APPLYING for a job can be a soul-destroying process. First you have to polish your CV and make sure your covering letter matches the job description. Then you send them off into the void, hoping they will catch the eye of a recruiter who will call you in for an interview.
But no human may ever read your résumé. For the past few years, the world’s biggest firms have been using AI recruitment software to filter job applications and streamline the process, and smaller businesses are increasingly interested, too.
Advertisement
That should make selection fairer and less open to abuse, in theory. But it can also make it harder to get a foot in the door (see “Found wanting – by a machine“). Now a number of companies want to change the way AI helps find the right person for the job.
Existing applicant tracking systems (ATS) typically scan applications for keywords that the employer has selected. The software might prioritise sections of the CV in which those words appear alongside more recent positions, for example. Any CVs that don’t fit the bill are instantly rejected; those that pass are stored and indexed for a human recruiter to look through.
Career coach Pamela Skillings believes applicants should assume that their application will be screened by an algorithm, and quite probably rejected, before a human ever looks at it. But such systems are hardly foolproof. They don’t just frustrate people looking for jobs with non-traditional CVs – they are also far from ideal for employers. For example, excessively rigorous screening can mean relevant sections of a CV will be passed over if they contain words and phrases only slightly different from the employer’s preselected ones, and the candidate rejected.
“Jobseekers should assume their application will be screened by algorithm and quite probably rejected”
“Sometimes there’s a little bit of conflict between HR and the actual managers because they’re saying, ‘What are these résumés I’m getting?’ says Skillings. “Or, ‘Someone I know applied and they didn’t get through – there’s something broken here between what you’re screening for and what we actually want.'”
Moves are afoot to make the recruitment algorithms more sophisticated. The biggest name in online recruitment is Silicon Valley giant LinkedIn, and a few weeks ago it unveiled its revamped Recruiter app at a glitzy event in Los Angeles.
On stage, Eduardo Vivas, head of LinkedIn Talent Solutions, showed how typing the names of two software developers he manages into a search box would cause the system to sift millions of LinkedIn users for matching talents. Then it displayed a list of candidates at other companies with like-for-like skills and experience.
“Instantly, I see people on the platform that look just like them,” he said. “The ability to take our best people and find people just like them is incredibly powerful.” Another of the firm’s algorithmic innovations is that employers will be able to narrow down huge lists of existing contacts to find those with the skills appropriate for a recently advertised job.
A button for itchy feet
The app will also indicate how often a prospective candidate has interacted with the company’s posts on LinkedIn and show whether he or she is actually looking for a new job right now. This bit of knowledge will be picked up thanks to the “something new” button coming to LinkedIn profiles. By clicking on it, users on the social network can tell employers – except, of course, their own – that they have itchy feet.
Other companies are going even further to cherry-pick the best candidates. Connectifier, founded by ex-Google staff, markets itself as a search engine for recruiters. The firm’s algorithms crawl a range of websites and candidates’ profiles, as well as CVs, to build up a picture of skills and expertise.
“We can look at things in finer detail – things that a recruiter might have a hard time taking into account. Like, does someone have a lot of friends in a certain location or at a certain company?” says CEO and co-founder John Jersin.
The system can also, for instance, harvest data from a site like GitHub, where programmers share and discuss code.
“If they’re posting a significant amount of code online, we know that they know the language they’re writing in,” says Jersin. “If they’re answering a lot of questions about a specific language or technology, we have a lot of information about their knowledge on those skills.”
It’s not the only such technology out there. is a small Danish start-up touting “a machine learning engine for your recruitment”. The software has been trained on professional vocabulary relating to job descriptions and is able to analyse databases of CVs, looking for candidates who might fit a post.
The firm’s algorithm uses statistical models that look at the distribution of words. It understands that “software engineer” or “software developer”are very similar roles, for example.
Six Danish firms have been testing the system on open job positions. “Because we had a lot of résumés in the database we could actually detect patterns,” says Jonas Krarup, one of Reveal’s founders.
Krarup and his co-founder Joel Raucq say their approach could help companies with large numbers of CVs on file to make the most of that data and identify good candidates as soon as new roles come up. The system can even predict how interested a candidate might be in a job change from their current position. This is done by assessing how many previous jobs candidates have listed on their CV and noting how frequently they have moved from role to role.
Of course, there are those who resist the encroachment of AI on recruitment. Unsurprisingly, headhunter Nick Corcodilos is not a fan.
“The reason companies need technology to sort through so many résumés is because they mindlessly solicit so many,” he says. Corcodilos thinks that firms should move away from online recruiting, and he worries that “potentially outstanding” candidates are being unfairly ruled out by some systems.
For Skillings, AI has great potential in recruitment but, she adds, it will work only if recruiters and algorithm designers are willing to take the time to carefully define what they really want in a candidate.
“AI selection will only work if recruiters and algorithm designers carefully define what they really want”
“The end result could be that algorithms could actually identify people who would be great at the job,” she says, “but might not have ever been given a second look under the traditional keyword scan.”
(Image: Qilai Shen/Panos Pictures)
Found wanting – by a machine
For 10 months, Alyssa Mathews has been looking for her first full-time job. She has sent off well over 100 applications, so far to no avail. It’s not that she’s underqualified – she has a master’s degree in environmental science. Instead, she thinks she is falling foul of CV-screening algorithms.
“I’ve got rejection upon submission, where I submit my application and it immediately rejects me,” she says. “I know that’s just a computer, and I failed some sort of algorithm or test.” She believes that because her CV is highly interdisciplinary, the computer perceives it as lacking specialisation.
Since discovering more about how the algorithms work, Mathews has changed tactics and begun approaching recruiters at job fairs. She is now getting many more interviews.
Some find the problem insurmountable, though. One 48-year-old graphic designer who wished to remain anonymous says she has sent out several hundred applications in recent years. A large number of rejections, she suspects, are to do with her age and gaps in her CV.
“I’ve met a lot of people in my exact situation,” she says. “Over 40, unemployed for over three years and they know that it’s over.”
This article appeared in print under the headline “The AI headhunters”
