> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sigmamind.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Call Analytics

The **Call Analytics** dashboard gives you a real-time view of how your AI voice agents are performing, from call volume and duration to cost breakdowns and quality scores.

<Info>
  Analytics are scoped to the agents and date range selected in the filter bar. Use the **Agent Name**, **Date Range**, and **Timezone** controls at the top of the page to customize your view.
</Info>

<Frame>
  <img src="https://mintcdn.com/sigmamindai/NAvWXFjx0aiBgyrX/images/agents/callanalytics.png?fit=max&auto=format&n=NAvWXFjx0aiBgyrX&q=85&s=21f7329a95c6c32e46f6edf8f9120028" alt="Callanalytics" width="1913" height="907" data-path="images/agents/callanalytics.png" />
</Frame>

***

## Key Metrics

<CardGroup cols={3}>
  <Card title="Total Call Minutes" icon="clock">
    Cumulative talk time across all selected agents and the chosen date range.
  </Card>

  <Card title="Number of Calls" icon="phone-incoming">
    Total calls handled. Use this alongside **Avg Call Duration** to understand traffic patterns.
  </Card>

  <Card title="Total Spent" icon="circle-dollar-sign">
    All-in cost including LLM inference, telephony, TTS, STT, and platform fees.
  </Card>

  <Card title="Avg Call Duration" icon="timer">
    Mean duration per call in minutes. A falling average on a high-volume day is normal and expected.
  </Card>

  <Card title="Average Cost per Call" icon="receipt">
    Total spend divided by number of calls. Useful for budgeting and capacity planning.
  </Card>

  <Card title="Call Terminated Reason" icon="circle-check">
    Breakdown (count and %) of how calls ended — caller hang-up, agent end, or transfer.
  </Card>
</CardGroup>

***

## Volume & Cost Trends

Each metric card on the dashboard includes a **time-series chart** grouped by the interval you select (Hours, Days, Weeks). These charts help you spot:

| Pattern                                   | What it means                               |
| ----------------------------------------- | ------------------------------------------- |
| Flat volume → sudden spike                | Testing session or new user on-boarding     |
| Rising avg duration early → falling later | Power users first, focused callers later    |
| Cost closely tracks volume                | Healthy — no runaway per-call cost          |
| Cost rising faster than volume            | Investigate LLM prompt length or model tier |

<Tip>
  Switch **Group By** to **Hours** when investigating a specific day's spike — it shows exactly when traffic arrived.
</Tip>

***

## Agent Activity

### Tool Calls

The **# Tool Calls** chart shows how often agents invoked external tools (e.g. CRM look-ups, appointment booking APIs) during calls. The **% Tool Calls (weighted)** metric normalises this against total call volume.

<Warning>
  A very low tool-call rate for agents designed to use tools may indicate missing intent triggers or misconfigured tool definitions. Review your agent's tool configuration if the weighted rate is unexpectedly low.
</Warning>

### Transfer & End Calls

| Metric                          | Description                                                                               |
| ------------------------------- | ----------------------------------------------------------------------------------------- |
| **# Transfer Calls**            | Calls escalated to a live agent. A non-zero value means your fallback routing is working. |
| **# End Calls**                 | Calls the agent actively ended (as opposed to the caller hanging up).                     |
| **% Transfer Calls (weighted)** | Transfer rate as a share of all calls — key self-service KPI.                             |

***

## Call Quality

### Blank Calls

The **% of Blank Calls** chart shows the percentage of calls with no meaningful audio or interaction per day. High blank-call rates are normal during testing phases but should fall toward 0% in production.

### Quality by Duration

The **Call Quality (% of call b/w duration)** stacked bar chart segments calls into duration buckets — longer calls generally indicate higher engagement and are treated as higher quality.

<Note>
  During ramp-up or heavy test days, you may see 100% of calls fall into shorter duration buckets. This is expected and will normalize as real traffic grows.
</Note>

***

## Cost Breakdown

The **Cost Breakdown** chart splits spend into five components:

<AccordionGroup>
  <Accordion title="LLM — Language Model Inference" icon="brain">
    The largest cost driver for most agents. Directly influenced by system prompt length, conversation turns, and the model tier selected in your agent configuration.
  </Accordion>

  <Accordion title="Telephony" icon="phone">
    Per-minute carrier charges for inbound or outbound calls. Fixed rate regardless of agent behaviour.
  </Accordion>

  <Accordion title="TTS — Text-to-Speech" icon="volume-high">
    Charged per character synthesised. Shorter, more concise agent responses reduce TTS cost.
  </Accordion>

  <Accordion title="STT — Speech-to-Text" icon="microphone">
    Per-second transcription cost. Longer silences and hold periods are included in billable STT time.
  </Accordion>

  <Accordion title="Platform Fees" icon="server">
    Fixed SigmaMind infrastructure overhead per call.
  </Accordion>
</AccordionGroup>

***

## Call Source

The **Call Source** donut shows where calls originated:

| Source           | Description                                             |
| ---------------- | ------------------------------------------------------- |
| **Playground**   | Calls made via the SigmaMind in-browser testing sandbox |
| **Web Widget**   | Calls initiated from an embedded widget on your website |
| **Phone Number** | Calls from a provisioned SigmaMind phone number         |
| **API**          | Programmatically triggered calls via the Calls API      |

<Info>
  Use the **↺ refresh icon** on individual charts to reload that metric without refreshing the full page.
</Info>