Many people assume that when they apply for a job online their application won’t be viewed by a human.
Turns out they are probably right, at least if they are applying to Unilever. And that could be a good thing.
The global consumer-goods company got a lot of attention earlier this summer for rolling out an algorithm-based recruiting strategy on college campuses to prescreen candidates and gather tangible evidence about whether they are right for a job before any interviews take place.
“Our prior college recruitment process focused heavily on recruiting at specific schools and reached a narrow percentage of the national student population,” said Mike Clementi, Unilever’s vice president of human resources for North America. “We felt there was a way to cast a wider net and bring in more diversity of thought and talent.”
Students interested in an internship are invited via job boards and social media ads to submit an online application, which culls data from their LinkedIn profile to reduce the amount of time to fill out the application form. If they meet basic eligibility requirements, they are invited to play 12 short online games designed to measure different cognitive, emotional and social traits, and evaluate the candidate’s level of fit for the company overall as well as specific roles. They are then asked to complete a digital interview recording themselves solving real-world problems using Unilever scenarios.
“It’s not about finding one single precise profile but how students’ overall characteristics match Unilever’s future talent needs,” Clementi said. Eventually the list is narrowed down to top candidates, who secure an interview with an actual human, and participate in a Discovery Center Day, where they experience a day in the life of a Unilever employee. “We are disrupting the traditional recruitment process for interns,” Clementi said. “The end result is a powerful talent landscape that will support the future of our business.”
A Wider Net
Using algorithms to scan candidates solves a lot of problems, said Kevin Parker, CEO of Hirevue, a video interview software company in Salt Lake City that helped Unilever develop the recruiting program. “Unilever’s challenge is scale,” he said. “They have a lot of openings and a lot of candidates, and in a lot of cases the right candidate applies for the wrong job.”
Those candidates might get cut from a traditional screening process, but the algorithms are designed to look beyond traditional résumé information for core attributes. Parker argues this approach benefits both candidates and companies, because it speeds the recruiting process, opens recruiting to a much broader pool and ensures candidates with great potential (rather than just great pedigrees) get offers.
“We are already starting to see the returns from this process, and they are very much in line with what we’d hoped for,” Clementi said. “We’ve been able to cast a wider net and remove unconscious bias from the process and are finding that our candidates have more diversity of thought and are coming from a wider range of schools and geographic locations.”
Unilever isn’t alone in using machine learning to screen candidates, though curious recruiters should know that this is not a simple solution to implement. Algorithms aren’t a pre-packed piece of software that can instantly find the best candidates for any job. They use specific data about what makes a high performer in the company to search for those traits in the candidate pool. Those qualities may include specific academic or job history, or they can be softer measures, such as a high level of empathy, curiosity or determination. “The algorithm can find those traits which helps narrow the search,” Parker said.
It also helps companies take a more holistic view of candidates, which isn’t always easy for human recruiters to do, said Nathan Kuncel, professor of industrial-organizational psychology and a McKnight Presidential Fellow at the University of Minnesota. “People tend to get overly excited about an individual piece of information, and that can make them myopic,” he said. For example, if a candidate shares the recruiter’s alma mater, that recruiter may seek out data that confirms they are a good choice while ignoring anything to the contrary. “Algorithms push back on the tendency to misweight information,” Kuncel said.
Conversely, if a company is seeking a more diverse workforce but their current team is predominantly male, the algorithm can tease out the gender-neutral traits that make some people more successful and seek them out in the entire pool of candidates. “The beauty of algorithms is that they are a testable way to bypass failures in human judgment,” Kuncel said.
But this all assumes you have this information in the first place. To define what makes a high performer, companies need to compile years of past performance data, and correlate it with high-potential résumés, interviews with managers and workers, and detailed profiles that describe the skills, experiences, and personality traits of the best employees. That data can come from a lot of different sources, if it exists at all, and compiling it in a way that makes it possible to uncover desired performance traits isn’t easy. The biggest challenge companies face is turning qualifying information into numbers that the algorithm can work with, Kuncel said.
It’s a time-intensive thing to do but worth the effort, he said. “Even the most basic algorithms trounce human judgment every time.”
Sarah Fister Gale is a writer in Chicago. Comment below or email firstname.lastname@example.org.