HR knowledge assistants: a practical design guide

Design a policy assistant that uses approved sources, respects access and knows when to hand a question to a person. This practical guide to HR knowledge assistant brings together a step-by-step approach, illustrative examples, a reusable worksheet and answers to common questions. Start with the section closest to your current challenge, then use the working session to turn the guidance into a clear next action.

Define the assistant’s job

An HR knowledge assistant should help someone find and understand approved information. Start with a bounded set of questions, such as locating the correct process for a routine request. Do not begin by promising that it can answer every employee question.

Define the audience, supported locations and topics that require a human response. Personal cases, disputed decisions and sensitive matters may need a specialist even when a general policy is available. Make the handoff visible and usable.

Prepare authoritative sources

Create an inventory of approved documents with owners, effective dates, audiences and superseded versions. Remove duplicate or outdated copies from the retrieval set. A system that finds an obsolete policy accurately still gives the wrong answer for current use.

Attach useful metadata such as location, worker group and topic. Test whether the assistant can identify the applicable source. When that context is missing, it should ask a clarifying question or provide a contact rather than selecting a policy silently.

Enforce access outside the prompt

The application must restrict what each user can retrieve according to approved permissions. An instruction saying “do not reveal confidential information” is not a substitute for access controls.

Test both retrieval and the final answer using users with different permissions. Check shared caches, conversation history, exports and source links. Treat instructions found inside documents as untrusted content; they must not override the application’s rules or grant new access.

Worked example: two leave policies

Illustrative scenario: the assistant has policies for two locations. An employee asks how to request leave without stating their location. The assistant asks for the relevant context or directs them to the appropriate HR contact.

Once the applicable policy is identified, the answer links to the section and states its scope. If the document is expired or conflicting, the assistant escalates rather than combining rules from both policies into an invented answer.

Knowledge-assistant worksheet

• Audience and scope: Who can ask which kinds of questions?
• Sources: Which documents are approved, current and owned?
• Context: What information is needed to select the applicable policy?
• Permissions: How is access enforced and tested?
• Answer standard: What source, uncertainty and next step should be shown?
• Handoff: Which questions go to a person, and how?
• Maintenance: Who updates sources and re-runs tests after a change?

Include unanswered questions in evaluation. A system that declines an unsupported answer appropriately may be more useful than one that confidently answers everything.

Test and maintain the service

Build cases involving missing context, conflicting policies, inaccessible sources and instructions embedded in a document. Check source relevance as well as factual wording. Ask real users whether the response helps them complete the task.

Monitor failures with proportionate logging and clear access rules. Do not retain sensitive employee questions indefinitely by default. The AI governance guide helps assign ownership; the NIST AI RMF Playbook offers broader risk-management considerations.

Put the guide into practice

Set aside a working session with the people who own this process and one or two people who experience it. Use a fictional or appropriately authorised case, so the discussion can be specific without sharing unnecessary personal information. The purpose is to leave with a usable decision or document, not just agreement that the topic matters.

Prepare the case

Choose a bounded set of employee questions and identify the approved sources needed to answer them. Record each source’s owner, audience and effective date. Include a question that should go to a person. The exercise should establish what the assistant may answer, not simply collect as many documents as possible.

Write the starting assumptions down before discussing solutions. If the group disagrees on what happened, identify the information needed to resolve that difference rather than building a plan on an untested story.

Work through the decision

Test a question against the correct policy and check every material statement in the answer. Verify that source links open for the intended user and that the cited section actually supports the response. Ask whether the answer helps the employee take the next step without adding rules absent from the source.

Ask each participant to explain the proposed decision in their own words. Differences in interpretation often reveal an unclear criterion, a missing responsibility or an instruction that will be difficult to follow.

Test an exception

Introduce conflicting versions, missing location context and a document containing instructions to reveal other information. Test access controls with different users. The assistant should not combine incompatible policies or follow instructions embedded in retrieved content. Escalation or a clarifying question may be the appropriate result.

Record what changes in this situation and what remains the same. An exception should lead to a clear next step, with an owner, rather than an informal workaround that nobody can explain later.

Agree the handoff

Give the service owner a source-maintenance schedule, regression cases and a clear human handoff. Confirm who handles unanswered questions and how urgent or personal matters are routed. The employee should not be trapped in repeated chatbot exchanges when the information requires an authorised person.

Finish by confirming the owner, the next action and the date when the result will be reviewed. Give the person receiving the work enough context to continue without repeating the whole discussion.

Frequently asked questions

Is an HR chatbot the same as an HR knowledge base?

A knowledge base is a collection of organised information. A chatbot is an interface that may search or generate answers from that information. The quality of the interface depends partly on the sources, but also on retrieval, permissions, context and answer behaviour. Improving the documents and ownership often matters as much as choosing the conversational technology.

How can an assistant avoid using outdated policies?

Maintain authoritative sources with owners, effective dates and clear version handling. Remove superseded material from current retrieval and test questions after updates. The application should use relevant metadata and detect conflicts where possible. A prompt telling the model to prefer current information is not enough if the retrieval set contains ambiguous or incorrectly labelled documents.

Should the assistant answer individual employment questions?

Only within a deliberately reviewed scope. Personal cases can depend on facts, local rules or decisions that a general policy does not resolve. Provide a human route for sensitive, disputed or consequential matters. The assistant can help locate the appropriate contact or source, but it should not invent a determination about an employee’s rights or circumstances from incomplete context.

Review the first cycle

Review unanswered questions, incorrect source selection and failed handoffs. Use proportionate logging and protect sensitive employee queries. Check that source updates trigger the relevant tests and that retired policies stop appearing. If users cannot understand the answer’s scope or reach a person, improve the service design before expanding coverage.

Keep a brief record of what was tried, what participants found useful and what needs to change. Compare the result with the original problem rather than judging success only by completion. If the process created extra work without improving clarity, quality or support, simplify it and test again. Share the agreed change with the people who will use it, and name the person responsible for keeping the guidance current.

About New Dynamics

New Dynamics connects goals, feedback, recognition and reviews around the way organisations work. This guide is published by the New Dynamics Editorial Team as part of our practical library for HR leaders, managers and People teams.

Use the examples and worksheets to structure your own discussions and adapt them to your organisation. Illustrative scenarios are not customer case studies. Policy and employment guidance needs appropriate local review before adoption.

For questions about this guide, corrections or a conversation about your performance management process, email contact@new-dynamics.com. Explore the complete guide library for related resources.