Chat Intents What the assistant recognises in a customer message, and what it does about it.

When a customer messages you, something has to decide what the message is about before anything useful can happen. That is an intent.

The transparency centre

The intents screen is deliberately read only. It answers four questions at a glance:

• Is the assistant on
• What does it recognise
• What does it do when it recognises something
• What is it connected to

Nothing is configured here. Every gap it shows deep-links to the surface that owns it, so there is one place to look and one place to change each thing.

That separation is the point. A screen that both reports and configures tends to be trusted for the reporting and quietly wrong about it.

The five things it can do

When an intent matches, one of five responses follows:

Response	What happens
Canned	A fixed reply you wrote
Knowledge	An answer drawn from your knowledge base
Booking	The conversation moves to scheduling
Handoff	A person takes over
Clarify	It asks a question, because it is not sure

Clarify is the one worth valuing. An assistant that guesses produces confident wrong answers to customers. One that asks costs an extra message and gets it right.

The dry-run console

You can type a message and see what the assistant would do with it, without anything being sent, booked or saved.

Use it before turning anything on, and again after changing your knowledge base. It is the only way to find out that a common question falls through the gaps that does not involve a customer finding out first.

When nothing matches

An unmatched message goes to a person. That is the correct failure: the fallback is a human, not a guess.

What it costs

Recognising an intent is a language model reading the message, so it draws on your organisation allowance rather than a person's seat. Nobody on your staff asked for it, and visitor volume has nothing to do with headcount.