Skip to main content
Contact

AI how-to guides People and HR

How to answer HR policy questions with AI

Short answer

Index your handbook, country addenda and role policies so an AI assistant answers employees in plain language, cites the exact article and version, and picks the variant that applies to them. When there is no source or the case needs nuance, it hands over to HR with the full conversation. Start with the most asked topics in one country, then extend.

Updated: · Thinkia

What AI does

  • Answers questions about leave, expenses, benefits or working time in natural language, in the employee's language.
  • Cites the exact policy article, version and jurisdiction behind each answer, and declines when there is no source.
  • Picks the country, region or role variant that applies, using the employee's profile.
  • Hands over to HR with the conversation, the suggested policy and its confidence.
  • Shows policy owners which content is asked about most and which is flagged as outdated.

What stays with people

  • Writing, approving and updating the policies themselves.
  • Personal and sensitive cases: health, conflicts, disciplinary matters, exceptions.
  • Any decision about a person: pay, promotion, performance or termination.
  • Deciding which topics the assistant answers and when it must hand over.

How to put it in place

  1. Find the questions that fill the HR queue

    Group recent HR tickets and emails by topic and pick the most repeated ones. Check that each has a written policy behind it: the assistant can only answer what is documented.

  2. Put the policies in order

    Gather the handbook, country addenda and role policies, with version, owner and review date. Resolve contradictions and mark which variant applies to whom, because the assistant will repeat whatever the documents say.

  3. Connect the content and the employee profile

    Index the policies so answers come with citations, and connect the HRIS so the assistant knows the employee's country, entity and role. Give it read access only, and only to what it needs.

  4. Set the grounding and escalation rules

    Decide that no source means no answer, which topics always go to a person and what context HR receives with each handoff. Write it down as policy with HR and, where relevant, employee representatives.

  5. Pilot in one country, then extend

    Open the assistant to one country or business unit, review samples of answers with HR and fix content gaps. Add countries, languages and topics when quality holds, and keep policy owners alerted to outdated content.

What you need

  • Current, approved policies with an owner and version for each.
  • Access to the HRIS for the employee's country, entity and role.
  • The channel where employees already ask, such as the intranet or Teams.
  • Clear escalation routes into the HR team or ticketing tool.
  • An HR owner who decides scope, reviews answers and keeps content fresh.

When it is not worth it

  • Small organisations where HR answers a handful of questions a week.
  • Policies that are mostly unwritten or contradictory: write and agree them first.
  • Topics that almost always need a personal conversation.

Risks and how to control them

  • Invented benefits or wrong entitlements: answer only from cited sources and say so when there is none.
  • The global answer given where a local rule applies: tie every policy to its jurisdiction and resolve the variant from the employee profile.
  • Employee personal data: apply data protection rules, limit what the assistant can see and log queries for review.
  • Scope creep into decisions about people: under the EU AI Act, AI used for decisions on promotion, termination, task allocation or performance evaluation may be high risk, so keep the assistant to information and review this with your legal team.

How Thinkia does it

We start by mapping your policy structure, jurisdictional variants, HRIS and the query types that drive HR ticket volume. Then we define the grounding policy, escalation rules, language scope and who owns each piece of content, and pilot in one country or business unit.

It runs on Enterprise Knowledge AI: answers cite the exact article and version, with no answer if there is no source, every query and citation is logged for HR and compliance, and HR owns both the content and the escalation policy.

The Thinkia solution behind this guide

Thinkia products involved

Frequently asked questions

Can a chatbot answer HR questions reliably?

Yes, if it answers only from your approved policies, cites the source and hands over when there is none. A general chatbot without access to your handbook will give generic or invented answers.

How does it handle different countries?

Each policy is tagged with the jurisdiction and roles it applies to, and the assistant uses the employee's profile from the HRIS to pick the right variant. Local addenda need to be documented for this to work.

Is an HR assistant high risk under the EU AI Act?

Answering policy questions is not, in itself, one of the employment uses the AI Act lists as high risk. Using AI to decide or evaluate people at work can be. Keep the assistant to information and check your case with your legal team.

What happens with personal or sensitive questions?

They should go to a person. Define the topics that always escalate, such as health, conflicts or disciplinary matters, and make sure HR receives the context so the employee does not have to repeat it.

Can't we just use Copilot or ChatGPT?

General assistants can search documents, but they do not resolve country variants, enforce answer-only-with-source or route to HR on their own. Those rules have to be designed and tested for HR content.

Related guides

Related decisions

Key terms

Sources

Want to do this in your organisation? Let's talk.

Talk to an AI Expert