> This is a page from the ElevenLabs documentation. For a complete page index, fetch https://elevenlabs.io/docs/llms.txt. For the full documentation in a single file, fetch https://elevenlabs.io/docs/llms-full.txt.

# Analytics

## Overview

The analytics dashboard provides granular, real-time metrics for your conversational agents. You can break down performance across multiple dimensions — by agent, branch, time period, language, call type, model, and more — to understand exactly how your agents are performing in production.

Analytics data is powered by a high-performance columnar database, enabling fast queries across large volumes of conversation data with flexible filtering and grouping.

## Accessing analytics

Navigate to the **Analytics** tab in your agents dashboard. You can view metrics across your entire workspace or filter down to a specific agent.

![Analytics dashboard General tab showing call count, average duration, total cost, and call
volume over time](https://files.buildwithfern.com/https://elevenlabs.docs.buildwithfern.com/docs/028eabc5e8536870547389bc6a07c2fe01d9e9440c4f930cc2f9d1efdf22de37/assets/images/conversational-ai/analytics-general.png)

When running [experiments](/docs/eleven-agents/operate/experiments), you can jump directly to branch-filtered analytics from the agent configuration page using the **View Analytics** button. This pre-applies the agent and branch filters so you can compare variant performance immediately.

## Time range and granularity

Select the time range for your analysis using the date picker at the top of the dashboard. You can choose from preset ranges or define a custom window.

The dashboard automatically adjusts the granularity of time-series charts based on your selected range — hourly buckets for short ranges, daily or weekly for longer ranges.

## Available metrics

### Conversations

* **Call count** — total number of conversations in the selected period
* **Total duration** — aggregate conversation time
* **Average duration** — mean conversation length
* **Total cost** — total spend across all conversations
* **Average cost** — mean cost per conversation

### Performance

* **Agent response latency** — time for the agent to respond (median and percentiles)
* **Error rate** — percentage of conversations with errors
* **Error breakdown** — errors categorized by type (tool failures, LLM errors, connection issues)

### Success evaluation

If you have [evaluation criteria](/docs/eleven-agents/customization/agent-analysis/success-evaluation) configured, the dashboard shows success, failure, and unknown rates for each criterion. This is the primary way to measure business outcomes across experiments.

### Data collection

If you have [data collection](/docs/eleven-agents/customization/agent-analysis/data-collection) configured, collected values are available as filterable dimensions in the dashboard.

### Language breakdown

See the distribution of conversations across languages. This is useful for understanding multilingual adoption and comparing agent performance across different languages.

### Active calls

The dashboard displays the current number of active calls in real time. This reflects ongoing sessions across your workspace and is also available via the API.

## Filtering

Narrow your analytics view by applying filters on any combination of dimensions:

![Tools tab showing average error rate and average tool latency grouped by tool
type](https://files.buildwithfern.com/https://elevenlabs.docs.buildwithfern.com/docs/381d3ab344234172f0cf9474ec7388e19dbb3da7b4b7698becbd839bb96a1e15/assets/images/conversational-ai/analytics-tools.png)

| Filter                  | Description                                                       |
| ----------------------- | ----------------------------------------------------------------- |
| **Agent**               | View metrics for a specific agent                                 |
| **Branch**              | Compare performance across experiment branches                    |
| **Call type**           | Filter by inbound, outbound, or web calls                         |
| **Language**            | Filter by conversation language                                   |
| **Conversation source** | Filter by how the conversation was initiated (widget, phone, API) |
| **LLM model**           | Compare performance across different language models              |
| **TTS model**           | Compare performance across text-to-speech models                  |
| **ASR model**           | Compare performance across speech recognition models              |
| **Tool type**           | Filter by specific tools used in conversations                    |
| **Error type**          | Isolate conversations with specific error categories              |
| **Evaluation criteria** | Filter by success evaluation results                              |

## Grouping

Group metrics by any of the filterable dimensions to break down aggregate numbers.

![LLMs tab showing LLM time to first sentence over
time](https://files.buildwithfern.com/https://elevenlabs.docs.buildwithfern.com/docs/abfe0baeebc774a3782213d19e9fcf9d106cf12112fc530333289d8247f9fc10/assets/images/conversational-ai/analytics-llms.png)

![Turn taking latency chart showing p50, p90, and p99
percentiles](https://files.buildwithfern.com/https://elevenlabs.docs.buildwithfern.com/docs/e010346a296dd85403e9b4519378eba78cefd60aca798645cb435ceefc793bd6/assets/images/conversational-ai/analytics-latency.png)

For example:

* **Group by branch** to compare experiment variants side by side
* **Group by language** to see how agents perform across languages
* **Group by LLM model** to compare model performance and cost
* **Group by call type** to understand differences between inbound and outbound calls

Multiple grouping dimensions can be combined for deeper analysis.

## Workflow analytics

![Workflow analytics tab showing per-node entries, durations, terminations, and edge flow overlaid
on the workflow graph](https://files.buildwithfern.com/https://elevenlabs.docs.buildwithfern.com/docs/a758029a005f0327c1e6319708577efd617399c72cec63d5c342d3af2f0db2a3/assets/images/conversational-ai/workflow-analytics.png)

For agents with a [workflow](/docs/eleven-agents/customization/agent-workflows), the **Workflow** tab in the analytics dashboard overlays usage metrics directly on the workflow graph. Select an agent that uses a workflow to enable the graph view — you can then visualize conversation flow and traffic volume directly on the workflow canvas, with detailed inflow and outflow for every node.

### Node metrics

Click any node in the workflow graph to open the inspector and see:

* **Entries** — how many conversations entered the node in the selected time range
* **Average time spent** — how long the node was active on average across those entries
* **Terminations** — how many conversations ended while the node was active
* **Incoming and outgoing flow** — the distribution of conversations entering from each upstream edge and exiting through each downstream edge

### Drilling into conversations

From the node inspector, click **See conversations** to jump to the conversation history page filtered to that node. This applies the **Node entered** filter — a history filter scoped to a specific agent that surfaces every conversation whose transcript entered the selected workflow node at least once.

The **Node entered** filter is also available directly from the conversation history page.

## Using analytics with experiments

Analytics is the primary tool for measuring [experiment](/docs/eleven-agents/operate/experiments) outcomes. The recommended workflow:

Select the agent running your experiment.

Break down all metrics by branch to see variant-level performance.

Look at the metrics that matter for your hypothesis — success evaluation results, conversation
duration, cost, error rates.

When one variant consistently outperforms, increase its traffic share or merge it to main.

You can jump directly to this view from the agent configuration page by clicking the **View Analytics** button next to your traffic deployment settings. This pre-applies the correct agent and branch filters.

## Next steps

Set up A/B tests and use analytics to measure the impact

Define custom criteria to measure conversation quality

Extract structured data from conversations for analytics

Observe live conversations and send control commands