AI can write a reasonable reply in seconds. The temptation is to let it send that reply automatically. We don’t, and we recommend you don’t either.
What can go wrong
- It states a price, a date or a promise you never agreed to.
- It misreads the tone of an upset customer and replies cheerfully.
- It answers from out-of-date information.
- It sends something to the wrong person.
Each of these is rare, but each one happens under your name, and it only takes one to cost a customer’s trust.
What an approval step looks like
- The AI drafts the reply, follow-up or post.
- The draft waits in a queue, with the sources it used shown next to it.
- A person reads it, edits if needed, and approves it.
- Only then is it sent.
Approving a good draft takes a few seconds. You keep most of the time savings, and nothing leaves the building without a person seeing it.
Showing sources builds trust
When an AI answer shows where each fact came from, like a booking, an invoice or a note, you can check it in a glance. Answers without sources ask you to trust them blindly.
Where it’s safe to automate fully
Fixed messages you’ve written yourself, like an appointment reminder or a missed call text, can go out automatically, because nothing in them is generated on the spot. The approval step is for anything AI writes fresh.
This is how our Command Center’s assistant works: it answers with sources and drafts follow-ups that wait for approval. Nothing is sent without you.