Support chatbot setup: scenarios, handoff, reports

I set up a support chatbot that takes over the questions your team retypes every day, and steps aside the moment a person needs a person. The work starts with your own history rather than with a blank scenario editor: I read through the requests you already receive, group them into the handful of topics that genuinely repeat, and only then draw the answer tree. Buttons are written in the words your customers use, not in internal jargon, so nobody has to guess which branch hides their question.

The bot introduces itself as a bot in its very first line. No invented name, no portrait photo, no pretending to be a member of staff — one moment of exposure costs more trust than the whole widget saves. A "talk to a person" button sits on the first screen and stays visible for the entire conversation; it is never something the visitor has to earn after three failed attempts. Outside working hours the same button says plainly that nobody is there right now, collects a contact and a short description, files it into your existing ticket system and names the window in which an answer will arrive.

Health, money and legal questions are kept out of the scripts entirely. Those conversations go straight to a qualified specialist with the collected context attached, because a scripted reply on such topics costs far more than it saves.

What the work includes:
— analysis of your real incoming questions and a map of repeating scenarios;
— an answer tree with plain buttons, no dead ends, and sane fallback wording;
— handoff to a live operator on demand, on confusion, and by schedule;
— contact capture with a promised reply window, wired into your ticket system;
— a greeting that appears once and does not chase the visitor around the page;
— a report showing which questions the bot closes and which leak to people;
— one round of scenario rework after the first month of real traffic.