Recruiting with an ATS?

Your ATS might be rejecting your best employee

The best person for the job doesn't always look like the person you imagined when you wrote the job description.

Updated September 2026

Some of the best people I've ever hired shouldn't, on paper, have been the best person for the job.

They didn't tick every box.

They hadn't necessarily worked in exactly the right industry. Their career hadn't followed the obvious route. Sometimes they were missing something we'd originally put down as “essential”.

But there was something about them.

·      Curiosity.

·      Energy.

·      Judgement.

·      An unusual way of looking at a problem.

·      The ability to connect things that other people didn't.

Sometimes halfway through the interview you found yourself thinking:

Hang on. This could really work.

The wild cards.

And I'm increasingly wondering what happens to those people when we ask technology to decide who gets through the front door.

Recruitment has a perfectly understandable problem

Employers are drowning in applications. Advertise the right job and hundreds of CVs can arrive remarkably quickly. Nobody seriously believes a recruiter can spend twenty thoughtful minutes contemplating the hidden potential of every one.

So we use technology: Applicant Tracking Systems, Automated screening, CV parsing, Keyword matching, Candidate ranking.

Increasingly, AI.

And there are perfectly good reasons for doing it.

The ICO recognises that automated recruitment tools can help organisations process large volumes of applications quickly and consistently. The UK government's guidance similarly acknowledges the efficiency and scalability AI can bring to recruitment.

That's useful.

The problem comes when: “Help us process applications” quietly becomes: “Decide which human beings are worth meeting.” Those absolutely aren't the same thing.

The 82% candidate

Imagine we've advertised for a role. We've carefully written our person specification.

·      Five years' experience.

·      Relevant sector background.

·      Specific system experience.

·      Degree preferred.

·      Management experience.

·      Commercial experience.

·      Stakeholder management.

·      Strategic thinking.

·      Excellent communication skills.

And somewhere inside our recruitment technology, candidates start being compared against what we've asked for.

One candidate looks fantastic.

92% match.

Another:

87%.

Another:

82%.

Excellent.

Except there's a fairly fundamental question.

What if the person specification isn't perfect?

We wrote it. And organisations are remarkably capable of writing job descriptions describing the person who did the job previously rather than the person who could do it brilliantly next. Perhaps that 82% candidate is exactly what we expected.

I'm more interested in something else.

What's in the 18% that doesn't match?

Because sometimes that's the most interesting thing about them.

Recruitment has always had a conformity problem

This isn't something AI invented. Humans have been perfectly capable of recruiting people who look, sound and think like them for centuries. “Culture fit” has occasionally been little more than: I can imagine having a pint with this person.

That's not particularly sophisticated recruitment either.

Good technology can actually help reduce some forms of inconsistency and human bias.

So this isn't an argument for replacing algorithms with Abi’s gut feeling.

Abi’s gut feeling has its own problems. But AI introduces a different risk.

If we're not careful, we can automate our existing assumptions and apply them faster, more consistently and at considerably greater scale.

UK government guidance specifically warns that automated screening can inherit bias from historical recruitment data. It also highlights risks from CV-matching systems built around predetermined “ideal” CVs, and notes that employment gaps can disadvantage groups including carers, disabled people and neurodivergent candidates.

That's not a theoretical concern employers should simply shrug at.

The ICO audited providers of AI recruitment tools in 2024 and made almost 300 recommendations around areas including fairness, transparency and data protection.

And in March 2026, following engagement with more than 30 employers, the ICO again called for stronger transparency, meaningful safeguards and proper monitoring for bias where automated decision-making is used in recruitment.

So yes, there are compliance questions. But I think there's a broader recruitment question too.

What are we actually asking the machine to find?

Imagine your current high performers.

Now imagine their CVs arriving before you'd ever met them.

Would they all have been obvious?

Probably not.

·      Someone changed industry.

·      Someone had a career break.

·      Someone took a strange sideways move.

·      Someone didn't have the qualification.

·      Someone had never managed quite that many people.

·      Someone's previous job title didn't accurately describe what they actually did.

·      Someone had transferable skills rather than directly comparable experience.

·      Someone was simply much better in person than they were at writing CVs.

·      And someone made you think:

“They're not what we were looking for. But perhaps what we were looking for was wrong.”

That's an incredibly human moment in recruitment.

I don't want us to automate it out.

The wild cards matter

Almost everything genuinely new comes from variation; Progress happens because somebody does something differently.

