---
sourceDocument: Zurich Customer Service Management
sourceDocumentLink: https://www.servicenow.com/docs/r/zurich/customer-service-management

 Release :

    - zurich

ft:locale :

    - en-US

ft:publication_title :

    - Zurich Customer Service Management

ft:clusterId :

    - csm

bundleId :

    - csm

workflow :

    - Customer and Industry


---

# Machine learning model setup and behavior

# Machine learning model setup and behavior {#ariaid-title1}

Release version: Zurich  
Updated July 31, 2025  
![](https://www.servicenow.com/docs/portal-asset/ico-clock) 2 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 Machine learning model setup and behavior

This guide explains how ServiceNow customers can set up and configure machine learning (ML) models to predict field values and analyze sentiment for customer service cases.
It covers creating, training, and applying models using historical data to improve case categorization and sentiment detection, ultimately enhancing customer service efficiency and accuracy.
Show full answer Show less  

## Field Prediction Model Setup

* **Role Required:** Users with the `mladmin` role can create and train models via the Task Intelligence Admin Console.
* **Training Data Sources:** Models can be trained using data from the Email (`sysemail`), Case (`sncustomerservicecase`), Interaction (`interaction`), and any tables extending the Case table.
* **Model Inputs and Outputs:**
  * **Output fields:** Fields you want the model to predict (e.g., Category, Priority).
  * **Input fields:** Fields used for predictions (e.g., email subject and body text). You can use recommended inputs or customize them.
* **Attachment Support:** The model can also use text from supported email or case attachments. The system checks attachments for supported content types and parses text accordingly before inputting it into the model. Unsupported attachments are ignored.
* **Supported Content Types:** Defined in the system property `sncsmmltask.categorization.allowedcontenttypes`, ensuring only appropriate file types are processed.
* **Multi-language Support:** The categorization model supports multiple languages including text in attachments. Detected language results are stored in the `mlpredictorresults` table.

## Case Sentiment Model Setup

* **Pre-trained Model:** The sentiment model is pre-trained on extensive customer communication data including emails, case descriptions, and comments to recognize typical sentiment patterns.
* **Input Text Sources:**
  * Initial email subject and body for sentiment prediction at case creation.
  * Subsequent email bodies and case comments update sentiment predictions.
  * Case short description and description fields are also used for initial sentiment detection.
* **Supported Tables:** Sentiment analysis can be configured on the Case table or tables that extend from it (supporting one level of extension).  
Set up models to predict field values and sentiment for customer service
cases.

## Training a model {#csm-task-intel-model-setup-behavior__section_gfc_lbn_g5b}

Training a machine learning model is when the model learns patterns in past data to make
predictions for new data. Models are trained using a lot of data so that they can learn patterns
and the large data set makes the learned patterns statistically significant.

## Setting up a field prediction model {#csm-task-intel-model-setup-behavior__section_esq_2dn_g5b}

Users with the ml_admin role can create and train a machine learning model to predict
field values from the [Task Intelligence Admin
Console](https://www.servicenow.com/docs/aMxE6h_yLicrO5Ddacaeyw "Use the Task Intelligence Admin Console to create, train, and deploy machine learning models that predict different types of information for case and interaction records.").  
Using the field prediction model as a starting point, you can choose the training data set that the model learns from. The model can be trained using data from the following tables:

* Email \[sys_email\] table
* Case \[sn_customerservice_case\] table
* Tables that extend the Case table
* Interaction \[interaction\] table
{#csm-task-intel-model-setup-behavior__ul_ppc_swm_zyb}Models can also be trained using data from email or case attachments.  
You then direct the model to learn a pattern between two types of fields from that data:

* Output fields are the fields that you want your model to predict. For example, the Category and Priority fields for cases.
* Input fields are the fields that the model uses as a basis for predictions. For example, text in the subject and body of an email.
{#csm-task-intel-model-setup-behavior__ul_ent_3m3_d5b}

You can use the recommended input fields or you can modify these fields and add your own
preferences.  
If the model is configured to use text from attachments, the system performs the following steps when a case or interaction is created:

* The system checks the record for attachments with supported content types and file extensions. It ignores the attachments that have unsupported file extensions.
* If the record has attachments in a supported format, the system parses the text and sends it as an input to the categorization model, along with text from the input fields.
* If the record does not have attachments, or no attachments in a supported format, the system sends text from the input fields to the categorization model.
{#csm-task-intel-model-setup-behavior__ul_y24_2hl_d5b}

Supported content types and file extensions are stored in the
sn_csm_ml_task.categorization.allowed_content_types system
property. For more information, see [Components installed with
Task Intelligence for Customer Service](https://www.servicenow.com/docs/40pUFf3JG3QPhlf7yaM1ZA#case-categorization-components__section_bqd_lqt_xrb).

## Supporting multiple languages {#csm-task-intel-model-setup-behavior__section_jrk_533_h5b}

Categorization supports multiple languages including attachments, if the models are
configured to include attachments. The categorization model returns the predicted
language and stores it in the Detected Language field in the
Predictor Result \[ml_predictor_results\]
table.

## Setting up a case sentiment model {#csm-task-intel-model-setup-behavior__section_vpq_2dn_g5b}

The case sentiment model is pre-trained with a large data set to learn communication patterns. This data comes from customer emails and case descriptions and comments and reflects typical communication between agents and customers.

* Email: The model uses the text in the subject and body of the initial email to predict sentiment when the case is created. Text from the body of subsequent emails is used to update the prediction.
* Cases: The model uses the text in the case short description and description to predict sentiment when the case is created. Comments added to the case are used to update the prediction.
{#csm-task-intel-model-setup-behavior__ul_tbz_kt3_d5b}  
The case sentiment model supports case types. When setting up a sentiment model, you select the table on which to run sentiment analysis. You can select:

* The Case table
* Tables that extend the Case table

{#csm-task-intel-model-setup-behavior__ul_iws_qzg_c5b}  
Note:  
The sentiment analysis feature supports one level of custom extension from the Case table.