Conversation Flow Builder Guide

Design effective conversation flows for AI voice agents with visual dialog trees.

How to Use

Step 1: Define Entry Points

• Main greeting/welcome
• Different call types
• Transfer destinations
• Error handlers

Step 2: Build Dialog Tree

• Add conversation nodes
• Connect with transitions
• Define conditions
• Set responses

Step 3: Add Logic

• Intent detection rules
• Entity extraction
• Conditional branching
• Fallback handling

Flow Components

Node Types

Transition Types

• Intent-based: User says specific thing
• Entity-based: Based on extracted value
• Condition-based: Business logic
• Default: Fallback path

Sample Flow: Order Status

Welcome
  └── "What can I help with today?"
        ├── [Order Status Intent]
        │     └── Ask Order Number
        │           └── Validate Order
        │                 ├── [Valid] → Get Status → Respond
        │                 └── [Invalid] → Error → Retry/Transfer
        ├── [Returns Intent]
        │     └── Returns Flow...
        └── [Other/Unknown]
              └── Clarify → Retry/Transfer

Best Practices

Conversation Design

1. Keep it natural

   • Use conversational language
   • Allow interruptions
   • Handle "um" and "uh"

2. Be efficient

   • Don't ask unnecessary questions
   • Remember context
   • Confirm important details

3. Handle errors gracefully

   • Don't repeat the same error message
   • Offer alternatives
   • Know when to transfer

Intent Design

Good Intents:

• Specific and distinct
• Have clear training examples
• Cover common variations

Bad Intents:

• Overlapping with others
• Too broad or generic
• Missing common phrases

Error Handling

No Input Timeout

1st timeout: "I didn't catch that. Could you repeat?"
2nd timeout: "I'm still here. What would you like to do?"
3rd timeout: "Let me connect you with someone who can help."

No Match

1st no-match: "I didn't quite understand. Could you rephrase?"
2nd no-match: "Let me give you some options: [list options]"
3rd no-match: "I'll connect you with an agent for help."

Technical Errors

"I'm having trouble looking that up. Let me connect you with someone who can help directly."

Advanced Features

Context Management

• Store conversation state
• Reference previous answers
• Remember user preferences

Slot Filling

Required slots:
- Order number: "What's your order number?"
- Verification: "And can you confirm the email on the account?"

Collected: Proceed to lookup
Missing: Re-prompt for slot

Confirmation Patterns

Implicit: "Checking order 12345..." Explicit: "Just to confirm, that's order 12345, correct?"

Testing Checklist

• Happy path works smoothly
• Error paths handle gracefully
• Timeouts behave correctly
• Edge cases covered
• Transfers work properly
• Context is maintained
• Natural conversation flow

Related Tools

• Persona Builder - Define agent personality
• Call Script Generator - Script templates
• Quality Scorecard - Evaluate conversations