By the end of this guide you will have an AI HR assistant that answers employee questions about policies, benefits, onboarding and requests, using documents your HR team has approved, and that passes anything sensitive to a person. The job is done by an AI assistant or agent platform that reads your policies, connects to the systems where employees already work, and is tested and monitored by HR before and after launch.
To build an AI HR assistant for employees, the hard part is the source material and the boundaries, not the software. This guide covers preparing policies, choosing between enterprise agent platforms and lighter assistants, setting escalation rules and measuring results. It compares six tools from our Automation and agents category.
Key takeaways
- Start with a short list of high-volume questions, write or tidy the policies that answer them, then let the assistant answer only from those approved sources and escalate everything else.
- The main decision is scale: Moveworks and Kore.ai are enterprise platforms quoted through sales, while Saifa AI has a free plan and Scribe and Opus Training help build the knowledge behind an assistant.
- Keep people in the loop for sensitive topics such as health, grievances, pay disputes and legal questions. An assistant is not a substitute for HR judgement or legal advice.
- Tell employees what the assistant is, what it can see and who to contact when it cannot help.
What you need before you start
- Approved HR content. Handbooks, policies, benefits guides and onboarding checklists that HR is happy for an assistant to quote.
- A list of common questions. Pull them from your helpdesk, inbox and chat channels.
- Where employees work. Slack, Microsoft Teams, an intranet or a portal, so you can meet people there.
- An owner. One person in HR who approves content and reviews the assistant's answers.
- IT and security support. For connecting systems and reviewing data access.
- Time. A pilot with a handful of policies can run in a few weeks. An enterprise rollout takes an implementation project.
- Difficulty. Moderate for lighter tools, high for enterprise deployments.
Which kind of AI tool fits this job
An enterprise employee assistant. Moveworks gives employees one place to find information and get tasks done across work apps, with built-in use cases for IT, HR, finance and other departments. It connects to systems such as Slack, Teams, Workday, ServiceNow, Okta and Microsoft 365, and employees reach it through chat, web, intranet, portals and mobile in more than 100 languages. It is aimed at large organisations with an implementation project.
An agent platform you configure. Kore.ai is an enterprise platform for building agents for customer service, employee support and business processes, including HR and IT assistants. It offers an Agent Studio, workflows for multi-step tasks, testing before deployment and tracing and audit of every session. Kore.ai says it is built for enterprise deployments rather than quick self-serve use.
A lighter assistant trained on your documents. Saifa AI learns from a company's own documents, products, policies and procedures, answers internal questions, helps onboard new hires and uses role-based access to control what staff see. It has a free plan without a credit card.
Tools that build the knowledge and the training. Scribe captures work as step-by-step guides, and its Sidekick answers employees' repeat questions from that knowledge. Opus Training delivers short mobile lessons for frontline teams and includes Ask Opus, an AI search across training content. Thesmia.ai helps HR teams write internal communications, which is useful for announcing the assistant and explaining how to use it.
Step 1: Pick the questions the assistant should answer
You finish with a short, ranked list rather than an open brief.
- Collect the questions employees ask most: leave, benefits, payslips, onboarding, equipment, policies.
- Group them into topics and count roughly how often each comes up.
- Choose a first set that has clear, stable answers in your documents.
- Write down what is out of scope for now, such as disciplinary matters, medical details and individual pay.
Check it worked by confirming the first set could be answered from your existing documents without anyone's opinion.
Step 2: Prepare and approve the source content
The assistant is only as good as the documents behind it.
- Update the policies and handbooks that cover your topics, removing outdated versions.
- Make each policy easy to find and clearly dated, with an owner.
- If you need step-by-step content, create it. Scribe turns captured work into guides with annotated screenshots, and its Sidekick answers repeat questions from that knowledge.
- Get HR and legal sign-off on any answer that describes entitlements or obligations.
Check it worked by asking a new starter to answer five questions using only your documents.
Step 3: Choose a platform and check what it needs
Match the tool to your size, systems and skills.
- For a large organisation with many systems, ask Moveworks or Kore.ai for a demo. Both quote prices through sales, and Kore.ai's Automation AI is billed per 15-minute conversation session.
- For a smaller team wanting to try the idea cheaply, look at Saifa AI's free plan: 100,000 tokens per month, 1 team member, a 10 GB data source limit and 1 messaging channel.
- For frontline teams, consider Opus Training, which lists a mobile app, auto-translation into 100+ languages and Ask Opus search.
- Confirm where each tool stores data and what employees' questions are used for.
- Use the comparison table below to see pricing and limits side by side.
Check it worked by confirming the tool connects to the channels and HR systems you use.
Step 4: Connect the sources and set permissions
Decide what the assistant can read before it reads anything.
- Connect the approved documents and, if needed, HR systems. Moveworks lists Workday, ServiceNow and Okta among its connections.
- Use role-based access, so employees only see what suits their role. Saifa AI lists role-based access control.
