HR Policy Chatbot: Policy and Contract Answers in Seconds, People Decide

HR teams answer the same questions every week. How many vacation days do I have? Does our dental plan cover orthodontics? What is the parental leave policy in this province? Each answer needs someone to find the right document, check the right version and reply carefully.

Legal and HR staff have a second problem. Which contracts contain a non-compete, a specific notice period or an unusual bonus clause? To find out, they open files one by one. That is slow, easy to get wrong, and pulls skilled people away from work that needs judgement.

What this assistant does

The assistant uses retrieval-augmented generation (RAG): it searches your approved documents, then has a language model answer using only what it found. It serves two audiences.

For employees

• Answers questions about the employee handbook, benefits, leave rules, expenses and workplace policies.
• Gives the answer for the employee's own location and group, and says which version it used.
• Links to the exact handbook section or benefits page.
• Hands off to HR for anything personal, sensitive or not covered.

For HR and legal teams

• Finds clauses across employment contracts, templates and vendor agreements, such as notice periods or confidentiality terms.
• Lists which documents contain a given clause, with a quote and a link to each.
• Compares a clause against your current template so differences are easy to review.

The principle is simple: it informs, it does not decide. It never approves leave, judges eligibility or interprets the law. Those decisions stay with people.

How it works, step by step

1. Ingest documents. We load the handbook, policy documents and benefits guides from your HR system, SharePoint or Drive. Contracts go into a separate, restricted collection. Each document is tagged with location, employee group, effective date and owner.
2. Split and index. Policies are split by section and contracts by clause, keeping headings and numbering. Each piece is stored as an embedding, a numeric fingerprint of its meaning, in a vector database.
3. Retrieve for each question. The system checks who is asking, their role and location, then searches only the documents that apply to them.
4. Answer with citations. The model answers from the retrieved sections and quotes or links them. It adds a reminder to confirm personal cases with HR.
5. Log and improve. Every question, retrieved source and answer is logged in an audit trail. HR reviews unanswered and flagged questions to improve policies and wording.

Tools we use

• Backend: Python with FastAPI, with single sign-on so the assistant knows each user's role and location.
• Vector database: pgvector or Qdrant, with separate collections for policies and contracts.
• Language model: OpenAI or Anthropic Claude under business terms, or a private open-source model for contracts and other sensitive files.
• Interfaces: Slack or Microsoft Teams for employees, and a secure web app for HR and legal clause search.

Check each vendor's current pricing, data-processing terms and retention settings before you decide. These change, and they matter for employee data.

What you need to get started

• The current handbook, policies and benefits guides, with a clear owner for each.
• A list of locations and employee groups with different rules.
• Agreement on who may search contracts, and which contracts are in scope.
• Fifty or more real questions HR has received, with the correct answers.
• A privacy review of the design with your counsel or privacy lead.

Typical scope and timeline

A first version is typically 2 to 4 weeks, depending on the number of documents, locations and integrations. This is an estimate. An employee policy assistant in Teams for one country sits at the short end. Adding multi-country rules and a restricted contract search pushes it toward the long end, or into a second phase.

A sensible order is:

• Phase 1: employee policy questions for one location, in the chat tool staff already use, with HR reviewing answers during a pilot.
• Phase 2: more locations and employee groups, with location tagging tested against tricky questions.
• Phase 3: restricted contract and clause search for HR and legal, with its own access rules and audit reports.

Starting with policies builds trust and surfaces document problems early. Contracts carry more risk, so they benefit from that experience.

How we keep answers accurate

• An evaluation set of real questions. HR supplies questions and correct answers, including tricky location-specific ones. Every change is tested against them.
• Citations on every answer. Employees and HR can check the exact section the answer came from.
• Refusal when unsure. If the policy is silent or unclear, the assistant says so and routes the question to HR.
• Effective dates. Only the current version of each policy is used, unless someone asks about a past period.
• Monitoring. HR sees which questions are asked most, which go unanswered, and which answers get negative feedback.

Risks and how we handle them

Privacy and employee data

Personnel files, medical notes and investigations stay out of the index. Logs are kept lean and on a fixed retention period. Privacy laws such as PIPEDA in Canada and GDPR in the UK and EU set rules on employee data. Their exact effect on your setup needs review with your own counsel. Our article on RAG privacy by design explains the engineering side.

Access control and audit trail

Contract search is limited to named HR and legal roles through single sign-on. Permissions are enforced at retrieval time, not by prompt instructions. Every query is recorded so you can review how an answer was produced.

Wrong or outdated answers

A wrong answer about leave or pay can cause real harm. Citations, refusal rules, effective dates and a clear "confirm with HR" message limit that risk. Policy owners are notified when their documents are frequently cited or questioned.

When not to build this

• Your policies are out of date or contradict each other. Update them first.
• Your HR platform already has a policy assistant that covers your needs and data rules.
• You have very few employees and HR questions take minutes a week.
• You want the tool to make decisions about individuals. That is not a safe use of this technology.
• Most questions your HR team gets are about personal situations rather than written policy. A better intake form may help more.

How UnlockLive can help

We build HR and legal assistants with access control, audit trails and evaluation built in from the start. See our RAG development service, and our cybersecurity work for reviewing access and data handling.

The same approach powers a general internal knowledge assistant in Slack or Teams, and pairs well with employee onboarding automation. For contracts that need to stay off public AI services, read private RAG for regulated data. To discuss your policies and scope, book a free 30-minute call.