Intents & Routing — How the AI Agent Decides Which Workflow to Run
Written By Anand Raj
Last updated 24 days ago
Intents are how the Brain decides which node to move the conversation into. Every action node (Custom Task, Note Taking, Call Transfer, End Call, etc.) is connected back to the Brain through an Intent — a short, plain-English description of when the agent should jump into that node.
Note: If you are a direct business user and are not familiar with technical workflow configuration, do not make changes to the workflow. Contact your agency or raise a support ticket so the team can investigate and make the necessary changes.
How Intents work
Each connection from the Brain to a node carries a Caller Intent with two parts:
When the caller expresses a desire to… — a short description of the situation that should trigger this node (e.g. "book, schedule or arrange an appointment", "cancel an existing appointment", "complain about billing", "talk to a human").
Intent signals — example phrases callers actually say (e.g. "I want to book an appointment", "Can I schedule a visit?", "I'd like a demo").
The agent listens to the live conversation, matches it against these intents, and routes the call into the matching node — Book Appointment, Cancel Appointment, Note Taking, Call Transfer, End Call, and so on.
Common intents you'll configure
Book / Update / Cancel Appointment — route to the matching Custom Task node.
Note Taking — when the caller wants to leave a message, lead detail, or complaint.
Call Transfer — when the caller asks for a human, manager, or escalation.
Spam / Robocall — when the call clearly isn't a real customer, route to End Call so the agent hangs up cleanly.
Billing & Pricing or Knowledge — for general business questions answered from the Knowledge base.
If the agent picks the wrong workflow
If the agent keeps routing to the wrong node — for example, going into Note Taking when the caller clearly wants to book — that almost always means the intents are not configured cleanly:
Two intents overlap and confuse the agent.
Intent signals are too vague, too short, or missing real caller phrasing.
An intent is missing entirely, so the agent falls back to the closest match.
Fix it by tightening the "When the caller expresses a desire to…" line, removing overlap between nodes, and adding more real-world Intent signals for each node.
Important: Intents and prompts look like plain English, but they're actually LLM prompts — small wording changes can change routing behavior. If you don't have experience with LLMs and how prompts work, contact your agency before changing intents or prompts on a live agent.
Frequently Asked Questions
Where do I configure intents?
Where do I configure intents?
Open the Agent Builder from Identity, click the connection (or the target node) coming out of the Brain, and edit the Caller Intent panel on the right.
How many intent signals should I add?
How many intent signals should I add?
Add as many real caller phrasings as you can — 5 to 10 per node is a good starting point. The more natural examples you give, the more accurate the routing.
Can two nodes share the same intent?
Can two nodes share the same intent?
What happens if no intent matches?
What happens if no intent matches?
The agent falls back to the Knowledge node and tries to answer from the knowledge base, or stays in the general conversation until a clearer signal comes through.
Should I edit intents myself?
Should I edit intents myself?
If you're comfortable writing prompts for LLMs, yes. If not, ask your agency — small wording changes in an intent can change how the whole call is routed.
More in Identity & Conversation Flow:
Was this helpful?
Still need help? Ask the team