Post-call analysis

​

What the report contains

• Summary - 3–5 sentences of prose in the language of the call.
• Sentiment - overall (positive / neutral / negative) with a score in the -1.0 .. 1.0 range, optionally split by caller and agent.
• Caller intents - up to 5 short phrases describing what they wanted.
• Topics - up to 6 tags for the threads discussed.
• Action items - up to 6 concrete next steps (e.g. “call back Thursday at 10:00”).
• Call outcome - a success flag with a one-sentence justification of whether the call’s goal was achieved.

​

Where you see it

• Call details - the full report with section icons (outcome, sentiment, topics, intents, action items). While the analysis is running, a “Generating full analysis…” spinner is shown.
• Overview and Analytics - aggregated indicators: call mood (average score) and the most frequent topics, also per agent.
• Agent improvement suggestions - the analysis feeds the AI prompt improvement proposals in the agent editor (always subject to your approval).

​

Example report (API)

{
  "summary": "The customer booked a table for 4 on Friday at 7:30 PM...",
  "sentiment": "positive",
  "sentiment_score": 0.62,
  "sentiment_per_speaker": {
    "user":  { "label": "positive", "score": 0.5 },
    "agent": { "label": "positive", "score": 0.7 }
  },
  "intents": ["table reservation for 4", "question about vegan options"],
  "topics":  ["reservation", "menu", "opening hours"],
  "action_items": [
    "Confirm the reservation by SMS before 6:00 PM",
    "Email the vegan menu"
  ],
  "success": true,
  "success_reason": "Reservation booked for Friday 7:30 PM.",
  "language": "en",
  "generated_at": "2026-04-23T22:15:01+00:00"
}

​

Notes

• The analysis requires a transcript of at least ~50 characters - very short calls are skipped.
• The report language follows the AI Summary Language setting on the agent’s Settings tab. The default, Auto, takes the language the agent itself speaks (from its transcriber and voice configuration), so an agent set to Polish gets a Polish report even after a short call. Only a multi-language agent follows the language of the call itself.
• The short call_summary and the full analysis are generated independently - even if the full analysis temporarily fails, the summary remains.