Artificial intelligence (AI) is shaking up the recruiting process, changing the way recruiting agencies and sourcers discover and hire tech talent.
AI and machine learning enable professionals to quickly analyze huge amounts of data and make decisions and predictions based on that, said Summer Husband, senior director of data science at Randstad Sourceright, in a presentation at the 2017 SourceCon event. Now, recruiters and sourcers are putting AI to work to help define a job posting’s “perfect fit,” better surface strong candidates from search pools, and improve their ability to fill job openings fast.
Shortening up the hiring window
Unfilled job postings are a significant drain on organizational productivity. To ensure the right recruiting resources are being applied to filling a particular job opening, some sourcers are taking a tip from the healthcare industry’s use of survival analysis.
[ Learn how to win the war for top tech talent, and retain your best by knowing the 9 reasons good employees leave — and how you can prevent it. | Keep up on the latest CIO insights with our CIO Daily newsletter. ]
In healthcare, survival analysis is a machine learning technique that analyzes time to an event, such as a patient’s expected time before recurrence of a disease or a death. Here, Husband sees a good analogy to sourcing and recruiting
“We take data on jobs we’ve filled for clients in the past, how long those took, how many candidates, open roles, information about the company as well as job market data from sources like the BLS and CareerBuilder, for instance, to find out how all of those things impact the ‘survival rate’ of our open jobs,” Husband says. “We’re obviously flipping the script, because we want our open jobs to die quickly, but the process is the same.”
That process allows sourcers to set reasonable expectations for clients and allocate appropriate resources to harder-to-fill roles, she says.
“Our goal is that when we see a new [job requisition], we can gather these features right away and we can see whether or not this will be a tough one to fill, and then we can decide whether we should put extra resources toward that now instead of waiting and potentially failing to deliver candidates,” she says.
Accuracy is also important; being able to evaluate the precision (the fraction of received information that’s relevant) and recall (the fraction of relevant information that’s received) of algorithms can help sourcing and recruiting professionals make sure they’re delivering the right candidates, she says.
“Sometimes, overall accuracy isn’t the most important thing. In recruiting, it’s okay if you’re missing a few people who might be a good fit. We’d rather that than send a whole bunch of bad results — or inappropriate candidates,” she says.
Reverse-engineering the ‘perfect fit’
Some forward-thinking recruiters and hiring managers are using AI and machine learning to reverse-engineer candidate “fit,” and to predict a potential candidate’s performance in the role, says Chris Nicholson, CEO of artificial intelligence software company Skymind.
“The best use case would be solving the matching problem so that you’re leveraging the tech to find the best candidate for the employer and vice versa. The question everyone’s trying to answer through all the interviews, screenings, tech and coding challenges, is, ‘How can I predict someone’s performance?’ So, the smartest recruiters and hiring managers would start gathering résumés, performance reviews, work product, any information at all about highly successful people that already work for them and plug that into an algorithm to figure out what you are looking for,” Nicholson says.
Ideally, each organization would be able to create algorithms customized for the unique needs of their organizations, taking into account the varying definitions of success, Nicholson says.
“Obviously what makes someone a stellar employee at one company isn’t necessarily going to translate to every other company. So, it’s a bit like ‘Moneyball,’ where you’re looking for specific skills, traits, expertise that will fill in where you are lacking. The problem is, most companies are using the wrong ones, and that’s not only hurting them generally, but it also contributes to the lack of diversity and inclusion,” Nicholson says.
“Sure, maybe Stanford degrees, tenure at Google, white, male, been coding since childhood works for some companies, but not only are you excluding hundreds, maybe thousands of people with those parameters, everyone else is chasing those folks, too. They’re going to be expensive, they’re going to be unavailable, they’re not going to give you the results you need,” Nicholson says.
Balancing recruiting risk
AI and machine learning technology also can help determine how and when sourcers and recruiters need help with their workloads by looking at who has a disproportionate share of medium- to high-risk requisitions that might take extra time or resources to fill, Husband says.
