
What do I need to prepare before an AI agent goes live?
What do you need from me?
Five things: your real opening hours including the awkward exceptions, the three questions that tell you whether an enquiry is worth your time, who a call should be handed to and when, access to the calendar it should book into, and the list of things it must never say.
Only your business name, your hours, and one channel are strictly mandatory to start. The rest makes the agent better, and the difference is large.
Why does the never-say list matter so much?
Because it is the only part nobody thinks to write down, and it is where the expensive mistakes live. Prices you no longer honour. A service you stopped offering. Any promise about delivery dates. Anything that sounds like medical, legal, or financial advice.
An agent with no never-say list will eventually answer one of those questions helpfully and confidently, and you will find out from a customer.
What about my past messages?
They are the most useful thing you can give us. Real past enquiries and the replies you sent show how you actually talk, which questions genuinely repeat, and where people get confused. An agent trained on those sounds like your business; one written from a template sounds like a template.
A few dozen real threads are worth more than any amount of describing your tone of voice.
What happens if my hours change later?
You send one sentence and we change it. Hours, an opening line, a handoff rule, a price it quotes — those are configuration, not a rebuild, and they are covered by the care plan.
The important thing is telling us. The most common way an agent goes wrong is that something changed in the business and the agent was the last to hear about it.