---
sourceDocument: Yokohama Enable AI
sourceDocumentLink: https://servicenow-prod.fluidtopics.net/r/yokohama/intelligent-experiences

 Release :

    - yokohama

ft:locale :

    - en-US

ft:publication_title :

    - Yokohama Enable AI

ft:clusterId :

    - platai

bundleId :

    - platai

workflow :

    - Platform


---

# Intent Discovery

# Intent Discovery {#ariaid-title1}

Release version: Yokohama  
Updated January 30, 2025  
![](https://www.servicenow.com/docs/portal-asset/ico-clock) 6 minutes to read
Summarize  
![AI sparkle icon](https://servicenow.com/docs/portal-asset/ai-sparkle-icon) Summarized using AI  
This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.  

## Summary of Intent Discovery

The Intent Discovery application in ServiceNow helps identify opportunities for incident deflection by analyzing historic incident or task data to uncover useful intents for Natural Language Understanding (NLU).
It assists customers in understanding which prebuilt or custom intents to activate, such as for Virtual Agent and AI Search, enhancing automation and user experience.
Show full answer Show less  
Intent Discovery is a standalone application available on the ServiceNow Store and appears under **All \> NLU Workbench \> NLU Advanced Features** after installation, but it is not part of the NLU Workbench - Advanced Features installation.

## Key Features

* **Intent Analysis:** Runs on historic data like incidents, analyzing key text fields (e.g., short description) to identify intents and classify records.
* **Taxonomy Support:** Allows running analysis against predefined intent taxonomies (e.g., ITSM domain) to match records to known intents and identify unmatched utterances.
* **Clustering:** Groups unmatched records into clusters by keywords, enabling manual review and creation or expansion of intents.
* **Intent Importing:** Recommended intents from the analysis can be reviewed and added directly to existing or new custom NLU models within the same application scope.
* **Manual Utterance Addition:** Users can add clustered utterances to intents and models with the ability to edit or delete utterances, supporting continuous improvement of NLU models.
* **Report Management:** Users can run, rerun, delete, and manage multiple versions of analysis reports to keep intent data current.

## How It Works

To create an Intent Discovery report, an admin selects a data source (e.g., Incident table), filters records, specifies a field to analyze (like short description), selects a taxonomy, and optionally enables clustering for unmatched utterances. The system then processes the data, generating a report showing matched intents, unmatched records, and clusters.

Users can then:

* Review recommended intents and add them to NLU models to enhance automated handling of similar future incidents.
* Examine clusters of unmatched utterances to create or expand intents by adding relevant utterances manually.
* Run repeated analyses to refine intent coverage over time.

## Benefits for ServiceNow Customers

* Improves Virtual Agent and AI Search accuracy by identifying relevant intents from historical data.
* Enables proactive incident deflection by surfacing common intents that can be automated.
* Supports continuous enhancement of NLU models with data-driven intent and utterance recommendations.
* Facilitates efficient management of intent models within the ServiceNow platform's application scopes.

## Getting Started

* Install Intent Discovery from the ServiceNow Store with admin privileges.
* Access it under NLU Advanced Features and run your first analysis on incident or task data.
* Review and import recommended intents into your NLU models or create new intents from clustered utterances.
* Iterate analyses regularly to maintain and improve intent coverage and incident deflection effectiveness.  
Use the Intent Discovery application to help identify opportunities for
incident deflection. For example, you can use it to identify which Virtual Agent
conversations to activate next.

## Summary usage {#intent-discovery__section_nhl_155_vnb}

For applications that consume NLU, such as Virtual Agent and AI Search, Intent Discovery helps
you to better understand which prebuilt intents you can benefit from, and which custom intents
would be useful to create.

Intent Discovery provides an analysis that you run on historic incident data or
other task data. You can also group the run's remaining records into different clusters so you
can manually add utterances to NLU intents. In addition, you can use
specific clusters to create new intents in a model.

In this example scenario, you're using Intent Discovery to identify the top
intents in your instance, and how much coverage they can provide across your historic incident
records.

## Installation {#intent-discovery__section_pch_q2l_2qb}

Intent Discovery is available from the ServiceNow Store. For more
information, see [Install Intent Discovery](https://servicenow-prod.fluidtopics.net/diJdbUSVIuId7YSFnN8Heg "You can install the Intent Discovery application (sn_nlu_discovery) if you have the admin role.").  
After Intent Discovery is installed and activated, it appears under AllNLU WorkbenchNLU Advanced Features.  
Note:  
Although organized under NLU Advanced Features in the navigation pane, Intent Discovery is a separate application that is not included when installing NLU Workbench - Advanced Features.

## Intent Discovery report details {#intent-discovery__section_y5b_mbz_5sb}

* When Taxonomy is selected, the generated report contains intent recommendations against the selected taxonomy. A taxonomy is a prebuilt library of intents in a specific domain. While you don't have access to the underlying intents, when you run Intent Discovery against a specific taxonomy, data that maps to any intent in the taxonomy will be identified.
* Unmatched records are the utterances which couldn't match to any intent in the taxonomy.
* Recommended intents are the intents which are found from utterances that data was run on.
* The percentage of Unmatched records (clustered) are the records that aren't classified (records that don't belong to any of the recommended intents).
* The percentage of unmatched records and the number of recommended intents don't need to match. It's a coincidence if they match.
{#intent-discovery__ul_ztg_nbz_5sb}

