Post-Call Intelligence

Automatically extract structured JSON data, compute caller sentiment, generate summaries, and tag call dispositions after every conversation.

Open Assistant Studio →

Overview

The Analytics tab enables automated post-call intelligence. When a call completes, Tunk asynchronously analyzes the entire transcript using an evaluation LLM to extract business data and evaluate call quality without affecting live call latency.

Assistant Analytics Configuration

1. Sentiment Analysis

• When enabled, Tunk computes an overall sentiment score for the interaction:
  • Positive: Caller expressed satisfaction, gratitude, or constructive agreement.
  • Neutral: Standard informational exchange without emotional skew.
  • Negative: Caller expressed frustration, complaints, or dissatisfaction.
• The sentiment score is attached to the call record and delivered in the post-call call.completed webhook (inside data.analytics.sentiment).

2. AI Call Summary

Generate automated summaries of the conversation:

• Preset Summaries: Use standard concise summary formats (e.g. Bulleted Key Points, Executive Brief).
• Custom Summary Prompt: Define your own prompt to enforce specific summarization standards:
  MarkdownCopy

  Summarize this call in 3 bullet points:
  
  1. Primary reason for the caller's inquiry.
  2. Resolution or actions taken by the assistant.
  3. Pending follow-ups or promised customer deadlines.

3. Structured Data Extraction

Extract structured JSON key-value pairs from the conversational dialogue to update your CRM, database, or ticketing system.

You define the extraction schema directly in the Analytics tab:

// Example Extracted JSON Payload
{
  "customer_name": "Jane Doe",
  "appointment_date": "2026-10-15",
  "vehicle_make_model": "Honda Civic 2022",
  "callback_requested": true,
  "order_id": "ORD-94821"
}

4. Custom Call Dispositions

Call dispositions categorize the final business outcome of the call:

• Create custom disposition tags (e.g., Appointment Booked, Voicemail Left, Wrong Number, Escalated to Human, Payment Confirmed).
• Provide an LLM evaluation rule for each disposition:
  Plain textCopy

  Tag as "Appointment Booked" if the caller agreed to a specific date and confirmed their email.
  Tag as "Voicemail Left" if the assistant reached an answering machine greeting.

• Dispositions appear in the Call History table for rapid filtering and team reporting.