HR · Process breakdown
Let AI answer handbook questions with the source, route sensitive topics to a person, and use the questions to improve the policies.
An HR chatbot is the most obvious use of AI in HR, and the one most often done badly. The reason an HR chatbot works for policy questions is simple: the answers already exist in the handbook, employees just cannot find them, and HR answers the same twenty questions every week. That is bounded, verifiable work. The reason it goes wrong is just as simple: a chatbot pointed at an outdated handbook, or one that starts answering questions about someone’s health or a conflict. The first thing to fix is the handbook itself: current, owned and complete enough to answer from.
AI interprets the employee’s wording, including typos and local terms, and retrieves the relevant section from the approved handbook rather than from general knowledge.
The reply quotes or closely follows the policy text, links to the section, and states when a question is not covered instead of improvising an answer.
When a question sits between two policies or depends on a contract type, AI drafts a reply with the relevant sections and an HR advisor checks and sends it.
Questions that touch health, harassment, performance, pay disputes or grievances are routed to a named HR partner immediately, with a short human-written acknowledgement.
Anything about an individual’s situation is a conversation between the employee and HR or their manager. The chatbot can only offer to connect them.
AI clusters unanswered and repeated questions and proposes clearer policy text or a new FAQ entry. HR approves and publishes; the next version becomes the new source.
Start with three policy areas, such as leave, expenses and working hours, for one department, with HR’s normal channel still open. Count answers given with a citation, escalations, answers HR had to correct, and questions that repeat. Extend to more policies only when corrections are rare.
Pointing a chatbot at an outdated handbook. It will answer confidently from the wrong policy, and every wrong answer becomes an HR case.
Anything about a specific person’s situation or wellbeing: illness, harassment, conflict with a manager, performance concerns, disciplinary matters, pay disputes and questions with legal weight such as dismissal or discrimination. The chatbot should recognise these topics, stop, and connect the employee with a named HR contact. Being conservative here costs little; getting it wrong costs trust.
HR helpdesk automation adds ticketing, routing, SLAs and reporting on top of the answering. If your volume is a few hundred questions a month, a well-built assistant on the handbook with a clear escalation path may be enough. If HR already drowns in email threads, ticketing first and the chatbot second is often the better order. Either way, the handbook quality decides the outcome.
Version the handbook and rebuild the assistant’s source on every change, review a sample of answers each week against the policy text, and treat every correction as a signal that the policy or the retrieval needs work. Show the citation in every answer so employees can check for themselves, and make it easy to report a wrong answer.
Illustrative workflow guidance by Arcgent. Each business needs its own assessment. No integration or savings claim has been verified for your systems.
Get an assessment based on your own steps, systems, and constraints.
Audit my process See an example report ↗