Overview
Call behavior settings control the timing, flow, and interaction dynamics of your voice agent conversations. These settings directly impact caller experience, conversation efficiency, and perceived naturalness of your AI agent.Key Call Behavior Settings:
- Response delay (how quickly agent responds)
- Interruption handling (can users interrupt?)
- User inactivity timeout (handling silent callers)
- Max call duration (preventing runaway calls)
- Background ambient sound
- Thinking voice (natural pauses)
Response Delay
What is Response Delay?
Response delay determines how long the agent waits after detecting that the user has stopped speaking before generating a response. Technical Definition: The minimum number of milliseconds between when the system detects speech has ended and when it begins processing the response.Configuration
Range: 100ms - 1500ms Default: 400ms Increment: 100ms To configure:- Open Call Settings section
- Locate Response Delay slider
- Adjust value (shows in milliseconds)
- Current value displays next to slider
- Save changes
Impact on Conversation Flow
Lower Values (100-300ms): ✅ Advantages:- Very responsive, snappy conversations
- Feels more dynamic and engaging
- Efficient for simple Q&A
- Better for impatient callers
- May cut off users who pause naturally
- Can interrupt multi-part questions
- Feels rushed for some users
- May miss end of long answers
- Balanced, natural conversation pace
- Allows brief pauses without cutting off
- Works for most use cases
- Professional feel
- May feel slightly slow for rapid exchanges
- Not ideal for very simple interactions
- Very thoughtful, deliberate responses
- Great for users who speak slowly or pause
- Better for complex topics requiring thought
- Comfortable for elderly or non-native speakers
- Can feel slow or unresponsive
- Inefficient for simple questions
- May frustrate fast-paced callers
Response Delay by Use Case
Customer Support (Quick Inquiries)- Recommended: 300-400ms
- Reasoning: Efficiency matters, questions are usually straightforward
- Example: “What are your hours?” → Quick response expected
- Recommended: 500-700ms
- Reasoning: Users need time to describe issues, check things
- Example: “My device shows error X when I…” → User may pause to read error
- Recommended: 300-500ms
- Reasoning: Dynamic conversation, but allow for objection handling
- Example: Balance between energy and letting prospects speak
- Recommended: 600-800ms
- Reasoning: Callers may be elderly, anxious, or need time to find information
- Example: “Let me find my insurance card…” → Needs patience
- Recommended: 700-1000ms
- Reasoning: People need time to think about answers
- Example: “On a scale of 1-10…” → Requires contemplation
- Recommended: 400-600ms
- Reasoning: Moderate pace for checking calendars
- Example: “Let me check my calendar…” → Brief pause acceptable
- Recommended: 600-900ms
- Reasoning: Complex information, users may need to take notes
- Example: Delivering important terms → Don’t rush
Testing Response Delay
Test Scenarios:- Natural Pausers: Have someone who pauses mid-sentence test
- Fast Talkers: Have rapid speakers test for cut-offs
- Question Sequences: “I have two questions…” (does it wait?)
- Listing Items: “I need help with X, Y, and Z” (catches all three?)
- Thinking Out Loud: “So I’m wondering…” (patient enough?)
Interruption Handling
What is Interruption?
Interruption handling determines whether users can speak while the agent is speaking, cutting off the agent’s response. Enabled (Default): Users can interrupt the agent naturally Disabled: Agent completes full response before listening againConfiguration
To enable/disable:- Open Call Settings section
- Find Interrupt toggle
- Toggle on (enabled) or off (disabled)
- Save changes
When Interruptions are Enabled
Behavior:- Agent speaks response
- User starts speaking
- Agent immediately stops and listens
- Agent acknowledges interruption (optional)
- User completes their input
- Agent responds to new input
- Customer service conversations
- Sales calls (engagement is key)
- Support where users may have urgent questions
- Natural, human-like conversations
- Interactive Q&A sessions
- Natural conversation flow
- Users don’t have to wait for full response
- Better for correcting misunderstandings
- More engaging experience
- Agent may be cut off before delivering full message
- Important information could be missed
- May be interrupted by background noise
When Interruptions are Disabled
Behavior:- Agent speaks complete response
- User speaking during this time is ignored
- After agent finishes, system listens for user input
- User must wait for agent to complete
- Critical information delivery
- Legal disclaimers
- Compliance statements
- Step-by-step instructions
- Confirmation of important actions
- Medical information
- Safety warnings
- Ensures complete information delivery
- No partial messages
- Better for structured scripts
- Clearer for complex multi-step instructions
- Feels less natural
- Frustrating if agent is verbose
- Users may want to interject
- Less engaging conversation
You can disable interrupts globally but still allow them for specific situations in your prompt. Example: “Users cannot interrupt normally, but if they say ‘stop’ or ‘wait’, acknowledge immediately.”
