The HR and planning department consists of a mix of administrative processing, text work, conversations, and assessments. Some of these tasks are purely executive and easy to automate. Others directly touch the interests of an individual, and that requires human judgment. We have mapped 53 tasks in this domain and classified them based on what the task itself demands, not based on what would be desirable for the organization.
Tasks that AI can take over generally have a clear input, a fixed structure, and no room for judgment that touches a person. Think of posting an approved vacancy on the right channels, scheduling job interviews based on available calendars, or creating a personnel file once the basic details are known. These tasks have one thing in common: a decision has already been made by a human, and the task is the execution of that decision. Recording a sick leave notification also falls into this category, as it is an administrative record of facts, not a judgment about the employee.
What characterizes this category is that a mistake is usually small and recoverable. An incorrectly scheduled appointment can be corrected, a file can be supplemented. That makes full takeover responsible, provided the source data is correct. Anyone using AI for this kind of task would do well to know which data your company may not use is, because even an executive task can go wrong if the input is not in order.
The second category is larger and more varied. Here AI drafts something or makes an initial assessment, but a human approves or rejects it and gives a reason for that. Drafting a job posting text is a good example of this: AI can write a text based on a job profile that fits the organization, but the hiring manager assesses whether the tone, the requirements, and the emphasis are correct before the text goes out the door. Drafting an employment contract also falls into this category: AI can generate a contract based on the role, salary, and collective labor agreement terms, but someone must check whether the conditions are correct before a signature is put under it.
Screening job applications also falls into this category, and it is precisely here that the reason for human oversight is clearest. AI can make an initial selection based on a CV and cover letter faster, but the outcome directly affects an individual's chances of getting an interview. Without a human who assesses the shortlist and can explain why someone does or does not proceed, a black box arises in which no one can any longer explain why a candidate was rejected. This is not only a quality issue but also an accountability issue: if a candidate asks why he was rejected, there must be a traceable reason, not an automatic outcome without explanation.
This category requires a fixed way of working: AI makes the proposal, a human approves or rejects it, and that choice is recorded with a reason. How that assessment looks in practice and which criteria belong to it, we explain on the page about how we assess a task.
The third category consists of tasks in which the conversation, the judgment about a person, or the negotiation itself is the core of the task. Conducting job interviews is the clearest example of this: it is not just about answering questions, but about assessing motivation, attitude, and fit, something that is and remains fundamentally human work. The same applies to negotiating employment conditions, where room, expectations, and relationships are probed in a way that follows no script.
Conducting sick leave conversations also remains human work. A conversation about recovery and workload capacity requires empathy and the ability to probe further in a way that fits the situation of that moment. Checking references sits in a greyer area: making contact and asking the questions can be supported to a limited extent, but weighing what a reference does and does not say, between the lines, is a human skill.
These tasks have in common that the outcome does not depend only on the input, but on the moment, the tone, and the interaction between two people. That is difficult to structure and therefore difficult to hand over to a system that works based on patterns from the past.
This classification says something about the task, not about the employee who currently performs the task. Whether freed-up capacity leads to different deployment, less hiring of external staff, or a different staffing setup is a choice that lies with the employer and for which their own legal requirements apply. Our classification provides no justification for this; it only describes which part of the work can be done with AI today and which part continues to require oversight or human action. Anyone using these outcomes to consider further steps does so on their own basis and with their own advice.
The classification into three categories, incidentally, recurs in almost every department we have examined. Anyone wanting to see how the same approach plays out for a different function can view the comparable breakdown for customer service or for the finance department. The pattern is always the same: execution without third-party interests can be taken over, assessment with consequences for a person requires oversight, and the human conversation remains human work.
This page describes the pattern for the department as a whole, but every team has its own mix of tasks and therefore its own outcome. If you want to know what that looks like for your situation, you can fill in the free quickscan from ftetoai: twelve questions, no account needed, resulting in an indication of which part of the hours in your profile can be taken over by AI today. The full work scan, which goes down to task level, is still under construction; we therefore do not yet offer that here, but the quickscan already provides a first, substantiated indication.
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.