Skip to main content

​
What are Extraction Templates?

Extraction templates define structured schemas for extracting specific information from call transcripts. They use LLMs to analyze conversations and extract data into predefined JSON structures, making it easy to capture lead qualifications, customer information, appointment details, and more.

​
Key Features

  • Custom JSON Schemas: Define exactly what data you want to extract
  • Flexible Instructions: Provide context-specific guidance to the extraction model
  • Multiple LLM Support: Choose from various models including GPT-4.1-mini
  • Template Reusability: Create templates once and use them across multiple agents

​
Endpoints

GET /extractions
GET /extractions/{template_id}

​
Common Use Cases

  • Lead Qualification: Extract budget, timeline, and decision-maker information
  • Appointment Booking: Capture preferred dates, times, and contact details
  • Customer Support: Extract issue type, severity, and resolution status
  • Sales Calls: Capture product interest, pricing discussions, and objections
  • Survey Data: Structure responses to specific questions

​
Extraction Output Format

Extraction results are returned in execution data as a nested JSON structure: 
{
  "extracted_data": {
    "Category Name": {
      "Extraction Name": {
        "subjective": "Free text response from LLM",
        "objective": "Pre-defined value or null"
      }
    }
  }
}
Example Response: 
{
  "extracted_data": {
    "Agent Handover": {
      "Agent Handover Needed": {
        "subjective": "The customer requested to speak with a human agent about pricing options.",
        "objective": "Yes"
      }
    },
    "Visit Details": {
      "Rescheduled visit date": {
        "subjective": "24/03/2026",
        "objective": null
      },
      "Visit Confirmation Status": {
        "subjective": "confirmed",
        "objective": null
      }
    }
  }
}

​
Field Descriptions

  • subjective: Free text response generated by the LLM based on the extraction prompt
  • objective: Pre-defined value selected from configured answer options, or null if not configured
For detailed usage examples and best practices, see the Using Extractions guide.