“When we call something a high-risk req, we see that 85 percent of the time those miss their target time-to-fill. So, we can see who has a heavier load of these kinds of reqs and then make decisions about what to do. Do we need to shift these workloads around? Do we need extra sourcers and recruiters working on these?” she says.
Intuitive search for surfacing stronger candidates
Combing through troves of job candidates for just the right fit breeds another problem: What terminology should you use in searches to ensure you are surfacing candidates who strongly match what you are looking for, regardless of what terms they use to describe their experience and skills?
Here, artificial intelligence can help — what Glen Cathey, senior vice president of global digital strategy and innovation at Randstad, describes as “the moment you realize that all searches work,” and you’re staring at a bunch of candidate leads without knowing the probability of their being “the right fit.”
“As sourcers and recruiters, what problem are we trying to solve? We’re trying to find ‘the best people.’ That’s easy to say, but it doesn’t really translate into a traditional Boolean search! What does ‘the best people’ look like? What does ‘the right fit’ look like?” he says.
Cathey compares it to the Where’s Waldo? series of puzzle books; it’s not difficult to search anymore, what’s of greater importance now is a data problem. That’s where semantic search, conceptual search and implicit search come in, Cathey says.
Semantic search seeks to understand a searcher’s intent and the context in which a search was performed to improve the relevance of results. Conceptual search doesn’t require a precisely worded query, but just a few keywords around which to form a concept. Implicit search pushes information and results to you based on information already assumed or gathered, much like how Google automatically pushes restaurant recommendations in your local area, or pops up traffic advisories when you walk out the door to commute to work.
“You can type in a one-word search, and it’ll work. It’ll return results. But you’re looking for skills, experience, culture fit, soft skills, candidates who probably are within a specific geographic range, that are affordable in the compensation range you have, and who also are interested in the job. How are you going to sort through these results and find that person when so many look similar?” he says.
The language problem also rears its head when, at the word/phrase-level, organizations, candidates, sourcers and recruiters use different terminology to describe job titles, roles, responsibilities and goals, Cathey says.
“Say you’re searching for a ‘web developer.’ If you are using a standard keyword search, you’re getting results! The search works! You can fill jobs with that search. But you’re only returning candidates who use that exact verbiage to describe what they do. Would every person in that role use that language? Maybe not. How do you find great people who excel at that but use different language? How would they say it? What wording can I use to identify them?” Cathey says.
Peering into ‘dark matter’ results
Traditional methods for searching for job candidates bring about another unfortunate side effect: “dark matter” results, Cathey says. When searching, there are more candidates out there, you’re just not seeing them. This creates a problem that many sourcers and recruiters don’t want to own up to.
“If we’re really trying to find the best candidate, though, then you’re excluding people with those searches. Doing it this way means you’re actually looking only for the best of the easiest candidates to find. And that’s hard to admit, right? But that’s what is happening here,” he says.
But using AI and machine learning can help unearth candidates missed by traditional screening, sourcing and recruiting methods. Once these “dark matter” candidates are unearthed, recruiters can focus on the human element of the recruiting process and dig deeper; even if a candidate’s résumé doesn’t appear to be relevant, perhaps they have incredible soft skills, leadership experience or other valuable skills your organization needs, Cathey says.
AI and machine learning tech are evolving at a fantastic rate, and they’re a great tool to add to a sourcer’s or recruiter’s toolbox. By speeding up the process and ensuring more relevant results, sourcing and recruiting pros can focus on connecting with candidates, engaging with them and getting them hired, Cathey says.
“Especially with some of these search techniques, even if I run out of specific people I can directly identify who would be the right person for a job, I can find people with a high likelihood of knowing the right kind of person. And that’s where the human element of this profession comes in. Instead of focusing on unbillable research, you can use AI to automate these mundane tasks to allow you to focus on your clients. So, instead of focusing on finding people, you can focus on recruiting people,” he says.