End of Call Report Event

The end-of-call-report event is triggered when a call completes and contains comprehensive analytics, recordings, cost breakdown, and AI-powered analysis results.

This is the most data-rich webhook event, containing full call metrics, analysis results, and recording information. It’s essential for post-call processing and analytics.

When It’s Triggered

This event is sent once per call when:

The call ends naturally (user or assistant hangs up)
The call is terminated due to timeout
The call is disconnected due to technical issues
Call forwarding completes

Event Structure

{
    "message": {
        "timestamp": 1772702523935,
        "type": "end-of-call-report",
        "call": { /* Enhanced Call Object with metrics */ },
        "assistant": { /* Assistant Object */ },
        "messages": [ /* Complete conversation history */ ],
        "phone": { /* Phone Object */ },
        "customer": { /* Customer Object */ },
        "analysis": { /* Complete Analysis Results */ }
    }
}

Key Features

Enhanced Call Object

The call object includes additional fields only available at call completion:

Field	Type	Description
`endAt`	string	ISO timestamp when call ended
`cost`	object	Detailed cost breakdown by service
`metrics`	object	Comprehensive call performance metrics
`recording`	object	Recording file location and metadata

Complete Analysis Results

Analysis Object

The analysis object contains AI-powered analysis results for the call.

Analysis data is primarily populated in end-of-call-report events. Other events may contain empty analysis objects that will be filled once the call completes.

Field	Type	Description
`summary`	string	AI-generated summary of the call conversation
`successEvaluation`	string	Assessment of call success: “Excellent”, “Good”, “Fair”, “Poor”
`structuredData`	object	Extracted structured data based on configured schema
`outcome`	string	Call outcome classification
`guardrails`	string	Guardrails evaluation result (or “DISABLED”)

Summary

Generated automatically if enabled in assistant configuration
Provides 2-3 sentence overview of what happened during the call
Uses configurable prompt and LLM model

Success Evaluation

Determines if call met objectives based on assistant’s system prompt
Uses descriptive scale: Excellent, Good, Fair, Poor
Configurable evaluation criteria and rubric

Structured Data

Extracts specific data points according to predefined JSON schema
Commonly used for capturing:
Customer information (name, phone, email)
Appointment details
Call metadata
Custom business data

Example Structured Data Schema

{
    "type": "object",
    "properties": {
        "Name": {
            "type": "string",
            "description": "Name of the patient"
        },
        "phone number": {
            "type": "string",
            "description": "Phone number of patient"
        },
        "appointment_date": {
            "type": "string",
            "description": "Scheduled appointment date"
        }
    },
    "required": [
        "Name"
    ]
}

Cost Breakdown

Detailed cost tracking across all services:

Field	Type	Description
`cost.callDuration`	number	Total call duration in seconds
`cost.interactlyCost`	number	Interactly platform fee
`cost.phoneCost`	number	Telephony provider charges
`cost.ttsVendorCost`	number	Text-to-speech generation costs
`cost.sttVendorCost`	number	Speech-to-text transcription costs
`cost.llmVendorCost`	number	Large language model API costs
`cost.totalCost`	number	Sum of all costs

Performance Metrics

Comprehensive metrics for optimization:

Audio Quality: Packet loss, delays, dropped chunks
Response Timing: LLM, TTS, STT latency distributions
Turn Statistics: User/bot turns, interruptions, barge-ins
Error Tracking: Failed requests, retries, timeout counts

Example Payload

{
  "message": {
    "timestamp": 1772702523935,
    "type": "end-of-call-report",
    "call": {
      "id": "WC-82015760-c3bd-427d-a23b-ba9b07e4ab85",
      "status": "finished",
      "startAt": "2026-03-05T09:21:20.063Z",
      "endAt": "2026-03-05T09:22:01.739Z",
      "assistantCallDuration": 41683,
      "callEndTriggerBy": "bot",
      "recording": {
        "s3Bucket": "interactly-qa-us-recordings",
        "path": "team-id/WC-82015760-c3bd-427d-a23b-ba9b07e4ab85.mp3"
      },
      "cost": {
        "callDuration": 41.676,
        "interactlyCost": 0.006946,
        "phoneCost": 0.0165,
        "ttsVendorCost": 0.010620,
        "sttVendorCost": 0.005348,
        "llmVendorCost": 0.0003385,
        "totalCost": 0.0397529
      }
    },
    "analysis": {
      "summary": "User contacted Apollo clinic to book an appointment but ended the conversation abruptly.",
      "successEvaluation": "Poor",
      "structuredData": {
        "Name": "",
        "phone number": "",
        "serial_number": 0
      }
    }
  }
}

Common Use Cases

Call Analytics & Reporting

def process_call_analytics(event_data):
    call = event_data["message"]["call"]
    analysis = event_data["message"]["analysis"]
    metrics = call.get("metrics", {}).get("callStats", {})

    # Store comprehensive call record
    call_record = {
        "call_id": call["id"],
        "duration_seconds": call["assistantCallDuration"] / 1000,
        "end_trigger": call["callEndTriggerBy"],
        "total_cost": call["cost"]["totalCost"],
        "success_rating": analysis["successEvaluation"],
        "summary": analysis["summary"],
        "user_turns": metrics.get("turns", {}).get("user", 0),
        "bot_turns": metrics.get("turns", {}).get("bot", 0),
        "interruptions": metrics.get("turns", {}).get("interrupted", 0),
        "avg_response_time": metrics.get("llmStats", {}).get("responseLatencyAvg", 0),
        "stt_confidence": metrics.get("sttStats", {}).get("confidenceAvg", 0)
    }

