Skip to main content

Getting Started with AI Insights

Configure AI Insights for your assistant in under 10 minutes.

Before You Start​

  • Have an AI Assistant created
  • Assistant should be handling calls (or ready for test calls)
  • Every assistant automatically includes a Call Summary insight (required)

Step 1: Navigate to Your Assistant's Insights Tab​

  1. Go to AI Assistants in the main navigation
  2. Click on your assistant to edit it
  3. Click the Insights tab

You'll see:

  • Call Summary insight (required, automatically included)
  • Add More Insights button
  • View Analytics button
  • Delivery Configuration section

Step 2: Add Insights to Your Assistant​

  1. Click Add More Insights
  2. Browse available insight templates by category:
    • Sales: Lead Quality Score, Purchase Intent, Budget
    • Support: Customer Sentiment, Issue Category, Satisfaction Score
    • Summary: Action Items, Next Steps, Key Details

Popular Templates:

  • Lead Quality Score (0-100): Measures buying intent
  • Customer Sentiment (text): Positive/Neutral/Negative tone
  • Action Items (array): Follow-up tasks from the conversation
  • Key Details (object): Customer name, budget, timeline, pain points
  1. Select the insights you want (start with 2-5)
  2. Click Add to attach them to your assistant

Processing Time Guide:

  • 1-5 insights: 5-15 seconds per call
  • 5-10 insights: 15-30 seconds per call
  • 10+ insights: 30-60+ seconds per call

Step 3: Configure Delivery (Optional)​

In the Delivery Configuration section on the same page:

Portal (Always Enabled)​

  • Insights automatically appear in the analytics dashboard
  • Accessible via the View Analytics button
  • Exportable to CSV

Webhook (Optional)​

  1. Check the Webhook checkbox
  2. Enter your HTTPS endpoint URL: https://your-api.com/insights
  3. Click Save
  4. Insights will be sent in real-time after each call

Requirements:

  • Must use HTTPS (not HTTP)
  • Must be publicly accessible
  • Should return 200 OK within 5 seconds

Email (Optional)​

  1. Check the Email checkbox
  2. Enter email address: [email protected]
  3. Click Save
  4. Receive insights via email after each call

Step 4: Test with a Call​

  1. Call your assistant's phone number
  2. Have a conversation that includes relevant information
  3. End the call
  4. Wait 30-60 seconds for processing

Example test conversation:

"Hi, I'm interested in your service. My budget is around $5,000
and I'm looking to get started next month. Can you help me schedule
an appointment?"

Step 5: View Results​

  1. Go back to your assistant's Insights tab
  2. Click View Analytics button
  3. Find your recent call in the list
  4. See extracted insights with their values

Example Results:

Call Summary: Customer inquired about service, mentioned $5,000 budget...
Lead Quality Score: 85
Action Items: ["Schedule appointment", "Send proposal"]
Key Details: {
  "budget": 5000,
  "timeline": "next month"
}

You can also view insights on individual calls:

  • Go to Analytics → Calls
  • Click on any call
  • Scroll to the AI Insights section

Creating Custom Insights (Advanced)​

If pre-built templates don't meet your needs:

  1. Go to Insights in the main navigation
  2. Click Create Custom Insight
  3. Fill in the details:
    • Name: Descriptive identifier (e.g., "Property Size")
    • Instructions: Detailed extraction rules
    • Output Type: Text, Number, Boolean, Object, or Array
    • Category: Sales, Support, Summary, or Custom
  4. Click Create
  5. Add the custom insight to your assistant via the Insights tab

Example Custom Insight:

Name: Property Size
Instructions: Extract the property size mentioned by the customer.
Look for square footage, lot size, or number of rooms.
Return the numeric value only (e.g., 2500 for "2,500 sq ft").
If not mentioned, return 0.
Output Type: Number

Next Steps​

Best Practices​

Do:

  • Start with 2-5 essential insights (Call Summary is automatic)
  • Test with 10-20 calls before going live
  • Use pre-built templates when available
  • Review extracted data regularly for accuracy

Don't:

  • Add too many insights at once (slows processing)
  • Skip the testing phase
  • Expect 100% accuracy (AI is very good but not perfect)
  • Forget to configure delivery if you need webhooks/email