​

Simulation Agents

​

What is a Simulation Agent?

• Simulate customer conversations based on prompts
• Control conversation flow and timing
• React naturally to your agent’s responses
• Provide consistent testing across multiple runs
• Follow scenario data to create realistic interactions

​

Why Simulation Agents Matter

• Eliminate human bias: Consistent behavior across all tests
• Scale infinitely: Run hundreds of tests simultaneously
• Cover edge cases: Test difficult customers and rare scenarios
• Save time & money: No need for human testers
• Provide 24/7 availability: Test anytime without scheduling

​

Creating a Simulation Agent

​

Step 1: Navigate to Simulation Agents

​

Step 2: Basic Information

​

Agent Name

• Budget-Conscious Insurance Shopper
• Skeptical Senior Customer
• First-Time Insurance Buyer
• Tech-Savvy Professional

​

Agent Type

• Choose between ‘voice’ or ‘chat’ depending on your use case.

​

Prompt

​

Language Model

• Select the language model for the simulation agent (eg. ‘gpt’, ‘claude’ or your custom model)

​

LLM Temperature

• Set the temperature for the language model (0.0 to 1.0) (default: 0.7)

You are calling an insurance company to inquire about {{insurance_type}} options. You are a {{age}}-year-old {{marital_status}} person with {{dependents}}, working as a {{occupation}} with an annual income of {{income}}.

Your characteristics:
- Budget concerns: {{budget_sensitivity}} level
- Insurance knowledge: {{knowledge_level}}
- Main objection type: {{objection_type}}
- Decision timeline: {{urgency_level}}

Variables will be automatically replaced with data from your scenarios when the test runs.

You are calling an insurance company to inquire about life insurance options. You are a 35-year-old married person with two young children, working as a software engineer with an annual income of $120,000.

Your characteristics:
- You're price-conscious but value good coverage
- You have basic knowledge of insurance but need explanations
- You're comparison shopping and mention competitors
- You ask about term vs whole life differences
- You're concerned about monthly premiums fitting your budget

Conversation style:
- Friendly but businesslike
- Ask clarifying questions when terms are unclear
- Express price concerns when quotes are given
- Mention you need to discuss with your spouse before deciding
- Take notes (mention this) during the conversation

Common questions you ask:
- What's the difference between term and whole life?
- How much coverage do I need for my family?
- What are the monthly payment options?
- Can I increase coverage later?
- What happens if I miss a payment?

Objections to raise:
- "That seems expensive compared to what I saw online"
- "I need to think about it and discuss with my spouse"
- "I'm not sure I need that much coverage"

• Reusability: One simulation agent can handle multiple customer profiles
• Consistency: Ensures all test variations use the same base behavior
• Scalability: Easy to test hundreds of scenarios with one agent
• Maintenance: Update the prompt once, affects all test cases
• Data-Driven: Automatically pulls values from your scenario datasets

​

Step 3: Voice Configuration

​

Voice Provider

​

Voice Name

​

Interrupt Sensitivity

• 0.0: Very difficult to interrupt
• 0.5: Normal conversation flow
• 1.0: Very easy to interrupt

​

Conversation Speed

• 0.5: Slow, elderly speaker
• 1.0: Normal speed
• 1.5: Fast, energetic speaker

​

Finished Speaking Sensitivity

• 0.0: Waits longer before responding
• 0.5: Normal pause detection
• 1.0: Responds very quickly

​

Step 4: Call Settings

​

Max Call Duration (minutes)

• Typical insurance sales calls: 15-30 minutes

​

Initial Message Delay (seconds)

• Simulates realistic call connection time

​

Initial Message (Optional)

• Leave empty for agent to speak first (inbound calls)
• Add greeting for outbound simulation: "Hi, I'm calling about life insurance options I saw on your website"

​

Example Simulation Agent Configurations

​

Example 1: Dynamic Customer with Variables

Name: Dynamic Insurance Customer
Model: gpt-4
Temperature: 0.7
Voice: en-US-JennyNeural
Conversation Speed: 1.0

Prompt:
You're a {{age}}-year-old {{family_status}} calling about {{insurance_interest}}. 
Your household income is {{annual_income}} and you have {{dependents}} dependents. 
Your main concern is {{objection_type}} and your budget is {{budget_monthly}}.

