An AI that is confidently wrong in front of a customer costs more than the automation ever saved. Four things go into every build, and they are not optional extras.
It works from your material, not its memory
Your real prices, hours, service area and policies are given to it at the moment it answers, rather than trusted to whatever it absorbed in training. Change a price on your site and the answer changes. That is also how the assistant on this page works.
It has written permissions
Before anything is built we agree what it may do on its own, what it must draft for you to approve, and what it must refuse and escalate. That list goes in the quote. Anything where being wrong is expensive, money, promises, legal, medical, stays on the approve side.
It gets tested against the awkward questions
Not the demo questions. The ones your customers actually ask, including the ones designed to trip it up and the ones where the honest answer is no. You see those results before it goes anywhere near a customer.
You can switch it off
Every build has an off switch you control, and a way to see what it did and why. If it starts behaving oddly at 9pm on a Saturday you are not waiting for me to wake up.
