A useful first response. A clear route to a person.

Start with the enquiry, not the voice

The first question is what a good interaction should achieve. Does someone need a basic answer, a message passed to the right person or help finding a next step? A natural-sounding conversation is only useful if it leads somewhere dependable.

• Answering a limited set of approved questions.
• Collecting enough detail to route an enquiry.
• Creating a message or follow-up task for a person.
• Explaining when and how a human can help.

Set clear boundaries

An AI receptionist needs clear limits on what it can say and do. Prices, availability, account information and promises should come from approved information or systems. Anything uncertain should have a safe next step.

For a proposed system, questions worth settling include whether it can make bookings or only request them, what happens when a connected system is unavailable, and which requests should immediately go to a person.

Make the human handover work

Someone should be able to understand who they are speaking with and how to reach a person. No Staff’s planned receptionist will be identified as AI and will not pretend to be Jonno.

A useful handover needs more than a button. Consider who receives the enquiry, what context they need and what the visitor should expect next. Avoid promising a response time that the business cannot support.

Think through privacy before a pilot

Any proposed system needs a clear decision about what information it collects, which providers receive it, how long it is retained and who can access it. Recording or transcribing calls introduces additional questions that need to be resolved before a system goes live.

Test with synthetic examples first. Include unclear requests, requests for a person, missing information and situations the AI should decline to handle. The aim is to understand the limits before relying on the system for real enquiries.