trends.

"We are talking about hundreds of variables, which range from simple elements like tenure, location, job name or level in the organization," Cook said. "There are more-complex factors such as changes in pay, the number of people on their team, the number of people who report to them, the length of time they have been in the specific role they hold and the number of times their manager has changed in the last 12 months."

According to Cook, success using people analytics tools isn't measured by attaching a percentage rate to the number of people identified as being at risk for leaving the company. Instead, success occurs when the technology identifies workers who are likely to leave and action is taken that keeps the employee, which in turn saves the company the money it spent developing that worker's skills and experience.

Visier executives also note that while the technology can gather critical data that suggests an employee will leave, only the employer's decisions can make that employee stay.

"What employers do with the data is always the differentiator," Cook said. "It's a human process. Technology is never going to take the place of a conversation that is required to engage somebody changing their minds and staying. I don't know a single piece of technology that would do that."

Nicole Lewis is a freelance journalist based in Miami.