The AI Intent trigger runs an automation when an incoming message means something, not when it contains a specific keyword. You describe the intent in plain language and a classifier decides whether each inbound message matches.
Keyword triggers require you to guess every phrasing a customer might use. "Can I send this back," "I ordered the wrong size," and "I changed my mind" all mean the same thing, and none contain the word "return." AI Intent handles those variants with one rule.
Before you start
- You need admin or automation permissions.
- AI Intent evaluates inbound messages only, not your team's outbound messages.
Create an AI Intent automation
- Go to Automations and select New Automation.
- Enter a Name, such as AI Intent: Return, then select the Inbox.
- Under Trigger, select AI Intent.
- Enter an Intent name. This is a short label used in the automation and in logs, such as Return or Reschedule.
- In Match when the customer..., describe what the message should express. Use at least 10 characters.
- Choose how often the automation runs:
- Once per conversation runs a single time, then runs again after the chat is closed. Best for acknowledgements and routing.
- On every incoming message runs on every matching message, including repeats in the same conversation.
- Optionally expand Examples and exclusions to tune accuracy.
- Add one or more Actions, then save.
Writing a good match description
Write it the way you would explain the intent to a new teammate. Name the goal, then the common reasons behind it. Avoid listing keywords.
Returns
Wants to return a product or send an order back, including because of wrong size, wrong item, damage, or a change of mind.
Reschedule
Wants to move, delay, or change the date or time of an existing appointment, delivery, or service visit.
Cancellation
Wants to cancel an order, appointment, subscription, or service, or is asking how to cancel.
Billing question
Is asking about an invoice, charge, payment method, refund status, or an amount they do not recognize.
Order status
Is asking where an existing order is, when it will arrive, or for tracking information.
Escalation
Expresses frustration or dissatisfaction, or asks to speak to a manager, supervisor, or human.
New lead
Is asking about pricing, availability, or how to get started, and has not purchased yet.
Wrong number
Indicates they are not the intended recipient or do not know who is contacting them.
Tips:
- One intent per automation. If a description covers two goals, split it.
- Describe the customer's goal, not your internal process name.
- Be explicit about scope. "Including because of..." helps the classifier.
Examples and exclusions
Optional, but the fastest way to improve accuracy on edge cases.
- Positive examples (up to 5) are messages that should match. Use real customer phrasing, including lowercase and shorthand.
- Exclusions (up to 5) are near misses that should not match.
Each is limited to 200 characters.
For the Return intent above:
| Positive examples | Exclusions |
|---|---|
| Can I send this back? | Do you accept exchanges? |
| I ordered the wrong size | When will you return my call? |
| I changed my mind about my order | I want to return to my old plan |
The exclusions matter. A keyword rule on "return" fires on "when will you return my call," and an exchange is a different workflow with a different playbook.
Add actions
AI Intent chains into the same actions as any other trigger:
- Assign To a user, so the right person picks up the thread.
- Send a Message using a template, to acknowledge immediately and ask for what you need next, such as an order number.
- Add Tag, to make the intent reportable later.
Combine them, or add multiple triggers to one automation with Add Trigger.
What it looks like in a chat
A customer texts hi can i start a return please. The automation matches, sends the acknowledgement template, assigns the chat, and records the action in the conversation timeline.
Testing your intent
- Save the automation.
- From a test phone, send a message that should match, then one that should not.
- Confirm the automation fired in the conversation timeline.
- If it missed, add the message as a positive example. If it overtriggered, add it as an exclusion.
Run two or three phrasings before relying on the automation in a live inbox.
FAQ
Does this replace keyword triggers? No. Keyword triggers are still right for exact strings such as STOP, a promo code, or a reference number. Use AI Intent when the wording will vary.
Can I use AI Intent and the AI Agent together? Yes. AI Intent handles deterministic work such as routing, tagging, and assignment. The agent handles the conversation.
What happens if two automations match the same message? Both run unless one has the override option checked. Check This automation should override subsequent automations when triggered on the higher-priority automation.
Does the classifier read the whole conversation or just the latest message? This setting can be customized per trigger.