Employees continuously ask HR questions: how many leave days do I have left, how does the travel expense scheme work, what is the procedure in case of illness, where do I report a change of address. This is a task that starts with a question and ends with an answer given or a referral to a colleague. Precisely that boundary makes it interesting to scan: it is not an open advisory process, but a bounded action with a clear end point.
Three axes determine the picture here, and they turn out favourably.
The volume is high. In every organisation with more than a handful of employees, the same questions come up time and again: leave balance, sick-leave reporting procedure, working-from-home policy, expense regulations. A high volume of similar questions is precisely where automation pays off, because the time saved per question is small but the sum total across hundreds or thousands of questions per year is considerable.
The degree of structure is reasonable, though not perfect. Arrangements are fixed in a collective labour agreement, staff handbook or HR system, and most questions can be traced back to a documented answer. But not every question is a simple lookup: "can I save up my leave for a sabbatical" touches on exceptions, transitional arrangements or individual agreements that are not always equally well documented everywhere.
The client contact is internal and low-threshold. An employee asking about their leave balance expects a quick, factual answer, not a personal conversation. That makes the difference with, for instance, an external client conversation where tone and relationship carry more weight. For internal, factual questions, an automated answer is often just as usable as an answer from a colleague, as long as it is correct.
The room for judgement is limited but not zero. A question about the number of leave days is a calculation. A question about "can I have my leave paid out because I was overworked during the coronavirus period" requires interpretation of a specific situation, possibly combined with earlier agreements. As soon as a question cannot be traced one-to-one to a rule, room for assessment is needed — and that is precisely the point where the task shifts from category one to category two: AI can produce a draft answer, but an employee approves it or sends it back.
Compliance also plays a role here. Leave and sick-leave arrangements touch on employment law and sometimes on the Gatekeeper Improvement Act (Wet verbetering poortwachter). An incorrect answer about sick leave can have consequences that go beyond an uncomfortable conversation. That does not mean AI cannot be deployed here, but it does mean that an answer carrying legal weight needs an escalation path to a human who bears final responsibility.
The cost of errors is moderate: a wrong answer about office hours is annoying but reversible, a wrong answer about notice periods or dismissal procedures can mislead an employee with real consequences. That difference in weight within a single task category is precisely why a blanket yes or no does not suffice here.
The outcome is not a wholehearted "yes, AI takes this over". It is a layered answer: the vast majority of standard questions — leave balance, procedures, where to find which form — is readily automatable because it is structured, high-volume and low-risk. Questions that touch on personal circumstances, exceptions or legally sensitive topics fall into the intermediate category: AI drafts an answer, an HR employee approves or rejects it, with reason given. Questions that require genuine tailoring — an individual case file, a conflict, a precedent not yet recorded — remain human work.
This outcome depends heavily on how well the underlying knowledge has been documented. An organisation with an up-to-date, unambiguous knowledge base of arrangements and procedures gets far more out of automation than an organisation where agreements sit loosely in emails and in the heads of HR staff. A sector with many collective-agreement variants, shift work or individual contract arrangements also pushes the degree of structure down, and thus pushes the task towards more human oversight. A company with a small, homogeneous workforce again has less of a volume advantage, which means investing in automation pays off less quickly.
Today, AI can only work with text here: answering questions based on documented arrangements. That requires an up-to-date knowledge base that is actually correct, and a clear escalation path for questions that fall outside the standard. Without these two building blocks, automation here mainly creates confusion, not time savings.
If this change is also linked to a redistribution of HR tasks or a smaller HR staffing level, it quickly becomes more than a tool choice. In that case, also look at when the works council has a say and how to inform the works council about AI. Personnel decisions that might follow from this are subject to their own legal requirements; that falls outside the scope of this scan.
Would you like to know what proportion of HR questions in your organisation qualify for automation, and which part still requires oversight? The free quickscan from ftetoai gives an indication, in twelve questions, without an account, of what proportion of the hours in your profile can be taken over by AI today. The full work scan, which goes deeper into individual tasks and systems, is still under construction.
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.