A customer opens the chat with a question or a problem. The agent finds out what is actually wrong, and asks for the email on the account only when the question touches the account. The actor handling this part holds no tools at all, so it cannot answer from thin air — it can only listen. Then it reads four things: your customers, your approved answers, your troubleshooting guides, and any ticket already open for this person. Answers come from your Answers tab and nowhere else. If the question is not covered there, it says so — "I don't have that written down, let me get someone to confirm it" — and raises a ticket. It will not improvise, and it will not dress a guess up as a policy. Answers you have marked as applying to one plan are never offered to customers on another. Faults are walked through your own troubleshooting steps, one or two at a time, in the agent's own words rather than pasted as a list. Each guide carries your own escalate_if condition: the point at which the problem stops being self-service. The moment that is true, the agent stops suggesting things and fetches a person. A customer who has already tried four things does not want a fifth guess. What it will never do is as important as what it does. It never asks for a password, a card number, a security code or a one-time code — not to verify anyone, not for any reason. And it cannot change anything on an account: no cancelling, upgrading, refunding or resetting. When someone asks for a change, it says plainly that it has been passed to your team, and records exactly what was asked. "I've asked the team to cancel it" is honest; "I've cancelled it" is a lie nobody discovers until the next bill. When it hands over, the ticket carries the one field that saves the most time: everything the customer has already tried. Your colleague picks up where the customer left off instead of asking them to check their spam folder again.