HR questions

You can ask FixControl an HR question the way you'd ask a colleague — about leave, a line on your payslip, a procedure, or a situation at work. What happens next is not "the chatbot answers": FixControl first works out who you are, what you may be answered from, and how sensitive your question is. This page explains that from your side.

HR answering is off unless an administrator has switched it on for a project. If nothing on this page happens when you try, HR is not enabled for you — see HR governance.

Where you can ask

Teams and Slack deliberately answer only in a one-to-one conversation. An HR answer can carry personal information, and a channel or group chat is a shared room. If you ask in a shared conversation, the bot replies with a short note explaining why and does nothing else — your question isn't processed, isn't stored as a case, and isn't even sent to the AI for classification. Ordinary work chatter in a channel (approvals, ticket questions) keeps working exactly as before.

Your chat account also has to be linked to your FixControl account. If it isn't, you'll get the link instruction instead of an answer. That's on purpose: FixControl would rather not know who you are than guess.

Who can see your question

• You always see your own question and the answer to it.
• For a normal question, the HR colleagues whose role covers that kind of work can see it — the same way a service desk sees the requests it handles.
• For a sensitive question, the circle is much smaller. Only the reviewer your case was explicitly assigned to can open it. Nobody gets in "because they're an administrator": being a tenant administrator is not HR access, and a technical administrator sees empty lists.
• Someone with the right to assign work sees only that a restricted case exists — its status, its sensitivity, its route — not what it says.

What happens to a sensitive question

Every question gets a sensitivity level: normal, sensitive, or highly restricted.

A first, rule-based check sets a floor. Today it recognises Dutch wording:

• questions about harassment, bullying, discrimination, unwanted behaviour or the confidential adviser (vertrouwenspersoon), and questions about illness, medical matters, therapy or pregnancy, are highly restricted;
• questions about your salary or payslip, or about your own leave balance or personal details, are sensitive;
• questions about the collective agreement (CAO), policy, parental leave or a procedure are normal.

There is no rule of its own for legal or compliance questions: they take the restricted route when the AI's proposal classifies them that way. The AI's proposal can raise the level set by the rules, but never lower it — nothing can talk a question down into a lighter route.

A question the rules do not recognise — which today includes any question that is not written in Dutch — is never answered directly. You get a clarifying question, or it goes to a person.

A sensitive question does not get a direct answer either. It becomes a case, and its level decides which permission a reviewer needs before the case can even be handed to them. Only after that assignment can anyone read it and work on it.

What the AI is shown

Only the text of your question, for the single purpose of proposing how sensitive it is.

Your personnel file, retrieved documents, and the body of the answer are not sent to a model. When an answer genuinely needs personal data, the server reads a small, named set of fields straight from the HR system — things like your employment percentage, your remaining leave hours with the date they were valid, or a payroll explanation code. Not your file.

When you get a direct answer, and when you don't

You can get a direct answer when the rule-based check recognised the question as a general policy or procedure question, the source is current and authoritative, no two sources contradict each other, and the system is confident enough. The answer names the source and its version.

Anything that touches your own situation becomes a draft or a review case instead. Before that answer reaches you:

• someone has to approve it, and the approval is tied to that exact answer text — edit the answer and the approval is void;
• the person who requested the approval can never be the person who grants it;
• the approval expires after 72 hours, on the server. Nobody can stretch that window from the app.

If a source is missing, expired, or contradicts another source, if a connection to an HR system is unavailable, or if your question isn't clear enough, you won't get an invented answer. You'll get a clarifying question, or the case goes to a person. A missing leave balance is never shown as zero.

Reporting an answer that's wrong

If an answer looks wrong, report it in FixControl:

1. Open the project and choose HR governance. This link only appears on projects where HR is switched on.
2. Find the answer under Answers.
3. Type why the answer is incorrect (at least a few words) and click Report incorrect.

Reporting is only possible on that page, not in Teams or Slack. You can report your own answers; HR staff who handle questions can do the same for the answers they are allowed to see. An answer that is already under review, replaced or withdrawn cannot be reported again.

Reporting doesn't erase anything. The answer goes under review, the original is kept exactly as it was, and a correction task is created for HR. After that, one of two things happens:

• A correction is published. The corrected answer is a new version, linked to the original. The original stays, marked as superseded, so it's always clear what you were told and what replaced it.
• The answer is withdrawn. When there's no safe replacement, the original is marked as withdrawn rather than quietly rewritten. You're told it no longer stands.

Either way there is no version of history in which the first answer never existed.

What gets recorded

FixControl records that a decision happened, not what was said. The record holds the project, identifiers for the case and the answer, the classification, which source versions were used, the policy decision, and the reason code behind it.

Deliberately not in that record: the text of your question, case summaries, medical content, and the body of the answer. Those stay inside the HR tables, behind the same permissions. That way an auditor or an HR manager can check that the process ran correctly without reopening what you asked.

Related pages

• Approvals — how approval gates work elsewhere in FixControl.
• HR governance — the administrator's view: roles, permissions, connections.