Mixed Approach (Prompt-Based)
Even with interrupts enabled, you can guide behavior in your prompt:Interruption Handling by Use Case
| Use Case | Recommendation | Reasoning |
|---|---|---|
| Customer Support | Enabled | Natural conversation, user needs matter most |
| Sales Calls | Enabled | Engagement is critical, handle objections quickly |
| Technical Support | Enabled | User may need to clarify or correct |
| Appointment Booking | Enabled | User may realize conflict mid-confirmation |
| Surveys | Disabled | Complete each question clearly |
| Legal/Compliance | Disabled | Must deliver full disclaimers |
| Medical Information | Disabled | Critical safety information |
| IVR Menu | Disabled | User needs to hear all options |
| Announcements | Disabled | Important broadcast information |
| Emergency Alerts | Disabled | Critical instructions must be heard |
User Inactivity Timeout
What is Inactivity Timeout?
The number of seconds of silence before the agent proactively reaches out to check if the user is still there. Purpose: Handle situations where:- User stepped away
- User is confused or doesn’t know what to say
- Connection issues
- User is multitasking (looking for information)
Configuration
Range: 5-30 seconds Default: 15 seconds Increment: 1 second To configure:- Open Call Settings section
- Locate User Inactivity Timeout slider
- Adjust value (shows in seconds)
- Save changes
How It Works
Flow:- User stops speaking
- Agent responds (if applicable)
- Silence begins
-
After [timeout] seconds of silence:
- Agent proactively asks if user is still there
- “Are you still there?”
- “Is there anything else I can help with?”
- “I’m here whenever you’re ready.”
- If user responds: Continue conversation
- If still silent: Repeat or end call (configurable in prompt)
Timeout Duration Guidelines
Short Timeouts (5-10 seconds): ✅ Good For:- High-volume call centers
- Simple transactional interactions
- Efficient, quick exchanges
- Queue management scenarios
- May feel rushed
- Not enough time to look up information
- Frustrating for multi-tasking users
- General customer service
- Balanced approach
- Most business use cases
- Default recommendation
- May be too long for simple interactions
- May be too short for complex lookups
- Complex information gathering (user checking documents)
- Elderly or deliberate callers
- Healthcare/medical (finding insurance cards)
- Technical support (checking error messages)
- Very long silence can feel awkward
- Inefficient for simple calls
- May indicate dropped connection
Inactivity Timeout by Use Case
Customer Support (General)- Recommended: 15-20 seconds
- Reasoning: Balanced patience, not too rushed
- Poke Message: “I’m still here if you need anything else.”
- Recommended: 20-30 seconds
- Reasoning: Users may be checking error messages, cables, settings
- Poke Message: “Take your time. Let me know when you’re ready.”
- Recommended: 15-25 seconds
- Reasoning: Users checking calendars
- Poke Message: “I’ll wait while you check your schedule.”
- Recommended: 10-15 seconds
- Reasoning: Keep momentum, but allow thinking
- Poke Message: “Do you have any questions I can answer?”
- Recommended: 20-30 seconds
- Reasoning: People need time to think about ratings/feedback
- Poke Message: “Take your time to think about your answer.”
- Recommended: 5-10 seconds
- Reasoning: Simple questions, quick answers
- Poke Message: “Is there anything else you need?”
Customizing Poke Messages
Define how your agent handles silence in your prompt:- Be reassuring, not demanding
- Acknowledge they may be busy (“looking something up”)
- Offer to wait or call back
- Don’t make user feel bad about silence
Max Call Duration
What is Max Call Duration?
