Working in the software and information industry for quite some time now, so I had my share of arranging technical interviews and team building. Besides that, I worked for an HR testing and placement company for a while, so I’m familiar with the backoffice aspects of recruitment including competency testing and doing statistical similarity between applicants resumes and job descriptions.
Usually the process follows the following guidelines:
- Define the job description.
- Meet with the HR to define together the profile of the candidate and having he/r exposed to the more soft aspects of the job and team, so that the HR interview will take into consideration those aspects
- Publish the job internally and/or in job boards and get ready to get flushed with Resumes.
- HR is doing the first screening using ATS (Applicants Tracking Systems) that score the relevance of the resume test to the job description.
- Out of hundreds of applicants only the top scoring candidate are contacted by HR and further screening interviews.
- Those that pass that stage are invited to a professional interview and/or some testing out of a bank psychological testing, and some time give some professional challenge to the candidates.
- Those that survive that stage are given a payment proposal and if they agree they join the team.
This sorting process is broken in two places:
- The first, is that ATS resume filtering is prone to low accuracy – Regardless of how good the filtering algorithm is! The problem lies in the fact that most ATS are not connected to the actual performance of the candidate once she’s on the job, so there is no positive feedback loop that correctes wrongly placed candidates. Nor is there a mechanism for identification of errors of type II (false negatives) since you never know what you miss!
- The second break line is the fact that job description and HR evaluation as well as the candidate resume are very sparse on the actual data needed for a successful job placement, and there is always pressure to reduce the cost of recruitment. So, the fitment is very rough, and that’s the reason companies are pushing “bring a friend bonus”.
Years ago, when developing a psychological application implementing test banks, I learned that the best predictors for job success is not GPA grades or creative logic problems, but the following traits:
- Cognitive ability – Logical reasoning always helps whether you are developing software or driving a truck.
- Emotional stability – Big Five Personality skills test is a good predictor for how good one can cope with setbacks and a hard nosed boss.
- Being creative – The ability to deal with the unforeseen challenges, originality and the ability to solve problems. This covers all the things you don’t know about the job description, and the candidate will have to deal with.
- Ability to change – Across all job types, if you’re able to change and adapt to changing conditions, you’ll survive, again, this works even if the job description is bad, since the candidate will adapt.
- Historical performance – If one did good in the past, there is a good probability he will do so in the future.
- Ability to learn independently and with agility – Again if you are creative, willing to learn and actually seeking new knowledge, the more you do better regardless of the specific details of the job.
What we can see is that good grading according to psychological traits is really more about comparing candidates and less about job fitment. So what if you have two candidates, one with better learning ability than the other and a worse job history, and I’m not talking about the cost of testing and their reliability?
How can AI change the equation? – We know that generative AI is already doing great assistive services to knowledge workers, whether by consulting it via chat or using it in automations. In a way, generative AI as a negative effect as it’s normalising five of the 6 traits mentioned and reducing the differences between candidates as you can see in the 2023 Mckinsey report below:
What we are left with is personal historical performance, and this will lead to an even more inclination to rely on friends’ recommendations. Yet there was never a candidate that provided bad references, and it’s not simple to interview referrals. So a better approach is to review the social network of a person, which can be done automatically and objectively and infer from the relevant compatibility and ranking for the job.
This is why LinkedIn is proliferating, in spite of being a boring platform filled with Cheerleaders and Yea sayers. It’s a perfect platform for mining links and relevant performance content automatically.
In a way, you can implement the saying “Tell me who your friends are and I’ll tell you who you are” quite literally, by finding and understanding the right connection profiles to an existing worker that is leaving or doing profiling of links that are with relevant content, skills, etc. The sky’s the limit!
LinkedIn can be integrated with internal HR tools, so one can mine data from the organization and join it with external data, or even implement the same logic if you have something like Microsoft graph in your organization or similar platforms.
So, to summarize LLMs will have a negative impact on recruitment processes which can be countered by better link analysis of internal and external candidates, as well as reducing barriers to entry and lowering fees for Juniors’ positions while increasing productivity overall.
