How Document Intelligence for Customer Service works

  • Release version: Yokohama
  • Updated January 30, 2025
  • 5 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 How Document Intelligence for Customer Service works

    Document Intelligence for Customer Service automates the extraction of relevant data from email and case attachments, populating fields on case records to streamline case management. This capability helps ServiceNow customers efficiently process and manage documents related to customer service cases by leveraging machine learning models and optical character recognition (OCR).

    Show full answer Show less

    Use Cases and Components

    A use case defines the structure and extraction logic for a specific document type. It comprises:

    • Fields: Specifies which data points to extract and map to Case table fields.
    • Field Groups: Groups related fields such as tables or checkbox lists to improve extraction accuracy.
    • Document Tasks: Collections of attached documents used to train the model by confirming or correcting extracted values, helping improve future predictions.
    • Integrations: Conditions that determine when to trigger extraction, based on case attributes like category. The system selects an appropriate use case based on these conditions.

    Users with appropriate roles (tiadmin or tianalyst) can create and manage use cases. If no specific use case exists for a case type, the system searches the hierarchy to find a suitable use case.

    Extraction Modes

    Extraction modes control how data is extracted and handled:

    • Recommendation: Provides AI-generated suggestions for fields that agents must review and confirm or correct.
    • Auto-fill: Automatically fills fields when AI confidence is sufficient; agents still review all fields.
    • Fully automated (Straight through processing): Automatically extracts and populates all fields without agent review when confidence exceeds thresholds, ideal for high-volume or well-understood documents.

    Confidence thresholds for auto-fill and fully automated modes can be configured to suit organizational needs.

    Interface and Field Identification

    Extracted fields are clearly marked for agents:

    • In CSM Configurable Workspace, fields display an AI icon and labels with additional context via information icons.
    • In Core UI, fields show a “Predicted from DocIntel” message below the extracted values.

    For fully automated mode, values and messages are added before the agent views the case. For other modes, agents review and submit extracted values via the Document Intelligence workspace.

    Performance and Notifications

    Extraction processing time depends on attachment volume and data center location. The system notifies users if extraction fails, is slow, or if attachments are missing or unexpected, ensuring transparency in processing status.

    How It Works

    • Upon case creation, Document Intelligence checks for attachments of supported types (configurable via system properties).
    • The system selects the appropriate use case based on case type and integration conditions.
    • A document task is created and sent to the prediction model, which uses OCR to extract data.
    • The task status updates to “Done” once processing completes.
    • Depending on extraction mode, extracted values are either automatically added to the case or presented to agents for review in the Document Intelligence workspace.
    • Agents can validate, correct, and submit extracted data to continuously improve model accuracy.

    Document Intelligence for Customer Service performs a series of steps to extract relevant information from email and case attachments and add that information to fields on the case record.

    Use cases

    A use case, previously known as a task definition, is a template that is used to define the structure of a type of document you want to process. A use case is made up of the use case record and its related fields, field groups, integrations, flows, and all the related machine learning (ML) models. For more information, see Set up document extraction use cases.

    Table 1. Document Intelligence use case components
    Component Description
    Fields Identifies a field that you want to extract a value for. A use case includes the list of fields to extract from case and email attachments and a mapping of the extracted values to fields in the Case table.
    Field groups Field groups help Document Intelligence extract data from documents with tables, check box lists, and other logical groupings of fields.
    Document tasks

    A document task includes one or more attached documents that are used to train the use case to identify and extract the correct information. Document tasks are stored in the Document Tasks related list on the Use Case form.

    The Document Intelligence feature creates these tasks for each case with valid attachments. Each task trains the model. In CSM Configurable Workspace, agents can view these tasks, confirm or correct extracted values, and continue to train the model.
    Integrations Select conditions that tell the Document Intelligence feature when to extract values. The system creates tasks for those cases that meet the specified conditions.

    For example, you can add a condition on the Category field where [Category] [is] [Credit Card].

    At run time, the system evaluates the active use cases and identifies the use case to use for the case based on the integration. The first active use case that matches the integration for the case or case type is selected.

    If there are use cases for the case or case type but none that match the integrations, then no use case is selected.