The maximum length of time (in seconds) a call can last before being automatically terminated. Purpose:- Prevent runaway costs
- Handle stuck calls
- Set clear boundaries
- Protect against infinite loops
Configuration
Range: 30-600 seconds (0.5 - 10 minutes) Default: 300 seconds (5 minutes) Increment: 10 seconds To configure:- Open Call Settings section
- Locate Max Call Duration slider
- Adjust value (shows in seconds)
- Save changes
How It Works
Flow:- Call begins
- Timer starts
- At [max duration - 30s]: Agent warns (optional via prompt)
- “We have about 30 seconds left. Is there anything else?”
- At [max duration]: Call ends with farewell message
- Call terminates gracefully
Max call duration is a safety net, not a target. Most calls should end naturally via Smart Call End or user ending the conversation.
Duration Guidelines by Use Case
Quick Transactions (30-120 seconds)- Use Cases: Balance check, hours lookup, simple FAQ
- Why: Simple information, in and out
- Example: “What are your hours?” → 45 second call
- Use Cases: Customer support, appointment booking, basic troubleshooting
- Why: Moderate complexity, multiple steps
- Example: Schedule appointment → 2-3 minute call
- Use Cases: Technical troubleshooting, detailed sales, consultations
- Why: Complex issues require time
- Example: Troubleshoot device → 5-7 minute call
- Use Cases: Customer satisfaction surveys, feedback collection
- Why: Multiple questions, time to think
- Example: 10-question survey → 3-4 minutes
Max Duration Recommendations
| Use Case | Recommended Max | Average Expected | Buffer |
|---|---|---|---|
| Balance/Hours Check | 60s | 30-40s | 2x |
| Simple FAQ | 90s | 45-60s | 1.5x |
| Appointment Booking | 240s | 120-180s | 1.3x |
| Customer Support | 360s | 180-240s | 1.5x |
| Technical Support | 480s | 240-360s | 1.3x |
| Sales Qualification | 300s | 180-240s | 1.3x |
| Detailed Survey | 300s | 180-240s | 1.3x |
| Complex Consultation | 600s | 360-480s | 1.3x |
Handling Max Duration in Prompts
Prepare for approaching time limit:Monitoring Call Duration
Use analytics to optimize max duration: Metrics to track:- Average call duration
- % of calls hitting max duration
- Completion rate vs. duration
- User satisfaction vs. duration
- If >5% of calls hit max: Increase limit
- If average is <50% of max: Consider decreasing for efficiency
- If abandonment increases near max: Add warning prompts
Background Noise (Ambient Sound)
What is Background Noise?
Adds subtle ambient sound to the agent’s audio to create a more human, realistic environment. Purpose:- Reduce “dead air” feeling
- Simulate call center environment
- Make AI feel more human
- Provide auditory continuity
Configuration
Options:- Disabled (Default): Pure AI voice, no background
- Enabled: Subtle background ambiance
- Open Call Settings section
- Find Ambient Sound toggle
- Enable or disable
- Save changes
When to Use Background Noise
✅ Good Use Cases: Call Center Simulation- Agent positioned as call center representative
- Simulates realistic environment
- Reduces AI “uncanny valley”
- Match existing IVR or call center sound
- Consistent experience across channels
- Some users find complete silence uncomfortable
- Gentle background provides continuity
- High-end brands may want “perfect” audio
- Background noise can seem cheap
- Medical information
- Legal disclaimers
- Complex instructions
- Creates processing artifacts
- Conflicts with noise removal
- Poor connections already have noise
- Adding more degrades quality
Background Noise Characteristics
Sound Type:- Subtle office ambiance
- Very low volume (not distracting)
- Consistent throughout call
- No specific identifiable sounds
- Mixed at -30dB to -40dB below voice
- Barely noticeable but present
- Never interferes with speech clarity
Background noise is subtle by design. If you can clearly hear it, it’s too loud. It should be almost subliminal.
Thinking Voice
What is Thinking Voice?