    # Store in analytics database
    analytics_db.call_records.insert_one(call_record)

    # Generate insights
    generate_performance_insights(call_record)

Recording Management

def handle_call_recording(event_data):
    call = event_data["message"]["call"]
    recording = call.get("recording", {})

    if recording.get("s3Bucket") and recording.get("path"):
        # Download recording for processing
        recording_url = f"https://{recording['s3Bucket']}.s3.amazonaws.com/{recording['path']}"

        # Store recording metadata
        recording_record = {
            "call_id": call["id"],
            "s3_bucket": recording["s3Bucket"],
            "s3_path": recording["path"],
            "recording_url": recording_url,
            "duration_ms": call["assistantCallDuration"],
            "file_size": estimate_file_size(call["assistantCallDuration"]),
            "created_at": datetime.utcnow()
        }

        db.recordings.insert_one(recording_record)

        # Queue for additional processing
        queue_recording_analysis(call["id"], recording_url)

CRM Integration

def sync_to_crm(event_data):
    call = event_data["message"]["call"]
    analysis = event_data["message"]["analysis"]
    customer = event_data["message"]["customer"]

    structured_data = analysis.get("structuredData", {})

    # Extract customer information
    customer_name = structured_data.get("Name", "")
    customer_phone = structured_data.get("phone number", "")

    if customer_name or customer_phone:
        # Create or update contact in CRM
        crm_contact = {
            "name": customer_name,
            "phone": customer_phone,
            "last_interaction": call["endAt"],
            "call_summary": analysis["summary"],
            "call_success": analysis["successEvaluation"],
            "call_duration": call["assistantCallDuration"],
            "call_recording_url": get_recording_url(call["recording"])
        }

        crm_client.upsert_contact(customer_phone, crm_contact)

        # Create activity record
        crm_client.create_activity({
            "contact_phone": customer_phone,
            "type": "phone_call",
            "subject": f"AI Assistant Call - {analysis['successEvaluation']}",
            "body": analysis["summary"],
            "duration_seconds": call["assistantCallDuration"] / 1000,
            "date": call["endAt"]
        })

Cost Tracking & Optimization

def track_call_costs(event_data):
    call = event_data["message"]["call"]
    cost = call["cost"]
    metrics = call.get("metrics", {}).get("callStats", {})

    # Store cost breakdown
    cost_record = {
        "call_id": call["id"],
        "assistant_id": call["assistantId"],
        "duration_seconds": cost["callDuration"],
        "total_cost": cost["totalCost"],
        "platform_cost": cost["interactlyCost"],
        "phone_cost": cost["phoneCost"],
        "tts_cost": cost["ttsVendorCost"],
        "stt_cost": cost["sttVendorCost"],
        "llm_cost": cost["llmVendorCost"],
        "cost_per_minute": cost["totalCost"] / (cost["callDuration"] / 60),
        "date": call["endAt"][:10]  # Date for aggregation
    }

    db.costs.insert_one(cost_record)

    # Analyze cost efficiency
    tts_usage = metrics.get("ttsStats", {})
    cache_hit_rate = tts_usage.get("cacheRequests", 0) / max(tts_usage.get("totalRequests", 1), 1)

    # Alert on high costs or low cache efficiency
    if cost["totalCost"] > COST_THRESHOLD:
        alert_high_cost(call["id"], cost["totalCost"])

    if cache_hit_rate < 0.5:
        alert_low_cache_efficiency(call["assistantId"], cache_hit_rate)

Performance Monitoring

def monitor_call_performance(event_data):
    call = event_data["message"]["call"]
    metrics = call.get("metrics", {}).get("callStats", {})

    # Check LLM performance
    llm_stats = metrics.get("llmStats", {})
    avg_response_time = llm_stats.get("responseLatencyAvg", 0)

    if avg_response_time > LLM_LATENCY_THRESHOLD:
        alert_performance_issue(
            "slow_llm_response",
            call["assistantId"],
            avg_response_time
        )

    # Check STT quality
    stt_stats = metrics.get("sttStats", {})
    avg_confidence = stt_stats.get("confidenceAvg", 0)

    if avg_confidence < STT_CONFIDENCE_THRESHOLD:
        alert_performance_issue(
            "low_stt_confidence",
            call["assistantId"],
            avg_confidence
        )

    # Check for audio quality issues
    audio_stats = metrics.get("userAudio", {})
    dropped_chunks = audio_stats.get("droppedChunksCount", 0)

    if dropped_chunks > AUDIO_DROP_THRESHOLD:
        alert_performance_issue(
            "audio_quality_degradation",
            call["id"],
            dropped_chunks
        )

Data Retention

Consider implementing data retention policies for end-of-call reports:

def cleanup_old_call_data():
    # Archive calls older than 90 days
    cutoff_date = datetime.utcnow() - timedelta(days=90)

    old_calls = db.call_reports.find({
        "call.endAt": {"$lt": cutoff_date.isoformat()}
    })

    for call_report in old_calls:
        # Move to archive storage
        archive_call_report(call_report)

        # Clean up large fields but keep summary
        db.call_reports.update_one(
            {"_id": call_report["_id"]},
            {
                "$unset": {
                    "messages": "",
                    "call.metrics": ""
                }
            }
        )