    You can create one integration for a use case. If a use case has more than one integration, the system uses the latest one.
    Note:
    When creating an integration, make sure that the Create Flow check box remains unchecked.

    Users with the ti_admin role or the ti_analyst role can create use cases for cases and for case types. At run time, the system identifies the correct use case based on the case or case type when a case is created and has a valid attachment. If there is no use case for a specific case type, the system goes up in the hierarchy chain until it finds a use case to use.

    Use cases are stored in the Use Case (di_task_definition) table.

    Extraction modes

    A use case includes an extraction mode that determines how the data is extracted in the document task and how the task is processed. The mode changes the behavior of the fields in the Document Intelligence workspace.

    DocIntel uses the following extraction modes

    Table 2. Document Intelligence use case extraction modes
    Extraction mode Definition
    Recommendation

    Provides recommendations for the extracted fields in the Document Intelligence workspace. Agents can choose the recommended value for a field or manually enter a value. All fields must be reviewed.

    Recommendations are ordered based on how confident the AI is in the prediction. As Document Intelligence continues processing your documents, recommendations can improve over time.
    Auto-fill

    Adds values for the extracted fields in the Document Intelligence workspace. All fields must be reviewed.

    Auto-fill works only if the AI has enough confidence to make the prediction. You can change the confidence threshold by updating the Auto-fill threshold field in the use case.
    Fully automated

    (Straight through processing)

    DocIntel extracts the data for all fields and processes the document task if the confidence scores for all required fields are above the defined confidence threshold.

    DocIntel becomes more confident over time, as it processes more and more documents. Choose Fully automated mode for frequently processed documents or if you’re confident in the system.

    Extraction labels

    In CSM Configurable Workspace, the fields on the Case form that contain Document Intelligence predicted values are identified with an AI icon (AI icon) and label. These fields also include an information icon that displays a message with additional context about the predicted values.

    In Core UI, the fields on the Case form that contain Document Intelligence predicted values are identified with the message Predicted from DocIntel.
    • For Fully automated mode, the message appears below each extracted field returned by the task. The values and messages for each field are added to the case before the agent views the case record.
    • For the other extraction modes, the message appears below the extracted fields after the agent reviews the extracted values in the Document Intelligence interface, confirms or corrects values, and submits those values.

    Predicted field values

    In CSM Configurable Workspace, the fields on the Case form that contain Document Intelligence predicted values are identified with an AI icon (AI icon) and label. These fields also include an information icon that displays a message with additional context about the predicted values.

    In Core UI, the fields on the Case form that contain Document Intelligence predicted values are identified with the message Predicted from DocIntel.

    Performance

    The Document Intelligence for Customer Service feature needs time to identify cases with attachments, scan the attachments, and make recommendations for case fields. The response time also depends on factors such as case with attachment volume and data center location.

    The system displays notifications to the user for the following scenarios:
    • If the extraction fails.
    • If the extraction is too slow.
    • If there are no attachments.
    • If there are unexpected attachments.

    How it works

    When a case is created, Document Intelligence for Customer Service checks to see:
    • If the case has one or more attachments.
    • If the attachment types are specified in the sn_csm_ml_task.case.docintel.parsing_supported_types system property.
    If yes, the feature:
    • Identifies the right use case to use based on the case or case type and the use case integrations.
    • Creates a task using the use case, the sys_id of each attachment, and the case reference.
    • Sends the task, attachment sys_ids, and case reference as inputs to the prediction model.
    • Uses optical character recognition (OCR) solutions to extract data from the documents.
    • Sets the status of the task to Done once the solution has completed.
    • If the extraction mode in the use case is set to Fully automated, the extracted values are added to the case.
    • If the extraction mode is set to Autofill or Recommendation, the agent can validate the extracted values in the Document Intelligence workspace.
    The agent can open a case and review the predicted fields or review the prediction task by selecting Review in DocIntel and opening the Document Intelligence workspace interface in a separate tab. From this interface, agents can:
    • Review each predicted field.
    • Confirm correctly predicted values.
    • Update incorrect or missing values.
    • Submit the changes.