Key behaviors:
- Ask questions appropriate to your {{knowledge_level}} knowledge level
- Express concerns based on {{objection_type}}
- Your urgency to buy is {{urgency_level}}
- Mention that {{current_insurance}} is your current coverage situation

​

Example 2: Price-Conscious Young Family (Static)

Name: Young Family Insurance Shopper
Model: gpt-4
Temperature: 0.7
Voice: en-US-JennyNeural
Conversation Speed: 1.0

Prompt:
You're a 32-year-old parent calling about life insurance. You have a 
3-year-old child and another on the way. Your household income is $75,000. 
You're very price-conscious and need affordable coverage. You often mention 
your tight budget and ask about payment plans. You compare prices with 
other companies you've researched online.

​

Example 3: Advanced Template with Conditional Logic

Name: Adaptive Insurance Shopper
Model: gpt-4
Temperature: 0.8
Voice: {{voice_selection}}
Conversation Speed: {{conversation_speed}}

Prompt:
You are {{name}}, a {{age}}-year-old {{occupation}} interested in {{insurance_type}}.

Background:
- Income: {{income}}
- Family: {{family_status}} with {{dependents}} dependents
- Location: {{location}}
- Current coverage: {{current_insurance}}

Your behavior depends on your profile:
- If income > $100k: Focus on comprehensive coverage and tax benefits
- If income < $50k: Very price-sensitive, ask about payment plans
- If age > 60: Ask about pre-existing condition coverage
- If dependents > 0: Emphasize family protection needs

Questions to ask:
- "What's covered under {{insurance_type}}?"
- "How does this compare to {{competitor_name}}?"
- "What happens if {{specific_concern}}?"

Always mention your {{objection_type}} concern during the conversation.

​

Best Practices for Simulation Agents

​

1. Write Detailed Prompts

• Include demographic details
• Specify knowledge level
• Define conversation style
• List common questions and concerns
• Include realistic objections

• Replace hard-coded values with {{variable_name}}
• Variables are populated from your scenario data
• One agent can test multiple customer profiles
• Example: {{age}} instead of “35”, {{income}} instead of “$75,000”

​

2. Match Voice to Persona

• Younger voices for millennials
• Professional voices for executives
• Regional accents if testing geographic markets

​

3. Calibrate Conversation Settings

• Seniors: Slower conversation speed
• Executives: Higher interrupt sensitivity
• First-time buyers: Lower finished speaking sensitivity

​

4. Test Different Scenarios

• Various age groups and income levels
• Different insurance knowledge levels
• Multiple conversation styles
• Various objection patterns

​

5. Iterate Based on Results

• Review conversation logs
• Adjust prompts for more realistic behavior
• Fine-tune conversation settings
• Update based on real customer patterns

​

Working with Variables

​

Available Variables

• Demographics: {{name}}, {{age}}, {{gender}}, {{location}}
• Financial: {{income}}, {{budget_monthly}}, {{credit_score}}
• Insurance: {{insurance_interest}}, {{current_insurance}}, {{coverage_amount}}
• Behavioral: {{objection_type}}, {{urgency_level}}, {{knowledge_level}}
• Family: {{family_status}}, {{dependents}}, {{spouse_employed}}

​

Variable Naming Conventions

• Use lowercase with underscores: {{annual_income}} not {{AnnualIncome}}
• Be descriptive: {{preferred_contact_time}} not {{pct}}
• Match your dataset column names exactly

​

Testing Your Variables

1. Check that variable names match your dataset columns
2. Preview a test with one scenario to verify variable replacement
3. Ensure all required variables have values in your dataset

​

Common Use Cases

​

Sales Training Validation

• Handle price objections effectively
• Explain products clearly
• Build rapport with different personalities
• Close sales appropriately

​

Compliance Testing

• Provides required disclosures
• Doesn’t make false promises
• Handles sensitive information properly
• Follows regulatory guidelines

​

Product Knowledge Assessment

• Explain different insurance types accurately
• Answer technical questions
• Provide appropriate recommendations
• Handle complex scenarios

​

Troubleshooting

​

Simulation Agent Too Predictable

• Increase LLM temperature
• Add more variety to prompt
• Include multiple persona traits

​

Conversations End Too Quickly

• Add more questions to prompt
• Increase engagement instructions
• Adjust finished speaking sensitivity

​

Unrealistic Behavior

• Review and refine prompt
• Check conversation speed settings
• Ensure voice matches persona

​

Next Steps

1. Create test configurations combining agents and scenarios
2. Execute simulation tests to evaluate performance