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

# Intent Discovery

[Back to NLP Topics](https://koreai.mintlify.site/ai-for-service/automation/natural-language/nlp-topics.md)

Intent Discovery auto-extracts popular intents from historical conversation transcripts. Use it to accelerate bot development by surfacing intents you may have missed, then add them as new dialog tasks/FAQs or use their utterances to train existing ones.

> [!NOTE]
>
> Available for English language and Enterprise users only.

**Go to:** Automation AI > Tools > **Intent Discovery**

***

## Workflow

1. **Create a Project** — Organize intent discovery by business use case.
2. **Upload Transcripts** — Upload historical conversation CSV files.
3. **Extract Intents** — Platform auto-identifies intents and qualified utterances.
4. **Train the Utterances** — Add new intents or train existing dialogs/FAQs.

***

## Create a Project

1. Go to **Automation AI > Tools > Intent Discovery**.
2. Click **Create Project**.
3. Enter a **Name** (max 256 chars) and **Description** (max 1000 chars). Click **Proceed**.

The project landing page lists all projects with:

| Column | Description |
| - | - |
| **Status** | No Files Uploaded / Intents Not Extracted / Intent Extraction In Progress / Intents Extracted |
| **Sessions** | Number of sessions from uploaded transcripts. |
| **Intents Extracted** | Count of extracted intents. |
| **Results** | Added Dialogs, Added FAQs, Trained Dialogs, Trained FAQs. Displays NA until extraction completes. |

> [!NOTE]
>
> Maximum 10 projects per app. Project name and description can be updated anytime.

***

## Upload Transcripts

### CSV File Format

All columns are required.

| Field | Description |
| - | - |
| **Id** | Session ID. All messages in a session must share the same ID. |
| **Sender ID** | ID of the AI Agent or human agent. |
| **Other Party ID** | ID of the user. |
| **Direction** | `in` (user input) or `out` (agent response). |
| **Message** | The message content. |
| **Message Type** | `text` or `html`. |
| **Created on** | Date and time the message was sent/received. |

**Limits:** Up to 5 CSV files per project. Max 1 MB per file.

### Steps

1. Click **Upload Transcript** and select your CSV files.
2. After upload, click **Manage Transcripts** to view the file list with:
   * File Name, Upload Time, Sessions count, Status (Conv. Extraction In-progress / Conversations Extracted / Intents Extracted / Invalid File / Upload Failed).

### View Conversations

After successful upload, the platform extracts conversations. Apply a **Date Filter** under **Contents** to browse session summaries (AI Agent message count, user message count). Delete individual conversations that aren't useful.

### Delete a Transcript

Hover over any transcript row and click the **Delete** icon. Use this to remove incorrect or unneeded uploads.

***

## Extract Intents

1. Click **Extract Intents** in the Transcripts dialog box.
2. Wait for extraction to complete (time depends on file count and size).
3. The extracted intents list shows: Intent Name, Session Count, Qualified Utterances, Actions.

### View Extracted Intent Details

Click any intent row to open a detail panel showing:

* **Session ID** — Session ID and timestamp.
* **Utterances** — Up to 3 utterances shown per session; click the chip to expand all. You can bookmark utterances.
* **Status** — `Yet to Review` (default) or `Reviewed`.
* **View Transcripts** — Opens the Session Details dialog with the full chat transcript and session summary.

***

## Train the Utterances

For each extracted intent, use the **Add/Train** dropdown to take action.

### Add a New Dialog Task

1. Select **Add > Dialog Task**.
2. Enter a name and description.
3. Select the required utterances from the qualified list (bookmarked ones are marked).
4. Click **Add to Bot**.

### Train an Existing Dialog Task

1. Select **Train > Dialog Task**.
2. Select the target dialog task from the **Intent** dropdown.
3. Select utterances and click **Add Utterances**.

### Add a New FAQ

1. Select **Add > FAQ**.
2. Enter a name and description.
3. Select utterances and click **Next**.
4. Enter the FAQ question text and bot response. Selected utterances become alternate questions.
5. Set channel, display name, reference ID as needed. Click **Add to Bot**.

### Train an Existing FAQ

1. Select **Train > FAQ**.
2. Select the target FAQ from the dropdown.
3. Select utterances and click **Add Utterances**.

### Change Utterance Status

In the Extracted Intent Details panel, change status between **Yet to Review** and **Reviewed** for one or multiple utterances at once.

The **Actions** column and the project-level **Results** column reflect all add/train operations with running counts.

***

## Reusing a Project

Add more transcripts to an existing project at any time to extract additional intents. You can also delete obsolete transcript files.

> [!WARNING]
>
> Uploading new transcript files deletes the following from the project: extracted intents, generated summaries, session utterance groupings, review status flags, and bookmarks. Previously loaded transcript files aren't deleted. Re-extracting intents restores the deleted data.

***

## Delete a Project

Click the delete icon on the project row and confirm. All project data is deleted, including transcripts, extracted intents, summaries, utterances, status flags, and bookmarks.

> [!NOTE]
>
> Deleting a project doesn't affect Dialog Tasks or FAQs already trained from that project.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.