This blog post is the second in a series of two. Read the first blog post here.
Automation – Programmatic Matching
Using our personality extraction algorithm, we can gather the given personality of a candidate and the required personality of a role. Ideally we would do this on very generic descriptions of the candidate and role, where the candidate would just describe him/her self; what motivates them in life, what are their hobbies, what is their passion. This way, we have one generic description that we can use for any role and it is as pure as we can get it.
In reality, because staffing agencies are new to this way of working, such generic descriptions are not available. Of course we can start collecting them but we can already start using what is available at hand. Using a CV’s description of oneself, a motivation letter or another profile description, we extract the candidate’s personality. We do the same for the vacancy as well as for the hiring company if we deem the latter necessary. We then overlay all the personality profiles and produce a fitness score.
The process described above takes no more than a couple of seconds. Scaling is essence here. We can now daily produce a list of candidates and roles that they best fit and set our recruiters out to better help the candidates and ultimately get more satisfied customers and placed candidates.
We call this Programmatic Matching.
One more step
Once we takcled the crucial step of matching on personality rather than on CV, or in essence, on randomly mentioned hard skills, we quickly got the feedback that while this is an improvement, we cannot just neglect the candidate’s CV and the vacancy’s required skills. We needed CV parsing too.
We were warned at first that parsing a CV, while seeminly simple, is not an easy task. At least not if you want to do it right. There are a handful of open source implementations but none of them really are of commercial quality. So we set out to it ourselves. We quickly ran into some challenges that made us believe the warnings were fairly given but we embraced the challenge and eventually tackled all of them.
We ended up having a CV parsing algorithm that performed exceptionally well and we used it as input for our matching algorithm that already used personality to now also consider skills mentioned in the CV and those mentioned in the vacancy.
The proof is in the pudding
While we already had a solid business running with matching based on personality, we wanted to experiment with how well our matching actually performed, using both personality and skills. Some of partners in the staffing & recruitment industry were kind enough to help us run some experiments to figure out just how well we did.
We ran multiple experiments, small to large in scale, with indicators ranging from satisfaction to duration of a placement, to figure out how a match (or oppositely, a non-match) based on our algorithm would perform.
We found that actual outcomes slightly depend on the type of role (ie. manager, call center employee, tech expert) and type of indicator we are looking for (ie. duration of placement, employee satisfaction, employer satisfaction), but that all experiments resulted in a significant increase when using our algorithm over the status quo (which is often a manual process of consultants calling and placing candidates on roles every day). Improvements over this baseline go up as far as 50%.
While this a great business case, what we value even more is that we can help candidates become happier at their jobs. Our goal is to turn the staffing & recruitment industry around by moving from a paradigm where roles are being filled with candidates to candidates finding their ideal roles.
Thank you for reading this lengthy blog post! Contact us if you want to know some more.