The HR department combines two types of work that behave very differently under automation. On one side are tasks that largely consist of fixed steps: drafting texts according to a format, transferring data from one system to another, scheduling appointments based on availability. On the other side are tasks in which someone's career, health or income is at stake. These two types of work require a different approach, and our taxonomy of 53 tasks in this domain shows that pattern consistently.
Part of HR work is essentially administrative or textual in nature, with a clear starting point and a clear end point. Drafting job postings based on a job profile is an example of this: the input is known, the structure is largely fixed, and the result still goes to the hiring manager for approval. Posting vacancies on channels falls into the same category: once a text has been approved, distribution to job sites and social media is a matter of following formats and links, without requiring any judgment.
Creating a personnel file for a new employee is also largely structured input: data and documents are transferred into the HR system according to a fixed pattern, with little room for interpretation. Recording absence – noting the date, reason and expected duration – is comparable: a report is logged, not assessed. These tasks fall into category 1 not because they are unimportant, but because the outcome is predictable once the input is correct, and because a human still looks at the document further along in the process before it is used.
A larger part of HR work sits in between: AI can deliver a proposal, draft or initial assessment, but someone must approve or reject it and be able to justify that. Screening job applications is the clearest example here. Assessing CVs and cover letters against job requirements can largely be automated with support, but the shortlist that goes to the hiring manager must be explainable: why these candidates and not those others. Without that explanation, there is a risk that selection criteria become uncheckable, and that directly touches on the diligence that also applies in diligence when cutting positions, namely that an automated outcome should never be the final word without someone being able to verify the reason.
Drafting employment contracts falls into the same intermediate category: a contract can largely be generated automatically based on position, salary and collective labor agreement provisions, but the document must be checked before it can be signed, because errors in employment terms have direct legal consequences. Scheduling job interviews can also largely be automated, as long as there is a human who can catch exceptions – a candidate who does not fit the standard schedule, an assessor who needs to shift. What these tasks have in common is that the AI delivers an outcome that someone with knowledge of the context must be able to confirm or send back, with a reason attached. What that oversight looks like in practice depends on the task; see human oversight of AI, concretely for how that oversight is set up.
A third group of tasks revolves around personal contact, judgment about people, and sensitive conversations. Conducting job interviews requires an assessment of motivation, attitude and fit that cannot be captured in criteria; the interview report with advice is the result of an interaction, not a calculation. Checking references requires something comparable: it involves interpreting what someone does and does not say about a former colleague, something where tone and nuance say just as much as the facts.
Conducting absence conversations also belongs here: a conversation about recovery and capacity affects an employee's health and income, and requires a relationship of trust that cannot be transferred to a system. Negotiating employment terms with a candidate or employee is equally human work: it involves weighing interests within a policy framework, where the outcome depends on what someone needs at that moment and what the organization can offer. These tasks remain human work not because automation is technically impossible, but because the core of the task is a human interaction that cannot be replaced without loss.
The shift in HR work, then, is not about entire jobs disappearing, but about tasks changing in composition: less time on input and distribution, more time on conversations, assessment and oversight. That is a different matter from workforce planning, and decisions that affect that are subject to their own legal requirements; those fall outside what a task analysis can substantiate. It is, however, wise to involve the works council in good time, for example via informing the works council about AI, and to look at how HR tasks relate to comparable shifts elsewhere, as described in what work can AI take over in administration and what work can AI take over in customer service.
Anyone who wants to know how this pattern plays out for their own role or team can fill in ftetoai's free quickscan: twelve questions, no account required, resulting in an indication of what portion of the hours in that profile can be taken over by AI today. The full work scan, which maps an entire department, is still under construction; it does not yet exist at this time, and we would rather report that honestly than offer something that isn't there.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.