Skip to main content

​
Save yourself hours in a few minutes

An agent improves through planning, testing, and iteration. Before building, write down what a successful call should accomplish and what information the agent needs to get there. Start with one important call experience instead of trying to handle every possible request immediately. Give less common requests a clear fallback, such as transferring the caller or collecting information for follow-up.

​
Define your main use case

Think about the most common reasons customers call. Choose one goal, such as scheduling an appointment, qualifying a lead, answering a recurring question, or routing a caller to the right team. Write down:
  • The questions your staff normally ask
  • The information required from the caller
  • The steps that complete the task
  • The decisions that create different paths
  • The limits or situations that require a human
  • What should happen when the task cannot be completed

​
Plan tools and data

List the tools the agent needs before or during the call and the actions that should happen afterward. For each integration, record what information it needs, what result the flow expects back, and what should happen if it fails.
If a flow needs to connect to a service through an API, see the API Request block.

​
Gather your knowledge base

Collect the business information the agent needs to answer customer questions. You can add documents, written information, or website content to the knowledge base. Keep each type of information in the right place:
  • Put shared tone, objectives, and behavioral restrictions in agent guidelines.
  • Put business facts and reference material in the knowledge base.
  • Put ordered conversation steps, decisions, integrations, transfers, and endings in flows.
When connecting a website, include only the pages that contain useful and current information for the agent.

​
Think about call routing

Decide how callers reach the agent and when the flow should transfer them to a person or another call experience. Include routing and fallback behavior in your test plan.

​
Start simple, nail one use case, and scale from there

Build the shortest complete path first and test it. Then add branches and integrations one at a time. This makes it easier to identify whether an issue comes from the conversation design, business knowledge, or an external service.

​
Next steps

1

Choose a starting point

Start from the initial agent setup or duplicate an existing agent that already matches part of your use case.
2

Customize your flow

Add the main conversation path, required information, integrations, and fallback behavior.
3

Test and improve

Test the successful path, missing information, important branches, integration failures, transfers, and early call endings before using the flow with callers.

​
Common questions

You can build the main conversation in the visual flow editor. API-based integrations may require technical details from the service you are connecting.
Add reference information to the knowledge base, then use agent guidelines for behavior that should apply across flows.
Design an explicit fallback path, such as transferring the caller or collecting information for follow-up, and include that path in testing.
Last modified on July 17, 2026