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

# Evaluations

> Analyze the quality of your agent's conversations with rich filters, scores, and per-conversation details.

**Evaluations** is a deep-dive analytics view of your agent's completed conversations. Unlike the Supervise > Conversations view which shows all conversations at a glance, Evaluations is built for quality analysis—it surfaces scored, classified, and filterable conversation data so you can identify what's working and what needs improvement.

## What you'll see

The Evaluations view shows a table of conversations with additional quality metrics attached to each one:

| Column           | Description                                                |
| ---------------- | ---------------------------------------------------------- |
| **Channel**      | Where the conversation took place                          |
| **Topic**        | The auto-classified topic of the conversation              |
| **Sentiment**    | Positive, neutral, or negative user sentiment              |
| **Task success** | Whether the agent successfully resolved the user's request |
| **CSAT**         | Customer satisfaction score (if collected)                 |
| **Date**         | When the conversation took place                           |

Click any row to open the conversation panel and read the full transcript, summary, and resolution details.

## Filters

Use the filter bar at the top to slice the data:

* **Date range** — select a period or use presets (MTD, QTD, YTD, custom)
* **Channel** — filter by the communication channel
* **Topic** — filter by one or more auto-classified topics
* **Sentiment** — positive, neutral, negative
* **Task success** — filter for resolved or unresolved conversations
* **CSAT score** — set a minimum and maximum CSAT threshold (1–5)
* **Tags** — filter by tags assigned to conversations
* **Conversation IDs** — jump directly to specific conversations

### Metadata Filters

Below the standard filters, you can expand a **Metadata Filters** section to filter conversations by **user metadata** passed from your integrated channels (for example, `customer_id`, `email`, or any custom attribute).

* Metadata fields are **grouped by channel** (e.g., *Freshchat Bubble*, *WhatsApp Meta*).
* Type a value into any field to filter conversations where that metadata matches.
* Use the **✕** button on a channel header to clear all metadata filters for that channel.
* Collapse or expand each channel group to keep the filter bar organized.

<Tip>
  Combine metadata filters with topic or sentiment filters for highly targeted analysis. For example, filter by a specific `customer_id` and negative sentiment to see all problematic conversations for a single customer.
</Tip>

<Note>
  The metadata fields displayed depend on what your integrated channels send. If no metadata has been received, the section shows an informational message.
</Note>

<Tip>
  Start by filtering for **unresolved conversations with negative sentiment** — this combination usually reveals the highest-priority issues to address in your agent's knowledge or prompts.
</Tip>

## Connection to Hotspots

Evaluations and **Hotspots** work together. When you click on a topic in the Hotspots view, you're taken directly to Evaluations pre-filtered by that topic and date range. This lets you go from "this topic has a high priority score" to "let me read the actual conversations" in one click.

## Best practices

* **Review low CSAT conversations** to understand what frustrated users and improve agent responses.
* **Sort by task success** to find patterns in unresolved conversations—are they all about the same topic?
* **Use topic filters** to focus your analysis on a specific area of your product or service.
* **Use metadata filters** to drill down into conversations for specific customers or user segments.
* **Export data** for external analysis or stakeholder reporting using the CSV export option in the table toolbar.