A voice AI that makes calls but doesn't systematically report on those calls is half a product. The call happened — but what was said? What did the customer feel? What was the outcome? What should happen next?
Agni answers all of these questions automatically, for every single call, within seconds of call completion.
What a Post-Call Summary Contains
At the end of every Agni call, the system generates a structured post-call summary containing:
- Call outcome classification: One of a configured set of outcomes (e.g., Payment Promised, Call Back Requested, Not Interested, Transferred to Agent, No Answer, Voicemail)
- Key data points extracted: Any structured data captured during the call — payment date promised, callback time requested, objection raised, address correction given
- Natural language summary: A 2–4 sentence human-readable summary of the call, generated by the LLM from the transcript
- Full transcript: Timestamped, speaker-separated text of the full conversation
- Recording link: Secure URL to the call recording (stored on India-based servers, accessible for 2 years)
- Sentiment score: A -1 to +1 scalar representing the overall customer sentiment across the call
- Emotion tags: High-granularity tags for detected emotion segments (e.g., frustrated: 00:45–01:20, cooperative: 01:20–02:10)
Auto-tagging at scale: For a campaign of 10,000 calls, Agni generates 10,000 structured summaries automatically — no human review required to get usable data into your CRM and analytics pipeline.
Sentiment Analysis: How It Works
Agni's sentiment analysis operates on two dimensions simultaneously:
Acoustic Sentiment (Emotion from Voice)
The audio signal itself carries emotional information — pitch variance, speaking rate, intensity, pauses. Agni's emotion model analyzes these acoustic features in real time throughout the call, producing a continuous emotion track. This track is then aggregated into the post-call sentiment score and emotion tags.
Acoustic sentiment catches things that words don't say: a customer who says "okay fine" in a frustrated tone is flagged differently from one who says it with genuine acceptance.
Semantic Sentiment (Emotion from Words)
The transcript is analyzed for semantic sentiment — what was actually said. Complaint words, positive acknowledgment, threat language, and appreciation are all detected and weighted into the final sentiment score.
The combined acoustic + semantic score is more accurate than either alone, particularly for Indian languages where indirect expression is culturally more common than direct complaint.
Webhook Delivery: Real-Time CRM Updates
Post-call data is delivered via webhook to any configured endpoint within 30 seconds of call completion. The webhook payload is structured JSON containing all summary fields — designed to be parsed and written directly to a CRM record without transformation.
Native integrations write directly to:
- GoHighLevel: Contact notes, custom fields, opportunity stage updates
- Salesforce: Activity log, custom object records, case updates
- Zoho CRM: Call logs, notes, follow-up task creation
- Freshdesk/Freshsales: Ticket updates, contact timeline
- Custom webhook: Any endpoint — your own database, your BI tool, your data warehouse
Analytics Dashboard
Beyond per-call summaries, Agni's analytics dashboard aggregates across your entire campaign and account history:
- Sentiment distribution over time — are your customers getting more or less positive?
- Outcome breakdown by agent, by campaign, by language cohort
- Emotion heatmaps — which conversation segments generate frustration?
- Transcript search — find every call where a specific topic was mentioned
"We used to have a team of 4 people manually reviewing call recordings and writing notes. Agni's post-call summaries replaced that entirely — and the data quality is better because it's consistent across 100% of calls, not a 5% sample." — Operations Director, NBFC (Pune)
Compliance Logging
Post-call summaries and recordings double as compliance artifacts. Every call's consent capture timestamp, disclosure text, and call recording are stored in a tamper-evident format for 2 years — satisfying both RBI recording retention requirements and DPDP audit trail obligations.
Ready to get started?
Post-call summaries and sentiment analysis are included on all Agni plans. Access your analytics dashboard at app.ravan.ai. Webhook documentation at docs.ravan.ai.