Adds natural pauses and verbal fillers to make the agent sound more human and thoughtful. Examples:- “Hmm…”
- “Let me see…”
- “Okay…”
- “Ah, yes…”
- Natural pauses before responses
Configuration
Options:- Disabled (Default): Direct, efficient responses
- Enabled: Natural pauses and fillers
- Open Call Settings section
- Find Thinking Voice toggle
- Enable or disable
- Save changes
How Thinking Voice Works
Without Thinking Voice:Benefits of Thinking Voice
✅ Advantages: More Human-Like- Mimics natural human behavior
- People pause before answering
- Reduces “AI” feeling
- Makes latency feel intentional
- Covers technical delays gracefully
- Users more patient when agent “thinks”
- Natural transitions between topics
- Less jarring subject changes
- Feels more thoughtful
- Users don’t feel rushed
- Comfortable pauses in conversation
- Time to process information
- Adds 1-3 seconds per response
- Less efficient
- Longer call durations
- Some contexts demand crisp responses
- Can seem uncertain or hesitant
- May reduce authority perception
- Technical information delivery
- Time-sensitive scenarios
- High-volume quick transactions
Thinking Voice by Use Case
| Use Case | Recommendation | Reasoning |
|---|---|---|
| Friendly Customer Support | ✅ Enabled | Natural, human conversation |
| Corporate Helpdesk | ❌ Disabled | Professional, efficient |
| Healthcare Scheduling | ✅ Enabled | Warm, caring approach |
| Technical Support | ❌ Disabled | Clarity and precision |
| Sales Calls | ✅ Enabled | Natural rapport building |
| Legal Information | ❌ Disabled | Authority and confidence |
| Surveys | ✅ Enabled | Casual, comfortable |
| Emergency/Urgent | ❌ Disabled | Direct, immediate |
| Appointment Reminders | ❌ Disabled | Clear, concise |
| Lead Qualification | ✅ Enabled | Conversational discovery |
Combining Thinking Voice with Other Settings
Natural Conversation Setup:Call Behavior Presets
Preset 1: Friendly Customer Service
Configuration:Preset 2: Professional Corporate
Configuration:Preset 3: Fast-Paced Sales
Configuration:Preset 4: Patient Healthcare
Configuration:Preset 5: Technical Support
Configuration:Testing Call Behavior
Test Scenarios
Scenario 1: Interruption Test- Have agent give long response
- Interrupt mid-sentence
- Verify agent stops and listens
- Verify it responds to interruption
- Don’t respond after agent speaks
- Wait for inactivity timeout
- Verify poke message triggers
- Verify handling of continued silence
- Engage in conversation
- Approach max duration
- Verify warning (if configured)
- Verify graceful end at max duration
- Ask questions with different natural pauses
- Check for premature responses
- Check for awkwardly long waits
- Adjust response delay accordingly
- Ask: “I have two questions about X and Y”
- Verify agent waits for complete question
- Verify agent addresses both parts
Measuring Success
Key Metrics: Completion Rate- % of calls where task is completed
- Higher = better configuration
- Should match expected duration for use case
- Too long = inefficient
- Too short = rushed/incomplete
- % of callers who hang up mid-call
- High abandonment may indicate poor timing
- Post-call surveys
- Feedback on conversation flow
- Natural vs. robotic feel
- If users frequently repeat themselves
- May indicate response delay too short
Troubleshooting Common Issues
Agent Cuts Off Users Mid-Sentence
Problem: Agent responds before user finishes speaking Solution:- Increase Response Delay (+100-200ms)
- Test with users who speak slowly or pause
- Verify no echo/feedback creating false endpoints
Agent Feels Unresponsive
Problem: Long pauses after user speaks Solution:- Decrease Response Delay (-100-200ms)
- Check LLM latency (may be model issue)
- Verify adequate server resources
Users Complain About Being Rushed
Problem: Not enough time to think or respond Solution:- Increase Inactivity Timeout (+5-10s)
- Enable Thinking Voice for more natural pace
- Increase Response Delay slightly
Calls Frequently Hit Max Duration
Problem: >5% of calls reaching time limit Solution:- Increase Max Call Duration (+60-120s)
- Review if agent is too verbose
- Check for conversation loops
Users Interrupt Constantly
Problem: Frequent interruptions disrupting flow Solution:- Shorten agent responses in prompt
- Keep responses to 20-30 seconds max
- Use “Let me share important info” for critical sections
Poke Messages Feel Annoying
Problem: Users react negatively to timeout prompts Solution:- Increase Inactivity Timeout
- Soften poke message language
- Reduce frequency of repeat pokes
Next Steps
Intelligence Features
Enable smart features like gender detection and smart call end
Write Effective Prompts
Optimize your agent’s instructions for better behavior
Voice Settings
Choose the right voice and language settings
Testing Agents
Comprehensive testing before production
Need advanced control? Explore Flow Agent Call Settings for node-level behavior customization.