Someone challenges an assumption, someone approaches the problem from another direction, someone combines experiences nobody had previously thought to combine.

Businesses say they want this constantly.

Innovation.

Disruption.

Different thinking.

Challenge.

Entrepreneurship.

Then we build recruitment processes designed to identify the candidate whose previous experience most closely resembles the experience we've already decided they should have.

There's a contradiction there. You can't spend Monday telling employees to think outside the box and Tuesday configuring your recruitment system to reject everybody who doesn't fit neatly inside one.

Transferable skills don't always announce themselves

This is another problem with overly rigid screening; People aren't just collections of job titles.

Someone who has managed a busy restaurant may have extraordinary experience of leadership, customer service, commercial decision-making, conflict, prioritisation and performing under pressure.

Someone who ran their own small business may understand accountability in a way no job description quite captures, someone moving from teaching into corporate learning might understand how people actually learn better than someone who's spent ten years administering an LMS. Someone returning after caring responsibilities hasn't necessarily lost the abilities they had before they left.

Human beings can make those connections.

Technology can too, - increasingly - but only if we ask it the right questions and retain enough human judgement to notice when its answer isn't the whole story.

Then things get even stranger

Candidates are using AI too. Of course they are. We shouldn't be remotely surprised. Ready for the process flow?

·      Employers write job descriptions using AI.

·      AI writes the LI post

·      Candidates use AI to interpret them.

·      Candidates use AI to improve CVs.

·      AI helps write covering letters.

·      Those applications arrive in an ATS.

·      Another piece of technology parses them.

·      Another system potentially ranks them.

And somewhere in the middle of all this automation sits a rather important question:

Which actual person should we hire?

There's something faintly ridiculous about the possibility of one AI optimising an application for another AI to approve. Particularly if the eventual job requires originality.

Scores look reassuringly scientific

This is another thing worth watching. Numbers feel objective.

Candidate A: 91%.

Candidate B: 84%.

Candidate C: 76%.

It looks considerably more scientific than: “Sarah thought Candidate A was good.”

But a precise number isn't automatically an accurate one.

The score still depends on what the system has been asked to measure, what information it uses, how criteria are weighted and what assumptions sit underneath it. If we've defined the wrong version of “good”, measuring it more accurately doesn't solve the problem.

It simply gives us a very precise answer to the wrong question.

Human oversight needs to mean human oversight

This is particularly important.

A human technically being involved doesn't necessarily mean they're exercising human judgement. If the system produces a ranked shortlist and the recruiter simply works from the top down, the technology may still be shaping the decision enormously.

The ICO's current position is clear that where employers use meaningful human involvement, it needs to be genuinely meaningful and applied consistently.

In other words:

“The computer said no and I clicked confirm” isn't much of a safeguard.

Humans need permission to disagree with the machine. Otherwise we're not using AI to support judgement.

We're outsourcing judgement and keeping a human nearby for decorative purposes.

So should we stop using AI in recruitment?

No. No. No.

That would be a strange conclusion.

Used properly, technology can make recruitment quicker, more manageable and potentially more consistent. It can:

·      Help recruiters find information.

·      Summarise applications.

·      Identify skills.

·      Manage enormous volumes.

·      Spot patterns humans might miss.

·      Remove tedious administration.

·      And give recruiters more time to do the bit they're actually good at:

Recruiting people.

The question isn't whether we use technology. It's what decisions we're comfortable allowing it to make.

Digital People HR says…

Use the technology.

Absolutely.

But understand what it's doing. Know what criteria it's applying. Test whether those criteria actually predict success. Monitor outcomes for bias. Be transparent with candidates where automated decision-making is being used. Make sure people can challenge significant automated decisions where required.

And most importantly:

Leave room for the wild cards.

The candidate from the wrong industry, the unconventional career, the missing qualification, the sideways move, the person whose experience doesn't quite fit into your boxes. The person who makes an experienced recruiter stop halfway through an interview and rethink what the organisation actually needs.

Because recruitment shouldn't simply be about finding the person who most closely resembles the job description.

It's about finding someone who could be brilliant at the job. Those aren't always the same person.

Use AI to help humans recruit.

Don't ask AI to decide who humans are allowed to meet.

Want recruitment technology without losing the human judgement that makes recruitment work?

Digital People HR helps growing businesses build practical, fair and effective people processes, using technology where it genuinely makes things better rather than simply because we can.

Get in touch.