Will job matching ever take off?
Job matching is the AI of the job board world.
It’s like this: for decades AI has been held out to computer geeks and normal folks alike as something that was ‘just around the corner’. Even the definition of AI has changed – sure, Big Blue could beat a chess master but that wasn’t really intelligence, right? Since 1950, the Turing test has been the gold standard for AI – and no computer program has been able to pass it. (I will leave for now the various strengths and weaknesses of this test).
Job matching has held a similar spot in the job board industry. As far back as I can remember, there have been vendors promising solutions – but the solutions have always fallen short in one way or another.
The promise: A job seeker can fill out a form (often of significant length) that details skills, likes, dislikes, and job history. The job matching software will use this information to ‘match’ the job seeker to a ‘perfect’ (or pretty darned good) job. The job seeker is happy because he/she has found a great position, the employer is happy because they’ve found the perfect employee, and the job board is happy because a position has been filled.
The problem: People are impatient, even lazy. They don’t want to spend 10, 20, or even 30 minutes filling out forms for an uncertain return. On the other side, programmers and designers must decide which factors are most important in matching a job to a person. Those factors can be overwhelming – for example, is it enough to have a B.A.? Or is it better to have a B.A. from Amherst? Or is it better to have a B.A. in English from Amherst in the late 80s? With a 3.5 GPA or higher? And believe me, I’m simplifying the problem here. Think about cultural fit, hard and soft skills, personality traits, and much much more.
Who has tried it? Well, there’s Jobfox, RealMatch, and many others. None dominates the job board industry, although each has made inroads. There are also stand-alone services such as ClearFit and QuietAgent that are unaffiliated with specific job boards. I suspect that all face the problems described above: job seeker resistance and the complexity of matching.
What’s next? Heck if I know. Certainly if a system could gather information from job seekers in a painless manner, and then use that information to match jobs to seekers, it would have a chance of succeeding.
One path forward could be the collection and collation of data left on various social media platforms (imagine analyzing ALL Facebook postings for everyone ages 18 to 26, for example), which could then be fashioned into candidate profiles. Talk about a challenging project! But it’s just data, after all (although there is a small issue of privacy there).
Or perhaps a perpetual resume site like LinkedIn could somehow extract enough information from its users’ profiles, comments, and actions to build candidate data that would be used in matching systems.
Sounds like a job for AI!
[Want to get Job Board Doctor posts via email? Subscribe here.]. [Check out the JobBoardGeek podcast archive!]This Post Has 6 Comments
Comments are closed.
[…] This post was mentioned on Twitter by Ben Baldwin and The ClearFit Team, Jeff Dickey-Chasins. Jeff Dickey-Chasins said: New Job Board Doctor post: Will job matching ever take off? – http://bit.ly/bnJKHn […]
Agreed.
My only addition would be that time plays a big part in these algorithms. Take GMail and their new priority filter. If you’ve been using their service for a while they know your email habits very well. Who you always read emails from, what you put off, and what you trash. The major I problem I see with job matching services is that the majority of their users are active job seekers who want great (perfectly matched) jobs NOW. While they can go thru and help the site build better custom results they don’t have the time to – or would rather spend the time hunting down and applying for jobs. After a few too many duds they’re likely to go elsewhere.
Another note – while pulling facebook data for the younger applicants could be the way to help build the data backlog there is one glaring problem. Just because someone majors in X doesn’t mean they want to work in that field. Same thing for location (attend UCLA but want to work in NYC).
The focus should be to build great, simple, search tools and get high quality job postings. Then, let the applicants do their thing.
Thanks, Jeff, for your input on matching job boards. After filling out and getting few matches, I have found that most of them have one big problem: past experiences don’t always match when it should. For instance, what skills a candidate currently uses in, say, pharmaceutical sales wouldn’t necessarily match if the job wanted experience in medical equipment sales. But what if the seeker does have medical equipment sales experience from past employment which would make them a “perfect” match? On the flip side, the matching may be so broad that candidates will think they have a lot of perfect matches when, in fact, they don’t. Then there is the candidate who will broaden their experience by tweeking their skills until it’s a better match.
In one instance I got an email saying that I was a perfect match but when I went to apply for the job I had to go through another company website only to find out that the job wasn’t open anymore. I’m picky about who sees my resume and now I feel that the job board is selling me out. How do you explain that?
Great analysis Jeff. Candidates describe themselves as accurately as … job matching services describe themselves.
From what i hear (not in the market myself), dating sites suffer the same fate.
Hey,
What you mentioned about data collection sounds a bit like what we’re trying to do at Uvisor.
Some of the other big players in either “AI” or Expert System based matching (rule based, not actually learning systems) are Burning Glass, Daxtra and Actonomy. These are mostly engines that work behind the scenes offering matching services, either directly to candidates through jobsboard sites, or to direct employers or agencies by matching candidates to job posts. An understanding of context and a semantic understanding of text terms and how terms relate to each other are at the root of a lot of these systems. Others rely on a complex set of rules and taxonomies (relations, groupings and hierarchies between keyword terms) to do their matching.
Others such as LinkedIn have created their own matching engines, and in my experience so far seem to have done a decent job of it (some recent job posts targeted at me on the site weren’t just job skill and job title focused, but also sector/context focused in that they were roles within the recruitment technology sector).
There are a vast range of matching engines of varying complexity out there – as you’ve put well in your blog post, the main problems come down to the complexity of the matching logic and the data entered into the matching engine (both in the job post AND candidate profile). Poorly written and incorrectly assigned job posts are just as much to blame for many mis-matches as lack of candidate data put into the system.