AI in Recruitment: Where should AI stop and recruiters should start?

Deterministic software produces a predictable, repeatable output. Every piece of recruitment technology used before AI such as your ATS, your payroll system, your job board posting tool - works this way. You put data in, you get a known result out. If something goes wrong, you can see it and fix it. 

Probabilistic AI generates outputs based on patterns and probability, which means the same input can produce different results at different times. Ask an AI tool to draft a candidate outreach email and you might get five different versions. Ask it to assess a candidate's suitability and the answer shifts depending on how the prompt is framed, what data it's drawing on, and factors you can't always see or control. 

For most tasks, that variation doesn't matter. A slightly different outreach email is still a useful outreach email. 

For hiring decisions, where the cost of a wrong call is measured in months of lost productivity, damaged client relationships, and real money, probabilistic outputs need a human in the loop.  

A Q1 2025 Gartner survey of 2,918 job candidates found that 32% were concerned about AI potentially failing their applications, and 25% said they trust employers less if AI is used to evaluate their information. Only one in four candidates (26%) trust AI to fairly evaluate them, even though just over half (52%) believe AI is already screening their application. (Gartner, Voice of the Candidate Survey Analysis, Q1 2025) 

For recruitment agencies, that trust gap is a commercial reality. Candidates who feel processed rather than assessed disengage. The ones with options gravitate toward processes that treat them like people. 

At the same time, candidates are increasingly using AI on their own side of the process. A Q4 2024 Gartner survey of 3,290 job candidates found that 39% used AI during the application process - generating CV text, writing cover letters, and answering assessment questions. The boundary between authentic skills and AI-generated presentation is getting harder to read, which means the human judgment required to assess real fit is becoming more valuable, not less. (Gartner, Voice of the Candidate Survey Analysis, Q1 2025) 

The admin burden compounds this. A 2024 SmartRecruiters survey of 533 talent professionals found that 45% of TA leaders spend more than half their working week on administrative tasks that could be automated - leaving limited time for the high-judgment work that actually drives placements. (SmartRecruiters, Talent Acquisition Benchmark Report, 2024) That's the time AI should be giving back - and it does, when the boundary between automated and human work is drawn in the right place. 

For every AI touchpoint in your recruitment process, ask two questions: 

1. Is the output predictable and verifiable? If you ran this task ten times with the same inputs, would you get the same result? And if the result is wrong, would you know? 
2. What's the cost of an error? If the AI gets this wrong, is it a minor inconvenience or a significant problem for the candidate, the client, or your business? 

Map the answers and a clear picture emerges: 

The top of that list is where AI earns its keep. The bottom is where your recruiter earns theirs. 

The tasks in the bottom half of that table share something: they require judgment that can't be reduced to pattern recognition. 

Dominic Waters, who has spent over 30 years studying what actually predicts performance in a role, is direct: it's rarely what shows up on a CV. 

There's a lot of concern about people who look very well prepared, obviously these days perhaps with a bit of help from AI at interview, and actually turn out being a very different person to the one that presented at that point.

The qualities that predict whether someone will thrive such as appetite, resilience, how they respond to rejection, whether they'll pick up the phone when it's uncomfortable, don't surface in a structured interview or a CV parse. They require a different kind of assessment entirely. 

Laura Tressler, General Manager at Access Recruitment, sees the same pattern from the client side: 

What is really surfacing up is that we need people who are not becoming lazy with AI. They are really exercising great leadership skills, and they show great emotional intelligence, those softer skills. What we need from the recruiters interacting with us is the recruiter themselves to actually showcase that they've really understood that from the individuals they're putting forward.

A candidate who has prepared thoroughly, with AI assistance, can present a version of themselves that looks ideal on paper and sounds compelling in a video screening. Whether that person has the resilience to push through a difficult quarter, the emotional intelligence to manage a complex client, or the drive to pick up the phone on a hard day - that requires a human to assess. 

When you automate the top of the task list, you accelerate throughput. More CVs processed, more candidates ranked, more outreach sent. Speed is real and valuable. 

The question is whether the judgment at the end of that pipeline has kept pace. If the algorithm is pattern-matching against your historical hires, you're replicating your existing talent pool faster. Biases that were already there get amplified. 

Neil Carberry OBE, CEO of the REC (Recruitment & Employment Confederation), raised this directly: 

Clearly we can improve throughput in the system by automation and by AI. But equally, that kind of [thing] might lead us to just accelerating our existing biases rather than improving decision-making.

The human in the loop is there to make sure efficiency gains at the top of the process translate into better decisions at the bottom. 

The recruitment businesses getting this right share a common characteristic: they've made a deliberate decision about what their consultants are for before they chose the tools. 

The question that matters is: if AI handles the deterministic work, what do your people do with the time? 

Laura Tressler describes what this looks like in practice: 

The recruiter's role is to challenge the brief before it even gets to assessment. You've asked as a client for ten years' experience, but what's the actual outcome you need? Because maybe you don't need ten years' experience today because these three tools that could be in that business might help. You want them to be able to have those conversations and advise, and then just use the systems to be able to deliver the throughput in the fastest way possible.

That shift from administrator to advisor is only possible when the deterministic work is genuinely off the recruiter's plate. And it only happens when adoption is led from the top of the business. 

Where I see technology working and adoption really driving well is when we start to think about how do we not use the technology to be just a system of record, but a system of action. And that means you've got to lead from the front and lead at the top of the organisation, showing people a new way of working and how they can reimagine recruitment.

Laura Tressler, General Manager, Access Recruitment 

If you want to apply the deterministic/probabilistic test to your own business, start here: 

1. Where are humans currently doing deterministic work? If your consultants are spending time on tasks that produce the same output every time such as posting jobs, updating records, sending standard follow-ups - that's time AI can give back. The question is what they do with it. 
2. Where is AI currently making probabilistic decisions without human review? If your process has AI ranking or filtering candidates without a consultant reviewing the logic, that's worth examining. The AI may well be right. The issue is knowing when it isn't. 
3. Where are your highest-cost errors? Work backwards from your most expensive mistakes: wrong hires, damaged client relationships, candidates who dropped out late in the process. What stage did the process fail? Was there a human in the loop at that point? 

The goal is to be deliberate about which steps need human judgment, and design the process around that. 

AI is genuinely useful in recruitment. The firms using it well free their consultants from admin and create space for the conversations, judgments, and relationships that actually drive value. 

That happens when the boundary sits in the right place. Automate deterministic tasks with confidence. Keep a person in the loop for probabilistic decisions, where context, character, and judgment matter. 

It's what great recruitment has always been about. 

Ready to draw the line in your own business?

We'll show you exactly how recruitment businesses are using Access Evo to automate their deterministic work and give their consultants back the time to do the rest.