View Task Intelligence prediction history

  • Release version: Yokohama
  • Updated January 30, 2025
  • 2 minutes to read
    • Summarize
    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 View Task Intelligence prediction history

    This feature allows ServiceNow customers to view and analyze historical prediction data generated by Task Intelligence models. Prediction results are stored in thePredictor Result [mlpredictorresults]table, which records details such as predicted language, predicted and actual values, and prediction confidence. This data helps monitor the performance of machine learning models used in customer service case categorization.

    Show full answer Show less

    Access and Usage

    Users with the roles Task Intelligence admin (sncsmmltask.tiadmin) and Customer Service manager (sncustomerservicemanager) can access the prediction history by navigating to All > Task Intelligence for Customer Service > History. The History module lists the prediction records for review.

    Key Details about Prediction Records

    • Records are created for every prediction event, including testing behaviors where predictions are not applied to cases.
    • When a case’s status changes to Closed, categorization feedback captures key values if the predicted value differs from the actual case field value and if case categorization is active.
    • Prediction solutions are trained using different data sources depending on the categorization type:
      • Email categorization uses the Email [sysemail] table.
      • Case categorization uses the Case [sncustomerservicecase] table or its extensions.
      • Interaction categorization uses the Interaction [interaction] table.
    • Prediction feedback is triggered by a flow and stores multiple data points in the Predictor Result table, including predicted and final input/output values and confidence percentages.
    • For email categorization, final input values remain empty because email content is not expected to be modified by agents, but final output values are stored once the case closes.
    • For record categorization, input field values are stored in the final input values and output values are updated upon case closure.

    Preventing Duplicate Entries

    Duplicate entries can occur in the Predictor Result table for email sentiment replies. To prevent this, it is recommended to deactivate the Predict Sentiment for Cases inbound email action. This helps maintain clean and accurate prediction history data.

    View details about historical predictions in the Predictor Result [ml_predictor_results] table.

    The Predictor Result table includes entries for each prediction result, including the predicted language, predicted and actual values, and prediction confidence. You can use the data in this table to monitor model performance.
    Note:
    Records are created in the Predictor Result table regardless of the model prediction behavior, including the Testing behavior that doesn’t add prediction values to the case.

    Viewing prediction data

    Navigate to All > Task Intelligence for Customer Service > History to view the Predictor Result list.

    Users with the following roles can access the History module:
    • Task Intelligence admin [sn_csm_ml_task.ti_admin]
    • Customer Service manager [sn_customerservice_manager]
    Categorization feedback captures values when the status of a case changes to Closed. It captures these values if:
    • Case categorization is active.
    • The predicted field value is different from the case field value.

    Categorization feedback works for record categorization solutions and for solutions with and without attachments. Email categorization solutions are trained using data from the Email [sys_email] table. Case categorization solutions are trained using data from the Case [sn_customerservice_case] table or from tables that extend the Case table. Interaction categorization solutions are trained using data from the Interaction [interaction] table.

    Categorization feedback is triggered by a flow.
    • When email information is used for prediction, the final input values remain empty.
    • When attachment information is used for prediction, the attachment sys_ids are not included in the final input values.
    Categorization feedback stores the following values in the Predictor Result table:
    • Predicted Output Value Name
    • Predicted Input Value Names
    • Predicted Output Value
    • Predicted Input Values
    • Final Input Value
    • Final Output Value
    • Predicted Confidence (percentage)
    If the values for a case have been predicted using an email categorization solution:
    • The Final Input Value remains empty because it is not expected that the agent will modify the values of the email, such as subject and body.
    • The final value in the Predicted Output Value is stored in the Final Output Value once the case is closed.
    If the values for a case or interaction have been predicted using a record categorization solution:
    • The values from the fields defined in Predicted Input Value Names are stored in Final Input Value.
    • The final value in the Predicted Output Value is stored in the Final Output Value once the case is closed.

    Email sentiment replies

    Duplicate entries can be posted in the Predictor Result table for email sentiment replies. To help prevent these duplicate entries, deactivate the Predict Sentiment for Cases inbound email action. For more information, see Inbound email actions.