> For the complete documentation index, see [llms.txt](https://docs.poweromni.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.poweromni.ai/en/administration-and-configuration/feedbacks.md).

# Feedbacks

The **Feedbacks** section allows users to rate Agent responses directly in the Power Omni chat, and allows administrators to track these ratings in a centralized panel in the Administration Portal.

This gives your organization a continuous improvement channel: every positive or negative rating is linked to the agent, the model and the run that generated the response, making it easy to identify where Agents perform well and where they need adjustments.

## Overview

The feature consists of two parts:

* **Rating in the chat (User Portal):** below each agent response, users will find the **Good response** (👍) and **Bad response** (👎) buttons. For negative ratings, users can provide a reason and a comment.
* **Feedbacks Panel (Administration Portal):** administrators can view all recorded ratings, with filters, technical details of the run and access to the rated turn (question and answer).

{% hint style="info" %}

## <mark style="color:blue;">Privacy by default</mark>

<mark style="color:blue;">**Power Omni only stores the identifiers and metadata of the rated turn (agent, model, tokens and run steps), never the text of the messages. The rated question and answer are read directly from Microsoft Foundry at the time they are viewed, and the rest of the conversation is never displayed.**</mark>
{% endhint %}

### Enabling Feedback Collection

Feedback collection is controlled by a toggle at the top of the Feedbacks screen.

{% stepper %}
{% step %}

#### Go to the Administration Portal

{% endstep %}

{% step %}

#### Open Feedbacks

In the side menu, click **Feedbacks**.
{% endstep %}

{% step %}

#### Enable feedback collection

In the **Feedback collection** card, enable the option **Allow users to rate agent responses**.
{% endstep %}
{% endstepper %}

### Behavior when collection is disabled

* The rating actions (👍 and 👎) are **hidden** in the chat for all users.
* No new ratings are recorded.
* Ratings **already collected remain available** in the panel for review.

### How Users Rate a Response

When collection is enabled, the rating buttons appear below each agent response, next to the copy and information options.

When hovering over the buttons, users are informed that, by rating, the organization's administrators will be able to see the question and the answer in order to improve the agent.

### Positive Rating

Click the **Good response** (👍) button. The rating is recorded immediately, with no additional information required.

### Negative Rating

{% stepper %}
{% step %}

#### Click Bad response

Click the **Bad response** (👎) button.
{% endstep %}

{% step %}

#### Select a reason

In the **What went wrong with this response?** window, select a **Reason** (required).
{% endstep %}

{% step %}

#### Add an optional comment

Optionally, describe your experience in the **Comment** field (up to 1,000 characters).
{% endstep %}

{% step %}

#### Send feedback

Click **Send feedback**. A confirmation message will be displayed.
{% endstep %}
{% endstepper %}

### Available Reasons

<table><thead><tr><th width="281">Reason</th><th>When to use</th></tr></thead><tbody><tr><td><strong>Incorrect or incomplete</strong></td><td>The response contains wrong information or leaves out important parts.</td></tr><tr><td><strong>Not what I asked for</strong></td><td>The agent did not understand the request or answered about a different subject.</td></tr><tr><td><strong>Slow or with errors</strong></td><td>The response took too long or failed during the run.</td></tr><tr><td><strong>Inappropriate tone or style</strong></td><td>The content is correct, but the language, format or tone is not appropriate.</td></tr><tr><td><strong>Legal or security issue</strong></td><td>The response contains sensitive or inappropriate content, or poses a potential legal or security risk.</td></tr><tr><td><strong>Other</strong></td><td>Any situation not covered by the previous options. We recommend adding details in the comment.</td></tr></tbody></table>

{% hint style="warning" %}

## <mark style="color:$warning;">Transparency for users</mark>

<mark style="color:$warning;">By rating a response, the organization's administrators will be able to see the question, the answer and the user's comment. The rating stays linked to the user for up to 12 months, or until the conversation is deleted. After that, it is anonymized.</mark>
{% endhint %}

### Feedbacks Panel (Administration)

Below the Feedback collection card, administrators will find the list of all recorded ratings.

### List Columns

<table><thead><tr><th width="171">Column</th><th>Description</th></tr></thead><tbody><tr><td><strong>User</strong></td><td>Email of the user who submitted the rating and the total tokens consumed in the rated turn.</td></tr><tr><td><strong>Agent</strong></td><td>Agent that generated the rated response.</td></tr><tr><td><strong>Model</strong></td><td>AI model used in the run.</td></tr><tr><td><strong>Rating</strong></td><td><strong>Positive</strong> or <strong>Negative</strong>. For negative ratings, the selected reason is shown right below.</td></tr><tr><td><strong>Comment</strong></td><td>Comment provided by the user, or "No comment" when left empty.</td></tr><tr><td><strong>Date</strong></td><td>Date and time the rating was recorded.</td></tr><tr><td><strong>Actions</strong></td><td>Menu with the <strong>View details</strong> and <strong>View rated turn</strong> options.</td></tr></tbody></table>

### List Features

* **Filters:** refine the list to find specific ratings.
* **Refresh** (↻ icon): reloads the list with the most recent ratings.
* **Pagination:** navigate between pages using the **Previous** and **Next** buttons.

### Reviewing a Rating

In the **Actions** column, click the menu icon to access the available options.

### View Details

Opens the **Feedback details** window, divided into two sections:

* **Feedback:** User, Rating (with the reason, when negative), Comment and Date.
* **Run:** Agent, Model, **Input** and **Output** Tokens, and the **Run Id**, which identifies the run in Microsoft Foundry.
* **Run Steps:** expand this section (**+** icon) to view the steps of the run, such as tool and connector calls. This is the main resource for investigating the cause of a bad response.

### View Rated Turn

Displays the user's question and the agent's response that were rated, along with the **User feedback for this response** card. This option is also available through the **View rated turn** button in the details window.

{% hint style="info" %}

## <mark style="color:blue;">Only the rated turn</mark>

<mark style="color:blue;">**Only the rated question and answer are displayed. The rest of the user's conversation remains private and cannot be accessed through the panel.**</mark>
{% endhint %}

### Privacy and Data Retention

* **What is stored:** identifiers and metadata of the rated turn (agent, model, tokens and run steps), plus the rating, reason and comment.
* **What is not stored:** the text of the messages. The question and answer are read from Microsoft Foundry only when they are viewed.
* **Retention period:** 12 months by default.
* **Anonymization:** when the conversation or the user is deleted, or when the retention period is reached, the rating is anonymized. The user, the comment and the link to the conversation are removed, preserving only aggregate data.

{% hint style="danger" %}

## <mark style="color:$danger;">Be careful with comment content</mark>

<mark style="color:$danger;">Instruct users not to include personal data, credentials or sensitive customer information in the comment field, since it is visible to the organization's administrators.</mark>
{% endhint %}