Algorithmic Hiring: Why Hire By Numbers?
By: John Rossheim
Hiring algorithms have been in the headlines — and for good reason. Study after study, including this recent whitepaper published by the National Bureau of Economic Researchers, suggest that number-crunching can produce higher-quality hires than recruiters and hiring managers.
What’s not to like about the promise of boosting objectivity in candidate assessment, reduced bias and the ability to identify applicants that are most likely to become productive employees with long tenure?
Before employers, recruiters or staffing firms jump on the analytics band wagon, they need to understand what goes into data-driven talent acquisition. So let’s take a look at what goes into hiring algorithms and how data analytics might help organizations set aside human foibles to hire the best candidate.
What is an algorithm, anyway? For our purposes, an algorithm is a set of operations – from calculating simple averages to performing complex statistical analysis. When applied to a large quantity of data, it generates information that decision makers can act on.
In evaluating candidates, data types can include resumes, publicly available information, as well as responses to candidate assessments that delve into personality, temperament, aptitude for skills such as problem solving and more. The rules and operations of this calculation might be as simple as a spreadsheet that consolidates multiple ratings of a candidate’s potential or as complex as state-of-the-art predictive analytics or for that matter, machine learning.
Can software and silicon offer more balanced evaluations than gray matter? Proponents of hiring algorithms — and their legion is growing — want to put human foibles in their place when it comes to finding employees.
“We haven’t concluded that human judgments have no value,” says Nathan Kuncel, professor of psychology at the University of Minnesota. “It’s just that these judgments come with a package that includes bias. People can get hung up on one piece of information and make too much of it.”
Using big data to evaluate individual candidates. More complex hiring algorithms use data science to correlate the performance of large numbers of employees with data gathered on candidates.
“We’ve collected 2.5 million assessments of professionals, studied human performance factors, looked at what profile works best for each given role,” says Mike Distefano, a senior vice president at recruitment firm Korn Ferry.
Consider the data sources. Hiring algorithms may work within a narrow scope of information –such as the results of a single assessment — or with a wide range of data.
“We’re collecting information from three kinds of sources: publicly available information, background info supplied by candidates such as a resume and interaction data,” which can include metrics of keystrokes, says Mike Rosenbaum, CEO of Pegged Software, a firm that specializes in hiring algorithms for the healthcare industry.
Odd as it may sound, in some cases the number of keystrokes or words that a candidate enters in response to an assessment question can sometimes be a better correlation with future job performance than the actual content of that response.
Correlating the candidate’s data with employee performance. “We compare each person’s data set with outcomes like first-year retention,” says Rosenbaum. “We’ve seen lower turnover for every client, with a median reduction of 38 percent.”
Pegged’s software also weighs in with predictions of whether each applicant is likely to drive up or down institutional quality metrics, such as patient satisfaction and medical errors. Some hiring algorithms also correlate candidate data with employment outcomes such as productivity and performance ratings.
Some employers put faith in black-box hiring algorithms. Let’s face it: Data science is complex. Thus many corporate decision makers won’t understand the inner workings of the software that will have major input into who gets hired.
“What appealed to me is that Pegged aggregates our own data with other data in their database,” says Carlyle Walton, CEO of Metroplex Health System in Killeen, Texas. “I love the concept, even though I don’t understand all the math.”
Scientific, maybe. Irreproachable, no. “The Pegged Software solution employs a strictly scientific approach,” says the company’s web site. The software uses mathematical equations to enable what Pegged calls evidence-based hiring.
But not all algorithms automatically eliminate the human tendency for individuals to favor people who are like themselves. And a lot of expertise may be required to evaluate the potential of an algorithm to avoid bias in the context of a given data set.
Algorithms can’t always ensure diversity. Even when an organization has won broad buy-in for diverse hiring, there are many recruitment factors that algorithms alone cannot compensate for — such as a homogeneous pool of applicants.
Google, for example, has very publicly aired its goal of changing the demographic makeup of its workforce, which is overwhelmingly male and white or Asian. Yet the 2015 edition of Google’s diversity report, which shows that only 2 percent of its staff are black, 3 percent Hispanic, and 30 percent are women, points to the challenges of changing workforce demographics.
Part two of this story: Recruiting Algorithms: Appraising their Limits and Benefits