Guides/Team & access

I want customer questions answered automatically, without invented answers

The AI Operator: an extra member of staff you teach yourself, who hands over when unsure. 8 min read

The AI Operator is an extra member of staff in your inbox. It answers customer questions, opening hours, prices, how to book, from the knowledge you taught it, and its most important rule is honesty: it is designed not to answer beyond the knowledge you approved. When unsure, the conversation is escalated to your team with a note, rather than answered with an invented price, schedule, or piece of medical advice. Every reply is logged and available for you to review.

Price: US$30 a month, including a quota of roughly 1,500 replies. Enable it from the Billing page. The bot is installed automatically once payment is received, switched off until you have filled in its knowledge and turned it on.

Before subscribing, prepare its knowledge, free

The bot only answers from your knowledge cards. Switched on empty, most chats will be handed to your team with a note saying outside its knowledge, and the subscription feels wasted. That is why the Assistant page shows a readiness checklist before you pay:

Both can run alongside business as usual. Nothing is lost by delaying activation until you are ready.

Four ways to teach it

Every route ends in the same place: no knowledge becomes active without your approval. Draft cards are reviewed first, and proposals are approved with one click.

It gets smarter from your team corrections

Every time the bot gives up and your team answers, that answer appears as a knowledge proposal on the Assistant page. One click turns it into a card, and the same question will not be handed over again. Every Monday you receive a summary email: how many chats it answered, how many it handed over, and which questions come up often with no answer yet.

Setting how the bot addresses people

The Assistant page has a field called "How the bot addresses people". Write the forms of address and the set phrases your business uses, and the bot will use them. This is about tone of voice, not content. The facts still come from the knowledge cards.

What makes the biggest difference: state the forms that are forbidden, not only the correct ones. Writing "address parents as Mum and Dad" invites the bot to shorten it to "Dad" when addressing one person. Writing "always Mum/Dad, never Dad alone" closes that gap.

If the list of phrases is long, add "do not use them all in one reply". Without that the bot tends to cram everything in and the result sounds forced.

Sending an image or a PDF, not just text

Every knowledge card can carry one file: a JPG or PNG up to 2 MB, or a PDF up to 5 MB. If the bot answers using that card and the file helps, the file is sent along with the answer.

Images suit a price list meant to be taken in at a glance. PDFs suit brochures, registration forms, or terms of service, things meant to be saved and read again. For a PDF, the filename you upload is what the recipient sees in WhatsApp, so give it a clear name such as Registration-Form.pdf.

One file per card. If you have three price packages, make three cards, each with its own file. Besides being the limit, this makes answers more precise: someone asking "what is the sealer package?" receives only the sealer image, while someone asking "what does it cost?" receives all three.

Quota and cost, with no way to overrun

What the quota is spent on, with nothing hidden

The quota only goes down when the AI model is actually called, and every call is recorded: input tokens plus output tokens, times the rate per 1,000 tokens. The Assistant page shows a table, "What the quota was spent on this month", that sorts everything into plain activities:

Does not use the quota: bot off, quota exhausted, bot busy, or an outage (recorded as 0 and still visible in the table); every non-AI auto-reply, such as greetings, menus, reminders, quiet-lead follow-ups, payment confirmation, and the catalogue; classifying customer feedback (covered by WACO); and WhatsApp/Meta message fees, which never pass through WACO.

You always hold the controls: the Turn off bot button takes effect immediately, and a human agent can take over a conversation at any moment. The bot only answers conversations your team has not touched.

Several channels? Several branches? One assistant is usually enough

One assistant per product is the recommendation, and it can stand on any number of your channels at once, WhatsApp, website live chat, Instagram, at no extra cost, on one shared quota. A new branch, a new number, or a new website selling the same product, all branches of one beauty clinic for instance, does not need a new assistant. Just tick the channel on the "Which channels does the bot stand on?" card on the Assistant page.

A second assistant only makes sense when you have a different product and want the knowledge kept apart so it does not answer from the wrong room, a beauty clinic and a circumcision clinic for example. Add one from the Assistant page and it gets its own knowledge, quota, and channels, at US$30 a month. The page states clearly which assistant you are working on, and you switch between them with the tabs above.

Have your own application outside WACO? The same brain is available through the developer API.

Prefer to manage the bot by chatting? Through WACO MCP, Claude or ChatGPT can decide on suggestions, teach new facts, distil a website into knowledge, and test the bot's answers for you — without opening this page.

Still stuck after following this guide? Contact us from the dashboard — attach a screenshot if you have one, it speeds everything up.