- Keep payroll, health and disciplinary records out of scope unless your security team has reviewed the connection.
- Record who approved each connection and when.
Check it worked by testing with a standard employee account and confirming it cannot see restricted information.
Step 5: Write the instructions and the escalation rules
This is where you protect employees and the company.
- Instruct the assistant to answer only from approved sources and to say so when it does not know.
- Add rules to hand over to a person for sensitive topics, such as harassment, health, grievances, pay disputes and legal questions.
- Provide a clear contact route, for example an HR inbox or ticket form.
- Decide on a tone and a short disclosure line so people know they are talking to an AI.
- Use Kore.ai's guardrail and testing features, or the equivalent in your tool, to check these rules before launch.
Check it worked by asking five sensitive questions and confirming each is handed to a person.
Step 6: Test, pilot and announce
Test with real questions before everyone has access.
- Run your list of common questions through the assistant and have HR mark each answer correct, incomplete or wrong.
- Fix source documents rather than just rewording prompts, since the cause is often a vague policy.
- Pilot with a small group, then collect feedback.
- Announce it clearly. Thesmia.ai writes comms plans, employee messages and manager messages for HR teams, which can help with the launch.
Check it worked by reaching a pilot where HR reviewers mark nearly every answer correct and every sensitive case escalated.
Step 7: Monitor, improve and retire outdated answers
Policies change, so the assistant needs an owner.
- Review a sample of conversations regularly, within your privacy rules.
- Track the questions that get escalated or go unanswered and fix the content.
- Update the sources whenever a policy changes.
- Watch usage against your plan. Saifa AI's plans differ by monthly tokens, and Kore.ai's cost depends on conversation volume.
Check it worked by seeing fewer repeat questions reaching HR and no outdated answers in the sample.
Tools for this job compared
| Tool | Good fit for | Pricing | Stated limits |
|---|---|---|---|
| Moveworks | Large organisations with IT and HR ticket volume (web and mobile) | Contact sales | Pricing quoted per customer; aimed at large organisations with an implementation project |
| Kore.ai | Enterprise teams building HR and IT assistants (web) | Contact sales | Pricing not published; Automation AI billed per 15-minute conversation session; built for enterprise |
| Saifa AI | Small teams answering questions from company documents | Freemium | Free plan: 100,000 tokens per month, 1 team member, 10 GB data, 1 channel |
| Scribe | Turning processes into guides that an assistant can answer from | Freemium | Desktop application is Pro and Enterprise only; plans apply to the whole team; dollar prices not published |
| Opus Training | Frontline teams across many locations (iOS, Android, web) | Contact sales | Prices not published and typically per location; single sign-on only on Enterprise |
| Thesmia.ai | HR teams drafting internal communications (web) | Freemium, paid from £30/mo (checked 24 Oct 2026) | Free plan has no chat memory and limited messages; output is in UK English |
Prices and plan limits change, so confirm the current terms on each vendor's site before you commit.
Common mistakes to avoid
- Feeding the assistant outdated or conflicting policies. It will confidently repeat what it reads. Fix it by retiring old versions and having a named owner approve every source.
- Letting it handle sensitive cases. Grievances, health issues and pay disputes need human care. Fix it by writing explicit escalation rules and testing them before launch.
- Giving it too much access. Broad connections to HR systems can expose personal data. Fix it by using role-based access, starting with documents only and involving your security team.
- Launching without telling employees. People may not know it is an AI or how their questions are handled. Fix it by announcing the assistant, explaining its limits and giving a human contact.
- Treating it as legal advice. An assistant can misstate entitlements or local rules. Fix it by getting HR and legal review for answers about rights and obligations, and by labelling answers as guidance.
How we put this together
This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.
Frequently asked questions
Can I build an AI HR assistant for free?
On a small scale, yes. Saifa AI has a free plan with 100,000 tokens per month, 1 team member and 1 messaging channel, which suits a trial rather than a company-wide rollout. Enterprise platforms such as Moveworks and Kore.ai quote prices through sales.
What can an AI HR assistant answer?
It can answer questions that your approved documents cover, such as policies, onboarding steps and benefits guidance. Moveworks lists built-in use cases for IT, HR and other departments, including password resets and software requests. Anything outside your sources should go to a person.
Is an AI HR assistant safe for employee data?
It can be, if you control access and review the vendor's terms. Saifa AI states that customer data is never used to train other companies' AI and uses role-based access, and Moveworks lists governance and security controls. Involve your security and legal teams before connecting personal records.
How long does it take to set up?
The listings don't give pilot timelines, and most of the work in a small pilot is preparing content. Enterprise deployments involve an implementation project, and Moveworks lists professional services and support packages. Allow time for testing and for a pilot group.
Does it replace HR staff?
No. It deals with repeat questions and routine requests so HR has more time for cases that need judgement. Keep people responsible for sensitive conversations, policy decisions and anything involving rights or obligations.
Compare more options in our Automation and agents category.
