A manager with a people question may need three things first: the procedure, the form and who to talk to. An AI assistant can supply those, with a checklist of what HR will ask for. The decision and the conversation stay with the manager and HR.
Won’t managers treat its answer as HR’s decision?
They may, unless every answer says what it is. Have it give the procedure and the next step, and state that the decision sits with HR and the manager.
1. Start with one question managers already ask.
A fictional example: “An employee has asked to change their working arrangement. What’s the process, and what should I bring to HR?”
The manager needs the procedure and a list of what to gather. Whether the change is agreed is a separate matter that the assistant leaves alone.
2. Have it do three things.
Find. The procedure that applies, the current forms and the people responsible.
Prepare. A checklist of the information the process already asks for.
Refer. A clear next step when the situation needs HR or another qualified person.
Practical check: Does every answer end with who to contact next?
3. Leave employee records out unless the task needs them.
A guide to process doesn’t need to open anyone’s file. A manager may type a question about a named person, and that alone is no reason to connect the assistant to personal records. Agree the permissions first if record access is ever added.
4. Test with made-up manager scenarios.
Write a set of invented situations and have HR agree the expected guidance for each before the test. Check the procedure it gives, the source it points to, what it does when details are missing, and when it refers the manager on.
Read more: What Can an HR Agent Do? · Answer, Prepare or Act: How Much Should an AI Agent Do? · How Does an HR Chatbot Know Which Policy Applies to Which Employee?
The question is fictional. This isn’t legal advice or a statement of anyone’s entitlement. This is general guidance, not advice for your situation.