AI in Hiring: What Technology Should Evaluate, and What People Must Decide

AI in hiring is not one thing

Recruitment technology can perform very different functions:

The first items on this list mostly help people work. The last items increasingly take decisions away from people. Treating all of them as the same "AI hiring" creates confusion in both directions. Some teams avoid useful tools because they fear automated rejection. Others adopt tools that decide more than they realise.

What is the difference between evaluating and deciding?

Evaluating establishes what evidence exists about a candidate against the role's requirements. Deciding is the choice to hire, and it is the part people should own.

Evaluation asks: what evidence do we have about this candidate against the requirements of the role? Decision asks: do we want to hire this person?

A decision draws on the evaluation, but it also weighs things an evaluation cannot contain: how many seats are open, which candidates are available now, what the supervisor knows about the team, what the candidate asked in their last message.

Technology can assist with evaluation. Responsibility for the decision should remain with the employer and the recruiter. One model keeps people at both ends:

The rest of this article uses that model to sort out what AI should and should not do.

Where AI helps: the evaluation side

Frontline hiring has a specific problem. The Signal Gap, a term Radius Hire introduced in The Radius Papers 01 (2026), is the distance between what a hiring decision rests on and what predicts the outcome. In many frontline processes, the decision rests on a CV, a short phone call and a start date. Important requirements, such as reliably reaching the site for a 05:30 start or accepting the real shift pattern, are never asked about before the offer.

The method with the strongest research support shows what could fill that gap. A structured interview asks every candidate the same job-related questions and rates each answer against a standard written in advance. Validity here is an estimated correlation between what a selection method says about candidates and their rated job performance, corrected for measurement error: 0 means no relationship, 1 a perfect one. On the revised 2022 estimates, structured interviews average 0.42, a moderate link to job performance, and the highest of the 25 methods reviewed. Unstructured interviews average 0.19. Years of job experience average 0.07 (Sackett and colleagues, 2022). These are averages from many studies, mostly outside the Netherlands, and they measure overall job performance, not retention.

The researchers behind the revised estimates note that, in their traditional in-person form, structured interviews are generally not a viable strategy for high-volume jobs, and that when technology takes over delivery, validity has to be checked again (Sackett and colleagues, 2022).

That is where technology can help: not by deciding, but by making the evidence-gathering people already value possible at volume. On the evaluation side, technology can:

Whether that keeps the method's validity in your workplace, only your own outcome data can show. Measure it.

Should AI make hiring decisions?

No. AI can support the evaluation, but the person accountable for a hire should be the person who decides it, with the evidence in front of them. These parts of hiring stay with people:

This oversight must be meaningful. A person who clicks "approve" on every recommendation, without seeing the evidence or having the authority to disagree, is not really deciding. Three tests show whether a person really decides: a named owner, a record and a real ability to overrule.

The law points the same way. Under Article 22 of the GDPR, a decision about a candidate that has significant effects may not be based solely on automated processing, unless a narrow exception applies, such as necessity for entering into a contract. Guidance from European data protection authorities (WP251rev.01) adds that human involvement must be meaningful rather than a token gesture. It must come from someone with the authority and competence to change the decision.

AI used to recruit or select people is high-risk under Annex III of the EU AI Act. After the Digital Omnibus on AI (Regulation (EU) 2026/1744), the main high-risk duties apply from 2 December 2027. Some rules already apply: the ban on inferring emotions at work from biometric data such as face or voice (since 2 February 2025), AI literacy duties (since 2 February 2025), and the duty to tell people when they are interacting with an AI system (since 2 August 2026).

Article 14 requires providers to build high-risk systems so the people overseeing them can understand their limits, stay alert to over-reliance and override the output. Article 26(2) requires employers to give that oversight to people with the competence, training and authority to use it.

Why does ranking candidates move the decision?

A count shows, for each candidate, which requirements are supported by evidence, which rest only on a claim and which are missing, for example "meets 4 of 5 requirements with evidence". That is evaluation. A ranking orders candidates against each other, usually by a score, and quietly tells the recruiter where to start.

A ranking can move part of the decision into the tool, even when a person formally makes the final choice. Counting has limits too: a count is still an automated assessment, and an interview answer stays the candidate's own claim until someone checks it or rates it against written criteria.

Who does what: a practical split

Adjust this split for the role, the risk and the law.

In the Netherlands, an employer may check an applicant's identity by looking at the original document, but may not copy, scan or photograph it, or ask for the citizen service number (BSN), until the person is hired (Autoriteit Persoonsgegevens).

What this means for frontline hiring

Frontline hiring is where this line matters most, and where thoughtful technology can help most. Frontline roles are often filled quickly and in large numbers.

That creates a familiar tension. Volume and limited recruiter capacity push teams towards fast, thin decisions. Consistency is hard when every recruiter runs a slightly different phone screen. Candidates wait days for a call, or never hear back. Structured evidence is valuable but slow to collect by hand.

Technology can take on the repetitive, volume-heavy evaluation work: asking every candidate the same questions whenever they are available, and organising the answers. That leaves recruiters with the part that needs a person: reviewing evidence and deciding. The risk is the opposite design, where technology also takes on the deciding, quietly, through filters and rankings nobody inspects. The other risks of AI in hiring sit on the same line: over-reliance on output nobody checks, bias that nobody measures, and uses the law already restricts.

Before you use AI: three questions

Before adopting any tool, ask:

How Radius Hire applies this line

We build for the evaluation side of this line. Radius Hire asks every applicant the same structured questions, built on the requirements the employer sets, and shows the recruiter the evidence behind each requirement, labelled Proven, Claimed or Missing. It does not rank candidates or reject anyone on its own. A recruiter decides.

Applicants answer by voice in their own language: Dutch, English, Polish, Romanian, Ukrainian, Russian or French. They are told at the start that the interviewer is automated and that a person reviews the result, and they can ask to speak with a recruiter instead.

Under Annex III we treat Radius Hire as a high-risk AI system and design it for human oversight ahead of 2 December 2027.