Virtual Agent and NLU Workbench integration

  • Release version: Xanadu
  • Updated August 1, 2024
  • 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 Virtual Agent and NLU Workbench integration

    ServiceNow Virtual Agent administrators can now integrate and manage Natural Language Understanding (NLU) models directly within the Virtual Agent Designer interface. This integration streamlines the process of creating, updating, and applying NLU models and intents to Virtual Agent conversation topics, enhancing conversational accuracy and flexibility.

    Show full answer Show less

    For customers using Now Assist in Virtual Agent, existing NLU topics can be migrated into new Large Language Model (LLM) topics through the topic migration feature, allowing a smooth transition to improved conversational capabilities.

    Integration Setup and Roles

    • Administrators must first create NLU models and intents in the NLU Workbench using the admin or nluadmin roles.
    • In Virtual Agent General Settings, admins must enable NLU, select the NLU service provider, and activate language-specific models if needed.
    • Within Virtual Agent Designer, admins assign NLU models and intents to conversation topics, configure topic switching behavior, and set entity extraction on input controls.
    • Dialog Acts can be optionally enabled (English only) to allow Virtual Agent to respond dynamically to user changes mid-conversation with response types such as Modify, Affirm, and Negate.

    Managing NLU Models within Virtual Agent Designer

    • Administrators with virtualagentadmin or admin roles can update intent utterances, and train, test, and publish NLU models directly from the Virtual Agent Designer interface.
    • Publishing a topic automatically maps and associates the selected NLU model and intent to that topic, enabling seamless deployment.
    • Important publishing prerequisites include having a trained model (not in training), ensuring the intent is enabled, and confirming the model isn’t already published with another Virtual Agent topic.

    Practical Benefits for ServiceNow Customers

    • This integration simplifies the lifecycle management of conversational AI models, reducing the need to switch between multiple interfaces.
    • It enables more precise intent detection and entity extraction within Virtual Agent conversations, improving user experience.
    • Dialog Acts support flexible, context-aware responses, making conversations feel more natural and adaptive.
    • Seamless publishing processes ensure that updated models and topics are quickly available for end users without complex manual steps.

    Virtual Agent administrators can access and update their NLU models from within the Virtual Agent Designer user interface.

    Diagram of NLU and Virtual Agent authoring.

    Note:
    If you have Now Assist in Virtual Agent, you can continue to use your existing NLU topics and migrate them into new LLM topics using the topic migration feature within Virtual Agent Designer. For more information on topic migration, see Migrate NLU topics to LLM topics.

    Integration setup tasks, roles, and details

    As Virtual Agent administrators create and configure their conversation topics, they must first create their NLU model and its associated intents in the NLU Workbench. This action requires they use the NLU Workbench and the admin or nlu_admin role.

    In addition, they must also complete the following tasks in Virtual Agent General Settings.
    • Enable NLU.
    • Select the NLU service provider.
    • If using language-specific NLU models, enable the languages for those models.
    Virtual Agent administrators must also apply their NLU model to a conversation topic by completing the following tasks in Virtual Agent Designer:
    • In Topic Properties, select the NLU model, the NLU intent, and the topic switching behavior.
    • For input controls used in the topic flow, set the NLU properties for entity extraction.

    Optionally, admins can activate Dialog Acts to enable Virtual Agent to respond flexibly when users make a modification in mid-conversation. Currently available response types are Modify, Affirm, and Negate, based on the last 5 exchanges in the conversation. Dialog Acts can be configured for English only, in Topic Properties. For more information see Dialog Acts for Virtual Agent.

    After the NLU model is complete and associated with a Virtual Agent conversation topic, administrators with the virtual_agent_admin or admin role can do the following from within the Virtual Agent Designer user interface.
    • Update NLU intent utterances.
    • Train, test, and publish the NLU model.

    For more information, see Natural Language Understanding (NLU) topic discovery in Virtual Agent.

    Publishing topics from Virtual Agent

    When your model is published in the NLU Workbench, it's ready to use in Virtual Agent Designer. When editing a topic, click the Properties tab to select a model and intent to map to that topic.

    Natural Language Understanding section of the Properties tab of Virtual Agent Designer.

    When you click Publish, the model and intent are mapped to that topic and published seamlessly.

    Publishing a topic with a mapped intent fails for the following reasons:
    • Model isn't trained, or training is in progress.
    • The last trained model is already published with a VA topic.
    • The intent is not enabled in the model.