Tech

Employee turnover is expensive—an algorithm could help (and also hurt)

New software cobbles together billions of data points to indicate the likelihood of someone leaving.
article cover

Gizela Glavas/Getty Images

· 6 min read

To predict when an employee might quit their job, you don’t need to don the garb of an ancient soothsayer and stare intently into a crystal ball. You might, however, try consulting predictive attrition AI tools, which seemingly provide a window into the future. Analyze a variety of data points (from public salary info and turnover rates, to external factors like economic trends and industry shifts), then behold: You suddenly know which of your employees has a higher likelihood of quitting and you can intervene before it’s too late. But like all good algorithms, there could be a dark side.

Compiling troves of data in the service of predicting quitters is nothing new. IBM’s Watson supercomputer, released in 2010, can forecast attrition rates with 95% accuracy, CEO Ginni Rometty claimed in 2019. A glut of companies, such as Retrain AI and Eightfold AI, also harness big data to better forecast attrition, which has become more enticing to HR departments in certain industries amid layoffs and budgetary shortcomings: In a recent Capterra survey of 300 HR leaders, 98% said they’d use algorithms to help reduce labor costs during the next recession.

Predictive attrition software—which falls under the larger umbrella of people analytics—can be used to engage, motivate, and even promote workers, explained Andrew Spott, president and co-founder of HR Signal, a vendor of retention software. “We’re a people-centric company,” he told HR Brew. “Our job is to try and help employees and people grow better in their careers, and it’s typically easier to grow where you are than leav[e] an employer.” But the fear among advocates like Wilneida Negrón, director of policy research at the advocacy organization Coworker, is that many of the algorithms on the market could be used for punitive measures, like laying off workers deemed a flight risk. “In theory, you won’t eliminate [the possibility of] a malicious manager” who might use the data to make the decision to punish a worker, she said.

At its core, the proliferation of various and disparate AI tools mean HR professionals will have to decide how to interpret data and how those interpretations ultimately affect workers, Negrón said. When it comes to predictive attrition software, there needs to be “goodwill [between] both the vendor that’s creating this to help worker morale, and then goodwill on the side of a business, that they’re trying to use this to improve the workplace.”

Different branding, similar uses. Vendors can pitch the tech in various ways, explained Negrón. She’s encountered companies trumpeted as providers of “workforce behavior analytics products, investigation management software, or risk-management software,” but from a practical standpoint, the common denominator in developing the tech is prodigious amounts of data.

HR Signal bills itself as “proactive retention software,” with the intention of facilitating career advancement for workers. The company purchases data from wholesale vendors, collected from billions of publicly available resources, including government reports, public résumé listings on websites like LinkedIn, and market data concerning salaries and employee tenure. According to Spott, the algorithm is fed data concerning criteria such as educational histories, various positions held, start and end dates, degrees and universities. All data is anonymized, he said, but when a new employer is onboarded, they compare their employee roster with HR Signals’s existing data set, and it spits out a retention risk score. Employers do not include information on employee “gender, age, ethnicity, [or] marital status,” he wrote in a follow-up email.

Quick-to-read HR news & insights

From recruiting and retention to company culture and the latest in HR tech, HR Brew delivers up-to-date industry news and tips to help HR pros stay nimble in today’s fast-changing business environment.

Once employee information is inputted, “each employee has a bespoke analytics page that pulls in a huge amount of position-based benchmark data, position-based average compensation date, position-based market demand data,” he said. The software assigns a retention risk score score between 1 and 99. The higher the number, and the greater likelihood of an employee seeking new opportunities within 90 days, according to HR Signal.

If an employee’s score crosses a certain threshold, the software issues a prompt for HR to conduct a stay interview. HR Signal’s aim is “to focus proactive retention efforts, [getting] the right people to have the right conversation at the right time to reveal people who maybe aren’t happy…and be able to act on it before someone leaves,” Spott said.

A warning for IRL humans. The data should spur human intervention, sources stressed. “Data is great, but a human knows how to…sit down and have the conversation,” Alexandria Brown, founder of the HR consultancy The HR Hacker, explained. “You still need a human being who knows the career arc of that person…so you can sit down and have an honest and transparent conversation about what their future is going to look like.”

Still the possibility remains that an organization could wield the data more like a cudgel than a neutral reminder to initiate a conversation about career paths. “In our service agreement with customers, they agree to not use this for bias or prejudice or inequity. Ultimately, we can’t control that. But we take it seriously,” Spott said.

To use the data productively, Negrón advised collaboration between vendors and companies, to instill a sense of accountability. At best, vendors can “require participation of people that buy their products to go through a series of ongoing education on the potential harms,” she said.

Moreover, it’s about understanding that workers are at the crux of all considerations when using the technology. “It’s not just the vendor, the product, and the employer. There’s actually a worker in the middle of this,” she said.—SB

Correction 03/01/2022: The piece previously stated HR Signal scrapes data from various resources online, including LinkedIn. This is incorrect. The company purchases its data from wholesale vendors.

Do you work in HR or have information about your HR department we should know? Email [email protected]. For completely confidential conversations, ask Sam for his number on Signal.

Quick-to-read HR news & insights

From recruiting and retention to company culture and the latest in HR tech, HR Brew delivers up-to-date industry news and tips to help HR pros stay nimble in today’s fast-changing business environment.