Is Recruitment Truly Broken?
Or we just plain lazy?
I came across the following video on Recruitment.
It describes the current macro-environment on skills. Exacerbated by the shrinking population, skills that are highly desired are now in a smaller talent pool. It’s a race to try and identify these talents quickly from large piles of CVs that are submitted.
Moreover, recruitment seems to be, for now, an AI vs AI. HR is using AI to write job descriptions that better describe the job role, followed by using AI to determine and sieve through the hundreds of CVs they are getting. Talents are using AI to write cover letters and resumes that can draw the attention of recruiters, customized to the job description. With all these in mind, …
Is Recruitment broken?
What we see right now is AI disrupting the recruitment function, imo.
Here is how I look at it. I am pretty sure LLMs have access to large amounts of job descriptions and relevant job titles. This is not forgetting that there might be access to details on how to write good cover letters for these job titles as well, given that a lot of career coaches will have written on these topics, as part of their portfolios. What does this translate? If you use AI to write job descriptions, followed by talents using AI to write cover letters and resumes. It will just be the left hand fighting the right hand. There won’t be any good results because the whole recruitment will suffer, even if it frees up the time for recruiters to focus more on interviewing candidates for culture and management fit (as shown in the later part of the YouTube video). Because it failed at the relevant skills dimension.
Here is what I proposed especially if speed is of the essence to reach the relevant talents. Recruiters will still use AI for recruitment but it’s a different form of AI rather, not ONLY LLMs.
Solution
Firstly, recruiters should stop writing job descriptions using LLMs without any edits/inputs from humans. Recruiters should take this opportunity to understand the skills and knowledge needed by the company and work with hiring managers to rewrite the job descriptions. Recruiters can still use LLMs but there is a strong need to edit these AI-generated job descriptions and add in more “customized” content, to look for skills and knowledge relevant to the job role and the situation in the company. As time passes by, update these job descriptions according to the macro situation and the basket of skills and knowledge within the organization. Why? This is because these job descriptions will be fresh and not seen by LLMs. Given that LLMs have not seen these job descriptions before, chances are high that the cover letters may have a lot more gaps if an LLM solely writes it without any candidate’s input. If a candidate does not put in any effort to write a good cover letter and solely relies on LLMs to generate one, it says a lot about the character of the candidate and their desire for the job.
After writing fresh job descriptions tailored to the organization's needs, there is still a good chance that the recruiters will still need to sieve through the piles of resumes and cover letters that are sent. How can AI help here then to find the relevant talent quickly, to kickstart the recruitment process efficiently?
There are two things you can do to achieve that.
Use cosine similarity to determine the “distance” between the cover letter & resume to the job description.
Use designed scorecards to score cover letters and resumes.
Cosine similarity is a very simple formula to quickly determine how far apart the cover letter and resume are as compared to the job description. Given that the job description is not seen by LLMs, there is a good chance that if the similarity is high, the candidate will have likely written the cover letter and resume after reading through the job description.
A scorecard that is designed according to the organization will expedite the selection process, moving the high cosine similarity cover letters and resumes and ranking them according to desirability. For instance, the scorecard could give a higher score to candidates who come from relevant universities, for example, a software engineer candidate from Stanford’s computer engineering school. Or been hired before for a similar role in a reputable organization. The scorecard should be rule-based, improved upon, and thought through by the HR and the hiring manager, PLUS the scorecard should be considered a corporate secret, otherwise it can be gamed easily.
Conclusion
Now is a good time to look at the recruitment process and design it in a way that AI can be tapped into while managing the impact of AI that is being used by the talent side. I offered a solution to start over again and write the job description from scratch, working together with the hiring manager and having a process to continuously update it. This is followed by using an algorithm called cosine similarity to see how far apart the newly written job description is from the cover letter and resume. On top of that, use a scorecard to measure the desirability of the candidate and be able to identify and expedite the recruitment process for these desired candidates.
What are your thoughts on this? Will be keen to hear especially if you are in the recruitment function. PM me on my LinkedIn.
In case this is the first time you are here, I have a blog site for long posts. Below are a few recent posts. And if you feel generous today, you can make an anonymous donation here without signing up for an account (from what I know). Thank you for being so supportive! :)
Blog posts:
Past Popular Issues:
Celebrating the Monkey or Pedestal?
Was shown the following IG post from Simon Sinek (who shall not need any introduction). It is from his podcast and interview with Dr Astro Teller. whom at the time of writing is the “Captain of Moonshots at X”.
Show the Value, Not Algo
One important skill a data scientist or analyst should have is to be able to communicate the insights and what is the business value it can translate into. If I were to make it more straightforward it is about three things, more revenue, reduce costs, and increase profits.