AI in recruitment? Start small. But start somewhere.

For many recruiters, AI in recruitment feels like a choice between two bad options.

Do nothing, because GDPR, the EU AI Act and systems you cannot fully explain feel risky.

Or automate too much, too quickly, and end up with a score that decides who gets to move forward.

Neither is a great plan.

The European Commission lists AI tools used in employment and recruitment, such as CV-sorting software, as ‘high risk’ under the EU AI Act. GDPR also has rules around solely automated decisions that have significant effects on people. That does not mean every use of AI in HR is banned. It does mean you cannot start with: ‘The tool said no, so no.’ [1]

AI is not a recruitment strategy

Adding a tool because everyone is talking about AI is not innovation. It is another tab.

A better question is: which moment in our process do we currently understand too superficially? At the first screen, the interview or the debrief, we often compare language and gut feeling while we still do not see behaviour. That is where AI can play a useful, bounded role. Not as the decision-maker, but as another layer of observation.

Start with one question you want to answer better

Not: ‘How do we automate selection?’

Instead: ‘How do we have a richer conversation with this candidate?’

A game-based assessment can show how someone makes choices, approaches problems and adapts as conditions change. Not through another self-report or a perfectly rehearsed STAR answer, but through behaviour in a concrete task.

That does not produce a magical truth about a candidate. It produces signals: starting points for better follow-up questions that you can test and qualify. At Assess4me, that is the point: behavioural data does not replace a recruiter’s judgement. It helps inform it.

Human oversight is not a checkbox

‘There is still a human in the loop’ is not, by itself, a strong process. Meaningful human oversight means the recruiter understands what the information does and does not say, can question a pattern and takes responsibility for the final decision.

• Bound the use case: use one assessment as additional input, not as an automatic gatekeeper.
• Make the role clear: tell candidates what you are assessing and why, in plain language.
• Keep it testable: use the output to improve questions, not to blindly rank people.
• Review it together: look back at whether the insights were genuinely useful, fair and relevant.

The legal details vary by use case, data flow and organisation. This is not compliance advice. But one principle is hard to argue with: the greater the impact on someone’s opportunity, the more carefully, transparently and humanely you should run the process.

The future does not have to be fully automated

Recruitment does not improve because ‘AI’ appears somewhere in the workflow. It improves when you guess less, see behaviour where a CV only shows claims, and ask better questions instead of reaching faster conclusions.

So start small. Pick one moment where you are not seeing enough today. Add a purposeful layer of insight. And keep the recruiter at the wheel.

That is not a slow start.

That is a grown-up start.