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

 Release :

    - xanadu

ft:locale :

    - en-US

ft:publication_title :

    - Xanadu Enable AI

ft:clusterId :

    - platai

bundleId :

    - platai

workflow :

    - Platform


---

# Train and try your NLU model

# Train and try your NLU model {#ariaid-title1}

* Release version: Xanadu
* 
* Updated August 1, 2024
* 
* ![](https://www.servicenow.com/docs/portal-asset/ico-clock) 3 minutes to read

Train and try your model
iteratively so that its intents and entities are validated,
compiled, and saved to your model.

## Before you begin

* Make sure that the NLU Workbench - Core plugin, NLU Workbench plugin, and Predictive Intelligence plugin are all installed and activated.
* Create an NLU model. For more information, see [Creating models](https://servicenow-prod.fluidtopics.net/EN~zZkL5GYkqAEESELkXtQ "Creating models is the first step to taking advantage of Natural Language Understanding (NLU) in your instances. Create models for Virtual Agent and AI Search in the NLU Workbench.").
* Create one or more NLU intents and their associated entities for your model. For more information, see [NLU intents](https://servicenow-prod.fluidtopics.net/cNycXRzeIdVnj0UVXOJOBw "Intents drive your models' responses by matching a system action to user inputs. Models with good intents help Virtual Agent and Search respond to your users accurately.").
* If any utterance references a table vocabulary source, ensure that the source has been synchronized so that its values are available to your model. For more information see [Sync a table vocabulary source](https://servicenow-prod.fluidtopics.net/PsdIz2vFddioBuGgsSvgDQ "Synchronize your table vocabulary sources to obtain the latest changes to the ServiceNow source table. Synchronizing your vocabulary sources ensures your NLU models have the latest values when predicting intents.").
* Role required: nlu_editor, nlu_admin, or admin. The NLU editor must be assigned to the model.
{#test-train-nlu-model__ul_j5d_thp_zhb}

## About this task

Training your model saves any changes you made to the content, and checks for
conflicts or errors. Training also makes a model available for publishing.  
After training, you can try your model by manually entering individual utterances to see what intents are predicted.  
Note:  
To run a test of your model against a list of test utterances, see [Test and publish your model](https://servicenow-prod.fluidtopics.net/dhylMV9ceAvrFnKgnGYwIA "Assess the performance of your NLU model to identify areas for improvement. Then publish your model to make it available to other applications such as Virtual Agent.").

The mid-conversation responses of Dialog Acts can't be tried or tested in NLU Workbench.

In this example scenario, you've already built sufficient model content by adding intents, utterances, entities, and their associated annotations. Following the example procedure, you first train your NLU model. Then you try your model by manually entering utterances so you can check the prediction results and confidence scores.

## Procedure

1. Navigate to AllNLU WorkbenchModels.  
   The Virtual Agent tab opens by default.
2. Select the tab for your model's application, then select the name of your model.
3. In the Model details tab of the model overview, make sure there is enough content in Intents, Entities, and Vocabulary.
4. On the Build and train your model card on the model overview, select View phase.  
5. When the Build and train your model phase opens, ensure that the Train model tab is selected.  
   Result: The Train model tab displays the last time the model was trained, and also summarizes content changes since the last training, if any.
6. Select the Train button.  
   Result: The system displays a progress bar during training. When finished, the system displays one of two recommendations:
   * When less than 60% of the model's intents are covered in the default test set, the system recommends adding more test utterances. See [Test set creation and management](https://servicenow-prod.fluidtopics.net/y3xFPXipXEZ~RyjkoUWSRg "Use the default test set of your NLU model to test the model's performance and accuracy. Manage your test set over time by building or updating its content in the NLU Workbench.").
   * When over 60% of the model's intents are covered in the default test set, the system recommends proceeding to testing. See [Test and publish your model](https://servicenow-prod.fluidtopics.net/dhylMV9ceAvrFnKgnGYwIA "Assess the performance of your NLU model to identify areas for improvement. Then publish your model to make it available to other applications such as Virtual Agent.").
   {#test-train-nlu-model__ul_nfk_lf1_hwb}
7. To manually try individual utterances, select the Try model tab.
8. In the text field under Enter an utterance to test, type an utterance and select Go.  

## Result

In this example, you entered <kbd class="ph userinput">I need to update my home address</kbd> as the utterance to try.

1. The system displays the model's confidence threshold, which is 76% in this example.
2. Under Top prediction(s), the system displays all intents that were predicted with a confidence score greater than the threshold.
3. In the example, the intent UpdateAddress is predicted with a confidence score of 97%, which is greater than the threshold of 76%.

{#test-train-nlu-model__ol_i1b_ms1_hwb} The Try model results also display thumbs-up and thumbs-down icons for you to provide feedback. For more information, see [Test panel feedback](https://servicenow-prod.fluidtopics.net/_YwL~LM5HXjUhYj3vHmU_A "When testing your NLU model on the Try model section of the test panel, use this feature to provide feedback on the model's intent predictions.").

## What to do next

* Continue trying various utterances to check that your updates to model content are effective. See [Compare draft and published versions of your NLU model](https://servicenow-prod.fluidtopics.net/xm8dItBZoNmMm9jtbhvMMQ "Compare a draft trained Natural Language Understanding (NLU) model to its most recent published version. Test and review the changes to make sure that your draft model will have increased performance.").
* To test your model against a list of test utterances, use its default test set in the [Test and publish your model](https://servicenow-prod.fluidtopics.net/dhylMV9ceAvrFnKgnGYwIA "Assess the performance of your NLU model to identify areas for improvement. Then publish your model to make it available to other applications such as Virtual Agent.") phase, or navigate to [Multi-model Batch Testing](https://servicenow-prod.fluidtopics.net/r6jbvRAegl_J_FRyNHVV4g "Test multiple Natural Language Understanding (NLU) models against a large set of utterances to evaluate the performance of the models. Add test sets, test multiple models, and see test results.").
* To adjust the model's confidence threshold, use the Settings tab on the model's overview page. For more information, see [NLU model settings](https://servicenow-prod.fluidtopics.net/2rtHGKMTjAXLLzKmnstedg "Change your NLU model's name, description, or confidence threshold on the Settings page of the model overview.").
* If you're satisfied with the results of your testing, [Publish your NLU model](https://servicenow-prod.fluidtopics.net/mDp7cp8e~qzgsMC1NnGK5w "Publish your Natural Language Understanding (NLU) model to activate it and make it available for use in other applications that consume NLU.").
{#test-train-nlu-model__ul_x14_dzx_g3b}

*[\>]: and then