## Creating an Intent Discovery report {#intent-discovery__section_fyy_155_vnb}

1. Using the admin or nlu_admin role, navigate to AllNLU WorkbenchNLU Advanced FeaturesIntent Discovery.  
2. Select either Run analysis or Find recommendations.Figure 1. Intent Discovery landing page

## Running an analysis on the report {#intent-discovery__section_l2s_ssd_wnb}

1. For this example report, you configure the following fields on the Intent Discovery \> Create new screen.

* Data Source: Select the Incident (incident) table.
* Filter by: \[Created\] \[on\] \[This quarter\]
* Field to analyze: Short description (short_description). You choose Short description because it's a highly used string field that references words that can help the system identify an intent.
* Taxonomy: Select ITSM. This field tells the system to run classification processing on your ITSM incident records. It has 3 options: Classification, ITSM, or blank, which defaults as Classification.
* Cluster unmapped utterances by keywords... : Select the check box. When you check this box, the system groups your incident records that weren't classified into clusters.
* Report name: The field automatically defaults to Incident \<month/day/year\>. You can edit the name if you prefer. In this example scenario, you enter <kbd class="ph userinput">Incident 12/16/2020 - SF Test</kbd>.
{#intent-discovery__ul_qjk_m33_wnb}

2. Select Run analysis.  
Figure 2. Selecting data sources in Intent Discovery for a run analysis

Result: Your report appears on the Intent Discovery
screen, showing its status as the analysis begins. The subsequent status values appear in the
following order during the analysis: Preparing to run, Work in progress, Clustering, and Done.
This can take from 5 minutes to 30 minutes to complete. The fewer the records you have in a
cluster, the less time it takes. Turning clustering off can also speed up the process.  
Figure 3. An ongoing run analysis  
When the analysis is complete, the column values on the screen appear, with the Status column value set to Done, as shown in the image below.  
Note:  
If you want to delete the report and start over, point to the right of the Status column to invoke the Delete report icon.

3. Select the Name of your report.  
Figure 4. A completed run analysis

Result: The screen refreshes, showing the analyzed incident records and
the remaining incident records that were not classified.

## Importing recommended intents to new or existing custom models {#intent-discovery__section_xrn_pqd_wnb}

Before importing intents to an NLU model, ensure that you are in the same application scope as the model. For more information, see [Select an application from the application picker](https://www.servicenow.com/docs/access?context=t_SelectAnAppFromTheAppPicker&version=yokohama&pubname=yokohama-platform-administration&ft:locale=en-US).

1. On the Records covered by recommendations section of the screen, select the caret icon on a recommended intent you want to add to a custom model.  
Figure 5. Reviewing a recommended intent

Result: The details of the recommended intent appear so you can review
them, as shown in the image below.

2. Select Add to Model.  
Figure 6. Adding a recommended intent to a model  
3. On the Select a destination model screen that appears, choose a model you want to add the recommended intent to. If you can't find an appropriate model, create a new one, return to the report, and add the new model.  
Note:  
The model you choose must have the same application scope as your current scope.

4. Select Save.  
Figure 7. Saving a recommended intent to a model

Result: A banner appears on the screen, confirming the intent is added
to the target model.  
Figure 8. Confirmation of adding a recommended intent to a target model

The recommended intent also appears on the Model screen of the target model, as shown in the
image below.  
Figure 9. View a recommended intent in the target model

## Adding clustered utterances to an intent and its model {#intent-discovery__section_fmv_pqd_wnb}

1. On the Remaining records section of the intent discovery records screen,
select and open a cluster of utterance and short description data that you want to add to an
intent and its associated model.

As you continue to build out new intents from these clusters, you can click the
Ignore icon to remove any unwanted intents from the report.

There's also a Show Additional filter you can use to show or hide the
added intents, and the ignored intents as well.

2. Select Add to intent.  
Figure 10. Adding a cluster to an intent

3. In the Add this cluster to an intent and model screen, select an intent and
model pair you want to associate to this cluster.  
Figure 11. Adding a cluster to an intent and model

4. Enter a few utterance examples into the open text field. Select Add each time you complete your entry to save it in the system. Use the pencil icon or the trash can icon respectively to edit or delete
your entry.

5. Select Save.  
Figure 12. Adding paraphrased utterances to an intent

Result: The records screen appears, showing a banner confirming you
added two new utterances to the target intent and its associated model. The model and intent
pair appears in the Added To column, as shown in the image below.  
Figure 13. Confirmation of adding paraphrased utterances to an intent

Use the Show Additional filter if you want to show or hide the clusters
that have added intents, and the clusters that are ignored.  
Figure 14. Viewing or hiding clusters and ignored clusters

## Running another analysis on your Intent Discovery report {#intent-discovery__section_gwv_sqd_wnb}

1. Select Run Again.  
Figure 15. Selecting the version of the analysis to run

Result: The new run begins. When it's in progress, the option to cancel
the run appears, as shown in the image below.  
Figure 16. The Cancel Run option

When the run is complete, a new banner appears that states you have a new version of the
report.

2. Select the new version, then select Run Again.  
Figure 17. Selecting the new version of the report

Result: The time stamp you selected for the most recent run appears in
the Run date column of the Intent Discovery screen.  
Figure 18. View the new time stamp of the Intent Discovery report
* **[Install Intent Discovery](https://servicenow-prod.fluidtopics.net/diJdbUSVIuId7YSFnN8Heg)**   
  You can install the Intent Discovery application (sn_nlu_discovery) if you have the admin role.

*[\>]: and then