Create a model to predict record fields

Release version: Australia

Updated March 12, 2026

3 minutes to read

Summarize

Summarized using AI

Summary of Create a Model to Predict Record Fields

This guide provides instructions for creating and training a model to predict fields for case and interaction records within ServiceNow's Task Intelligence for Customer Service. Utilizing this functionality can help streamline processes by reducing handle time through predictive field choices.

Show full answer Show less

Key Features

Role Requirements: mladmin, tiadmin are necessary to initiate this task.
Model Setup: Access the Task Intelligence Admin Console via All > Task Intelligence for Customer Service > Setup and select "Set up model" to begin.
Purpose Definition: Choose the relevant table (Cases or Interactions) and define the trigger for predictions (e.g., new customer email or interaction creation).
Training Inputs: Optionally include attachments in training to enhance prediction accuracy.
Model Training: Select input and output fields to train the model, ensuring at least 500 records are available for effective training.
Prediction Preferences: Choose how predicted fields are managed, including options for autofill, recommendations, monitoring, or turning off predictions.

Key Outcomes

Once the model is trained and deployed, it will provide predictions for case and interaction fields, thereby enhancing efficiency in handling customer interactions. Users can evaluate model performance through sample results before final deployment, ensuring alignment with operational needs.

Create and train a model to predict fields for case and interaction records.

Before you begin: Role required: ml_admin, ti_admin
About this task: The field prediction model is a guide that includes some recommended settings. You can use these settings, such as the recommended input fields, or add your own preferences.
You can create and train multiple field prediction models for cases, case types, and interactions.

Set up your model

Navigate to All > Task Intelligence for Customer Service > Setup to access the Task Intelligence Admin Console.
Select Set up model on this model: Predict field choices to reduce handle time.
This opens the model and displays the first of five pages. Each page in the model asks you questions and helps you select the information you need to build an effective model.

Define the purpose

Select the table and the trigger for the model's predictions.

Model page displaying options for selection like case or interaction data to predict fields as well as the prediction trigger.

You can have the model predict case or interaction fields when a new customer email arrives or when an interaction is created. Base your decision on the data that your model should use to make the predictions.

Table 1. Select the table and trigger
Choose the type of table that has the fields you want to predict	Select the table type: Cases Interactions The model uses data from the selected table to make predictions.
Choose the moment or channel that should trigger a prediction	Select the trigger for the prediction: Cases Emails Interactions
Choose to include optional training inputs	Enable the check box to include attachments when training the model. Email and record attachments can have information that is useful for routing records correctly.

Choose a table type for the model.
Choose the prediction trigger.
If desired, enable the check box to include attachment data.
Include text from email or record attachments if this information is useful for making predictions. The model can evaluate attachment data, along with email or record text, to make predictions.
Select Save & continue.

Train your model

Select the input fields and output fields so your model can learn patterns. Output fields are the fields that you want the model to predict. Input fields are the fields that the model uses to make predictions.

Selecting this information tells the model what to look for during training.

Note:

You can use the recommended settings or select different ones.

Model page used to choose fields and conditions for your model to make predictions.

Provide a name for the model.
Choose the output table and the output fields for the model to predict.
Select conditions to choose a set of records for training.
The selected conditions determine both how the model is trained and act as a filter for the conditions that a record has to meet in order for predictions to be made.
Select the fields in the training data that the model should use to make the predictions (input fields).
Choose the input fields.
Review the resulting number of cases in the training data based on the selected conditions.
The model needs a minimum of 500 records for effective training. If this minimum number isn't available, try selecting different conditions.
Select Launch training.
Training can take some time, particularly if you are training the model on a large amount of data. You can request that the system send you an email when the training is done.

Assess your model

Assess the results from the training and view sample results for the predicted fields. Reviewing the results gives you a preview of how your model will perform after being deployed.

Select the prediction preference for each field. The model provides flexible options to autofill field values, provide recommendations for field values, monitor only, or turn off predictions depending on the sensitivity of those fields.

Model page displaying the estimated number of predicted field values and a sample selection of the test results.

Select a Prediction preference for each enabled field.


Autofill	Adds the best predicted value to the field on the record.
Recommendations	Shows the recommended value in a message below the field.
Monitor only	The system makes the field predictions and stores the information in the Predictions History but does not add any information to the case records.
Turn off predictions	Turns off predictions for the field.

Select View sample results to see sample results for each predicted field.
Select Save & continue.

Deploy your model

Review your choices from the previous pages and information about how the model was trained. Then you can select Deploy to deploy the model.

Model page displaying choices and inputs for review before deploying the model.