AI in HR, what to automate and what to leave alone
Where artificial intelligence speeds up hiring and people processes, and why certain decisions must stay human for legal and practical reasons.
HR is an area where artificial intelligence offers obvious savings and equally obvious risks. The line between sensible and problematic use is fairly clear: automating the preparation of information is safe, automating decisions about people is not.
What is worth automating
Internal employee questions. How many leave days do I have, how do I claim an expense, what does the insurance cover, what's the remote work policy. This is the case with the best ratio of volume, risk and benefit. An assistant over current policies resolves most of these and takes load off the team.
Summarizing and organizing applications. Reading a hundred CVs and extracting experience, education and availability into a comparable format. The system organizes the information; a person decides who to interview.
Drafting job descriptions. From real responsibilities and requirements, generating a first draft consistent with the company's tone. Saves time and improves consistency across openings.
Interview preparation. Generating relevant questions based on the role profile and the candidate's background, so the interviewer arrives prepared.
Analysing internal surveys. Processing hundreds of open-text responses, identifying recurring themes and quantifying them. It turns material that usually goes unread into actionable information.
Process documentation and onboarding. Assembling induction materials from existing documentation and keeping them current.
What is not worth automating
Automatic candidate rejection. Beyond the legal question, it's a bad idea in practice: the model learns from past hires and reproduces the organization's historical biases. If a certain profile was favoured before, the system will favour it again.
Performance reviews. A summary of objective information can be a useful input, but the review itself is a conversation between two people, with context no system holds.
Promotion or termination decisions. Not only because of legal exposure: these are decisions where responsibility has to be identifiable and explainable.
The legal framework worth knowing
The European AI regulation classifies systems used in recruitment, selection, promotion and termination as high risk. That brings concrete obligations: documentation, human oversight, risk assessment and transparency toward affected people.
In the United States, several jurisdictions require bias audits for automated employment decision tools, plus notification to candidates.
Even if your company doesn't operate in those markets, the standard is reasonable: if a candidate asks why they were rejected, there has to be an answer a person can give and stand behind.
How to reduce bias risk
Three concrete practices:
Explicit, written criteria. If the system ranks applications, it should do so against criteria the function defined in advance and can explain, not an opaque score.
Sample review. Periodically review a batch of rejected or low-ranked applications to spot patterns. It's the equivalent of quality control.
Distribution measurement. Compare the composition of candidates entering the process against those advancing. A systematic difference means something in the criteria needs examining.
What improves candidate experience
An under-discussed and highly valued use: keeping applicants informed. Much of the poor reputation of hiring processes comes from silence. A system that reports status, answers process questions and communicates decisions within a reasonable time improves how the company is perceived at very low cost.
Where to start
The recommended order is the reverse of what's usually proposed: start with internal employee questions, move to organizing information in recruitment, and only then approach candidate evaluation — with written criteria and human oversight on every decision.
Frequently asked questions
Can AI be used to screen CVs?
Yes, to rank and summarize applications against explicit criteria the company defines. Automatically rejecting candidates without human review is risky: it creates biases that are hard to detect and, in several jurisdictions, triggers legal obligations to explain decisions.
What legal risk comes with using AI in hiring?
Automated decisions about people are regulated in the European Union and in several US jurisdictions, with requirements around transparency, bias auditing and the right to human review. Involve legal before implementing.
Which HR use of AI has the best return?
Answering internal employee questions about policies, leave and benefits. High volume, low risk, and it frees the HR team for higher-value work.