Skip to main content

Conversational Forms

Create AI-powered conversational forms that collect data through natural language dialogue instead of traditional form fields.

Conversational Form Example - IT Helpdesk

Example: An IT Helpdesk conversational form collecting issue details through natural dialogue. The AI extracts structured data (issue category, urgency level, description) from the conversation in real-time.

Creating a Conversational Form​

Step 1: Choose Form Type​

When creating a new form, select "Conversational" as the form type:

  1. Click "New Form" from the dashboard
  2. Select "Conversational Form" option
  3. Choose a template or start from scratch

Step 2: Configure Basic Settings​

Set up the basic form information:

  • Form Name: Descriptive name for your form
  • Description: What this form is used for
  • Objective: What you want to accomplish (e.g., "Collect IT support ticket information")
  • Context: Business context and background information

Step 3: Define Topics​

Topics are the areas of information you want to collect. For each topic:

  • Topic ID: Unique identifier
  • Name: Display name
  • Description: What information to collect
  • Priority: Required, Important, or Optional
  • Depth: Surface, Moderate, or Deep questioning

Example Topics (IT Helpdesk):

1. Issue Description
   - Priority: Required
   - Depth: Deep
   - Description: Detailed description of the technical issue

2. Urgency Level
   - Priority: Required
   - Depth: Surface
   - Description: How urgent is this issue?

3. Device Information
   - Priority: Important
   - Depth: Moderate
   - Description: Make, model, and specifications

4. User Contact
   - Priority: Required
   - Depth: Surface
   - Description: Email or employee ID

Step 4: Configure AI Persona​

Customize how the AI interacts with users:

Persona Style:

  • Professional: Formal, business-appropriate
  • Friendly: Warm, approachable
  • Casual: Relaxed, conversational
  • Empathetic: Supportive, understanding

Tone: Custom tone description (e.g., "Helpful and patient")

Behaviors: List of behaviors (e.g., "Ask one question at a time", "Provide reassurance")

Restrictions: What the AI should not do (e.g., "Don't make technical diagnoses")

Step 5: Set Conversation Limits​

Configure conversation boundaries:

  • Max Turns: Maximum number of conversation exchanges (e.g., 15)
  • Max Duration: Maximum conversation time in minutes (e.g., 10)
  • Min Confidence: Minimum confidence threshold for extracted data (0-1, e.g., 0.8)

Step 6: Define Extraction Schema​

Specify what structured data to extract from the conversation:

{
  "issueCategory": {
    "type": "select",
    "required": true,
    "options": ["hardware", "software", "network", "other"]
  },
  "urgencyLevel": {
    "type": "select",
    "required": true,
    "options": ["low", "medium", "high", "critical"]
  },
  "description": {
    "type": "textarea",
    "required": true,
    "minLength": 10
  },
  "userEmail": {
    "type": "email",
    "required": true
  },
  "deviceInfo": {
    "type": "text",
    "required": false
  }
}

Step 7: Configure Storage​

Set up where to store the extracted data:

  • MongoDB Connection: Select from connection vault
  • Database: Target database
  • Collection: Target collection
  • Document Structure: How to structure the saved document
Transcript Storage by Deployment Mode

Conversation transcripts are stored differently depending on your deployment mode:

  • Cloud/Self-Hosted: Transcripts stored at _formMetadata.conversational
  • Standalone: Transcripts stored at root level conversational

See Deployment Modes for complete document structure details.

Form Configuration​

Objective​

A clear statement of what the form should accomplish:

"Collect IT support ticket information including issue description, 
urgency level, device information, and user contact details."

Context​

Background information that helps the AI understand the business context:

"We're an IT support team for a 500-person company. We handle 
hardware issues, software problems, network connectivity, and 
account access requests. Our team responds within 2 hours for 
high-priority issues."

