Why AI won’t be able to help you filter candidates by resume.
You’ve put up a job post. A day later, you’ve got 1,245 applicants.How are you supposed to figure out which of these applicants should progress to a first round interview?
AI has presented itself as the ideal solution. Offering to take those 1,245 applications and reduce them down to a simple score that will tell you if they should progress to an interview.
But is this the best approach?
An alternative would be to use platforms like Atlas which uses AI to ensure the data on each candidate is perfect. This means that 10 seconds of filtering can get you down to 50 candidates, and an AI copilot helps you understand each candidate more deeply.
But lets get back on track.
The goal of candidate screening
Screening is required when all the three points are true:
You want to hire the best people for your company.
You only have a set number of hours to devote to the hiring process.
You have a large number of applicants.
If you have 1,245 applicants and it takes one hour to interview each one of them, that’s four months of full time work.
The only other option is removing the quality bar (pick 20 CVs at random).
Deciding on who to interview
Reducing the number of applicants through a quick screening is the only way of running a good hiring process, which means filtering them based on criteria.
Let’s get this out the way. Any type of screening is imperfect. No matter what method you choose, you will screen out otherwise good candidates. This makes the interview process inherently unfair but that’s life.
The options you have for pre-screening candidates
The best screening methods will vary depending on the role and level, but that’s not what we’re here to talk about.
AI has promised you perfect resume screening
The first wave of Gen AI tools offers AI based candidate selection. You provide the AI with a job description and a set of resumes, AI will whittle down those 1,245 applications so that you can focus on the candidates most likely to be a match.
How does it work?
AI recommendation engines will typically extract the following items from a job description:
Number of years’ experience
Required technical skills
Required experience
It will then match those requirements to a resume and create a score based on how important those aspects are to the user.
The eagle-eyed readers will notice that AI isn’t really necessary for points one and two. It takes 10-15 seconds to enter those keywords into an ATS platform and these 15 seconds alone will significantly reduce the number of applicants that you’ve received.
AI’s offer is that it will also find the evidence of your required experience hidden in the CV.
Let’s take an example from a Front-End Software Engineer role that Rightmove posted in August of 2024:
“Is enthusiastic about the importance of testing, promotes TDD and has experience of react testing frameworks like Jest and React testing library”
Language models are perfect for matching the intent of two sentences. Yes, you could use TDD as a keyword, but what if the candidate has written Test Driven Development? The keyword would not find the candidate but an AI resume parser will.
The three problems that AI parsing creates
PROBLEM 1: Resumes are not complete documents of everything the candidate is capable of doing
AI parsers work with the assumption that every single thing that the candidate has done is on their resume. But when resumes are only a page or two in length, they simply outline what the candidate believes is important.
Therefore, an AI matching algorithm will reject people based on an experience mismatch even if there is an ability match!
PROBLEM 2: Experience with something doesn’t mean you’re good at something
Just because someone hasn’t got specific experience with a particular thing, doesn’t mean they won’t be able to do a good job.
In the same way, just because someone does have experience with something, doesn’t make them good at that thing.
Most people regularly encounter new problems to solve in their jobs. But that doesn’t mean you are constantly firing and rehiring to find someone who has that experience at that specific time.
PROBLEM 3: Candidates also use AI
If recruiters are using AI to match resumes to JDs, candidates can use AI to match resumes to JDs.
Obviously there is a distinction between lying and focusing on experience more relevant to the job but the truth is, most people have some experience in most things and it is easy to dress that up in a way that the hiring manager is looking for.
As an example, have you ever seen a resume of a sales person which shows them not beating their target?
So are AI matching algorithms a big step forward?
In short, AI matching algorithms allow recruiters to match required experience with demonstrated experience.
They will help you focus on the people that have written in their resume that they have done the things you need to be done but:
You will interview more people with the experience but with low quality execution of that experience
You will filter out everyone with the ability to do the job but without the experience to do the job
In our opinion, the cons outweigh the pros.
Atlas is an end-to-end recruitment platform, built entirely with generative AI. It handles everything – from intake, sourcing, and screening to interviews, pipeline management, and business development. It’s incredibly fast, easy to use, and AI is integrated into every feature.
