AI in recruitment and selection: what works for SMEs (and what is off-limits)?

Key Takeaways: AI can noticeably speed up hiring, writing job ads, answering candidate questions, summarising interviews, but the moment AI starts to assess or rank people, you enter heavily regulated territory. Under the EU AI Act, selection AI systems count as high-risk, with requirements for transparency, human oversight and non-discrimination. The practical line for SMEs: use AI freely for the supporting tasks (low risk, immediate time savings), and be cautious and careful with the assessing tasks. This article shows what works, where the limits lie, and how to deploy AI in recruitment responsibly.

The divide: supporting versus assessing

The whole recruitment story breaks down into two categories, and the difference determines both the return and the risk.

Supporting AI does not touch the selection decision. Think of drafting a job ad, answering candidate questions, scheduling interviews and summarising interviews. This is low risk and delivers immediate time savings.

Assessing AI ranks or selects people: automatically scoring CVs, ranking candidates, predicting who is suitable. This affects someone's chances of employment and therefore falls under the strictest rules. Here AI may support, but never decide on its own.

For most SMEs the gains, certainly at the start, lie in the first category.

What already pays off safely

Four supporting applications deliver value quickly with little risk.

Writing and improving job ads. AI drafts a clear, inclusive job ad in minutes and helps avoid jargon and unintentionally exclusionary language. You refine and publish.

Candidate communication. A chatbot answers applicants' frequently asked questions (about the role, the process, the terms) and schedules interviews, 24/7, without your HR team handling every question manually. This builds on the same chatbot technology as customer service.

Summarising interviews. AI summarises job interviews with the key points, so you can compare faster and more consistently, without leaving the decision to the AI.

Administration and scheduling. Scheduling, confirming and following up on appointments, an ideal task for a lightweight AI agent.

Where the limits lie: the EU AI Act

As soon as AI assesses or ranks candidates, you are using a high-risk AI system within the meaning of the EU AI Act. That brings concrete obligations: transparency (candidates must know AI is being used), human oversight (a human assesses and bears final responsibility), documentation of how the system works, and measures against discrimination.

The most important limit is simple: fully automatic rejection is not allowed. An AI may support a pre-selection, but a human must be able to assess and justify every rejection and invitation. The full obligations are in our EU AI Act compliance checklist.

The bias risk, and how to manage it

The biggest danger of assessing recruitment AI is unintended discrimination. A model that learns from historical hiring decisions inherits the biases of the past: who was hired before determines who the model now considers suitable. This way you quietly reinforce existing skew instead of correcting it.

You manage it with four measures: do not let AI select on its own, actively check models for bias, document how decisions are reached, and make a human ultimately responsible. This is not only legally required, it is also the best protection against reputational and legal damage. Good governance belongs here as standard, just as with MLOps where models are continuously monitored.

How to start sensibly

Start with the supporting tasks: let AI speed up your job ads and candidate communication, and measure the time saved. That delivers immediate results with minimal risk. Only consider assessing applications at high hiring volume, and then only with the right governance, transparency, bias checking and human oversight, around them. Record the usage in your AI usage policy, so your team knows what is and is not allowed.

Stratalytic and AI in HR

We help you deploy AI in hiring where it pays off, within the limits of the law:

• Quick wins: AI for job ads, candidate communication and interview summaries.
• Responsible: high-risk applications only with transparency, bias checking and human oversight.
• Compliance: set up in line with the EU AI Act and GDPR.
• Subsidy: development often falls under WBSO; training under SLIM.

Schedule a 30-min intro call and we will look at how to speed up your hiring without legal pitfalls.

Frequently asked questions

Am I allowed to use AI to select job applicants? Yes, but under strict conditions. Selection AI is high-risk under the EU AI Act, with requirements for transparency, human oversight and non-discrimination. Fully automatic rejection is not allowed.

Which applications are low-risk? Writing job ads, summarising interviews, answering candidate questions via a chatbot and scheduling appointments. These do not directly touch the selection decision.

How do I prevent discrimination? Do not let AI select on its own, check for bias, document the decision-making and make a human ultimately responsible.

Does AI in recruitment pay off for a small company? For the supporting tasks, certainly. The heavier, assessing applications require more governance and only become interesting at high hiring volume.