Topics Configuration​

Each topic includes:

  • ID: Unique identifier (e.g., "issue-description")
  • Name: Display name (e.g., "Issue Description")
  • Description: What to collect (e.g., "Detailed description of the technical problem")
  • Priority:
    • required: Must be collected
    • important: Should be collected if possible
    • optional: Collect if relevant
  • Depth:
    • surface: Basic information only
    • moderate: Some detail expected
    • deep: Comprehensive information

Persona Configuration​

{
  "style": "professional",
  "tone": "Helpful and patient, but efficient",
  "behaviors": [
    "Ask one question at a time",
    "Provide reassurance when appropriate",
    "Clarify technical terms if needed"
  ],
  "restrictions": [
    "Don't make technical diagnoses",
    "Don't promise specific resolution times",
    "Don't ask for passwords or sensitive credentials"
  ]
}

Conversation Limits​

{
  "maxTurns": 15,        // Maximum conversation exchanges
  "maxDuration": 10,     // Minutes
  "minConfidence": 0.8   // 0-1 threshold
}

Extraction Schema​

Define the structured data fields to extract:

[
  {
    "id": "issueCategory",
    "name": "Issue Category",
    "type": "select",
    "required": true,
    "options": ["hardware", "software", "network", "other"],
    "validation": {
      "required": true
    }
  },
  {
    "id": "urgencyLevel",
    "name": "Urgency Level",
    "type": "select",
    "required": true,
    "options": ["low", "medium", "high", "critical"]
  },
  {
    "id": "description",
    "name": "Description",
    "type": "textarea",
    "required": true,
    "validation": {
      "minLength": 10,
      "maxLength": 1000
    }
  },
  {
    "id": "userEmail",
    "name": "User Email",
    "type": "email",
    "required": true
  }
]

Advanced Configuration​

Custom Prompts​

For advanced control, you can customize prompt templates:

{
  "greeting": "Hi! I'm here to help you submit an IT support ticket. What kind of issue are you experiencing?",
  "followUp": "Can you tell me more about {topic}?",
  "clarification": "I want to make sure I understand correctly. {question}",
  "confirmation": "Great! I've collected: {summary}. Is there anything else?",
  "completion": "Thank you! Your ticket has been submitted. Ticket ID: {ticketId}"
}

Confidence Scoring​

Each extracted field receives a confidence score (0-1):

  • High (0.8-1.0): Very confident in the extracted value
  • Medium (0.5-0.8): Moderately confident, may need verification
  • Low (<0.5): Low confidence, should ask for clarification

Configure minimum confidence thresholds per field.

Topic Dependencies​

Define relationships between topics:

{
  "deviceInfo": {
    "dependsOn": ["issueCategory"],
    "condition": "issueCategory === 'hardware'"
  }
}

Testing Your Conversational Form​

Test Mode​

Use test mode to try your form:

  1. Click "Test Form" in the form builder
  2. Start a conversation
  3. Review how the AI responds
  4. Check extracted data
  5. Adjust configuration as needed

Preview​

Preview the form as users will see it:

  1. Click "Preview" in the form builder
  2. Interact with the AI
  3. Test different conversation paths
  4. Verify data extraction

Publishing​

Once configured, publish your conversational form:

  1. Click "Publish"
  2. Get a shareable URL
  3. Share with users
  4. Monitor conversations and extracted data

Monitoring & Analytics​

Track conversational form performance:

  • Total Conversations: Number of conversations started
  • Completion Rate: Percentage completed
  • Average Turns: Average conversation length
  • Confidence Scores: Average confidence per field
  • Topic Coverage: Which topics are covered most
  • Extraction Success: Percentage of successful extractions

Best Practices​

  1. Clear Objective: Write a clear, specific objective
  2. Relevant Topics: Focus on essential topics
  3. Appropriate Depth: Match depth to information needs
  4. Test Thoroughly: Test various conversation paths
  5. Monitor Performance: Review analytics regularly
  6. Iterate: Improve based on real conversations

Next Steps​