Copy page

Core Concepts

Analytics

Neuratel provides analytics at two levels: aggregate dashboards for tracking overall performance, and per-call post-call analysis that runs automatically after every conversation.

Dashboard Metrics

The analytics dashboard shows real-time and historical KPIs:

Metric	Description
Total Calls	Number of calls in the selected period
Average Duration	Mean call length in seconds
Success Rate	Percentage of calls that completed successfully
Minutes Used	Total minutes consumed
Inbound / Outbound	Call direction breakdown
Active Calls	Currently in-progress calls

---

Summary Endpoint

Retrieve aggregated analytics for a date range. Supports grouping by agent or phone number.

Code
curl "https://api.neuratel.ai/v1/voice-sessions/analytics/summary?startDate=2025-01-01&endDate=2025-01-31&groupBy=agent" \
  -H "Authorization: Bearer nk_live_your_api_key"

Code
import os, httpx

response = httpx.get(
    "https://api.neuratel.ai/v1/voice-sessions/analytics/summary",
    params={"startDate": "2025-01-01", "endDate": "2025-01-31", "groupBy": "agent"},
    headers={"Authorization": f"Bearer {os.environ['NEURATEL_API_KEY']}"},
)
summary = response.json()

Code
const response = await fetch(
  "https://api.neuratel.ai/v1/voice-sessions/analytics/summary?startDate=2025-01-01&endDate=2025-01-31&groupBy=agent",
  { headers: { Authorization: `Bearer ${process.env.NEURATEL_API_KEY}` } }
);
const summary = await response.json();

Response:

Code
{
  "start_date": "2025-01-01",
  "end_date": "2025-01-31",
  "total_calls": 1543,
  "total_duration": 458720,
  "minutes_used": 7645.33,
  "average_duration": 297,
  "success_rate": 0.87,
  "inbound_calls": 980,
  "outbound_calls": 563,
  "completed_calls": 1342,
  "failed_calls": 201,
  "by_agent": [
    {
      "agent_id": "550e8400-e29b-41d4-a716-446655440000",
      "agent_name": "Customer Support",
      "calls": 892,
      "minutes_used": 4430.67,
      "average_duration": 298,
      "success_rate": 0.89,
      "avg_latency_ms": 720
    }
  ]
}

Parameters

Parameter	Description
startDate	Start date (YYYY-MM-DD) — required
endDate	End date (YYYY-MM-DD) — required
groupBy	agent or phoneNumber
agent_id	Filter to a specific agent

---

Volume Endpoint

Get call volume as a time series for charts and trend analysis.

Code
curl "https://api.neuratel.ai/v1/voice-sessions/analytics/volume?startDate=2025-01-01&endDate=2025-01-07&interval=day" \
  -H "Authorization: Bearer nk_live_your_api_key"

Code
response = httpx.get(
    "https://api.neuratel.ai/v1/voice-sessions/analytics/volume",
    params={"startDate": "2025-01-01", "endDate": "2025-01-07", "interval": "day"},
    headers={"Authorization": f"Bearer {os.environ['NEURATEL_API_KEY']}"},
)
volume = response.json()

Code
const response = await fetch(
  "https://api.neuratel.ai/v1/voice-sessions/analytics/volume?startDate=2025-01-01&endDate=2025-01-07&interval=day",
  { headers: { Authorization: `Bearer ${process.env.NEURATEL_API_KEY}` } }
);
const volume = await response.json();

Response:

Code
{
  "interval": "day",
  "data": [
    {"date": "2025-01-01", "calls": 42, "duration_minutes": 168.5, "inbound": 30, "outbound": 12},
    {"date": "2025-01-02", "calls": 38, "duration_minutes": 152.0, "inbound": 28, "outbound": 10}
  ]
}

Parameter	Description
interval	hour (max 7-day range) or day (max 365-day range)

---

Post-Call Analysis

Every call automatically runs through AI-powered analysis when it ends. Results are attached to the call record and delivered via webhook events.

Summary

A concise paragraph describing what was discussed, the outcome, and any follow-up items identified.

Success Evaluation

Scores each call against a rubric defined in the agent's analytics config. The rubric type determines the scoring format.

Rubric Types

SuccessEvaluationRubric enum values (PascalCase):

Type	Output
PassFail	Binary pass or fail
NumericScale	Score on a numeric range (e.g. 1–10)
DescriptiveScale	Labeled score (e.g. poor / fair / good / excellent)
LikertScale	Agreement spectrum (strongly disagree → strongly agree)
PercentageScale	Score as a percentage
Checklist	Multiple criteria, each independently pass/fail
Matrix	Multiple criteria, each scored on a scale
AutomaticRubric	AI infers the appropriate rubric from your criteria description

Configure the rubric under analytics.post_call.analysisPlan — the same tree as summary and structured-data extraction. Use successEvaluationPrompt for the criteria (free-form text — the prompt determines what the LLM scores against).

Code
{
  "analytics": {
    "post_call": {
      "analysisPlan": {
        "enabled": true,
        "successEvaluationRubric": "Checklist",
        "successEvaluationPrompt": "Score one pass/fail per criterion:\n- Agent greeted the customer by name\n- Agent confirmed the order number before lookup\n- Agent offered a clear resolution\n- Customer expressed satisfaction with the outcome"
      }
    }
  }
}

The verdict lands on the voice session as call_successful ("success" / "failure" / "unknown") with a free-text call_successful_rationale. See Post-call analysis for the access pattern and full field list.

Sentiment Timeline

Per-turn sentiment scoring across the entire conversation. Each turn receives a sentiment value from -1.0 (strongly negative) to 1.0 (strongly positive), producing a timeline that shows how the caller's mood changed during the call.

The overall sentiment trend (improving, declining, stable) is included in the call summary.

Structured Data Extraction

Extract specific fields from the conversation into structured JSON. Define a JSON Schema in the agent's analysisPlan and the analysis engine fills in the values from the transcript.

Extraction runs only when analysisPlan.enabled is true AND structuredDataSchema is non-empty. The optional structuredDataPrompt lets you steer the extraction (defaults to a generic instruction if omitted).

Code
{
  "analytics": {
    "post_call": {
      "analysisPlan": {
        "enabled": true,
        "structuredDataPrompt": "Extract the listed fields. Use null when the caller did not state a value.",
        "structuredDataSchema": {
          "type": "object",
          "properties": {
            "issue_type": { "type": "string" },
            "order_number": { "type": "string" },
            "resolved": { "type": "boolean" },
            "satisfaction_rating": { "type": "number" }
          }
        }
      }
    }
  }
}

After the call, the extracted data is available in the call record under extracted_data:

Code
{
  "extracted_data": {
    "issue_type": "missing_item",
    "order_number": "ORD-2025-8842",
    "resolved": true,
    "satisfaction_rating": 4
  }
}

The same extracted_data field appears in the outbound session_report and analysis_complete webhooks fired to your configured endpoint.

---

Latency Breakdown

Every call record includes a breakdown of processing latency across the voice pipeline:

Metric	Description
latency_ttft_ms	Time from end of caller speech to agent's first LLM token
latency_ttfb_ms	Time from end of caller speech to first audio byte reaching the caller
stt_turnaround_ms	Speech-to-text transcription time

Use these metrics to identify bottlenecks in the pipeline — whether latency is coming from transcription, the language model, or voice synthesis.

---

Export

Analytics data can be exported from the Analytics section of the dashboard. Use the date picker to select a range and click Export to download as CSV.

Rate Limits

Endpoint	Max Range
/v1/voice-sessions/analytics/summary	365 days
/v1/voice-sessions/analytics/volume (hourly)	7 days
/v1/voice-sessions/analytics/volume (daily)	365 days

---

Best Practices

Daily — Check success rates and review any failed calls. Watch for latency spikes.

Weekly — Compare agent performance using groupBy=agent. Identify top performers and agents that need prompt tuning.

Monthly — Export data for reporting. Review sentiment trends and data extraction results to surface systemic issues.

---

Next Steps

• Calls and Sessions — Per-call details, transcripts, and recordings
• Agents — Configure the analytics section on your agents
• Monitoring — Real-time event streaming and active call visibility