Topic AI

Topic AI is an analytics application that uses machine learning and Generative AI to automatically categorize customer interactions. By identifying Customer Intents (what the customer wants) and Agent Actions (what the agent did), Topic AI provides deep insights into customer issues and business processes without manual effort.

This release introduces the C26 Layout, a modern, AI-centric interface with improved navigation and purple-themed styling aligned with the NiCE AI suite.

Key Capabilities

  • Automated Categorization: Uses the TAS 5.7 engine to identify more specific business subtopics (L3) across your business.

  • LLM Validation: Uses LLM as a Judge to automatically calculate the accuracy and coverage of your models.

  • Global Support: Supports whitespace and non-whitespace languages, including Japanese, Chinese, Korean, and Italian.

Prerequisites

  • Licensing: Ensure your tenant has a valid Topic AI license.

  • Permissions: Enable the Topic AI privilege. To access it, navigate to Admin > Roles and Permissions > Profile > AI Studio Settings, and then toggle on Launch Topic AI

  • Analytics Policies: You must enable Topic AI in Analytics Policies to process interaction data. To access it, navigate to Admin > Analytics Services> Analytics Policies, select your policy, and check the Topic AI box.

  • Feature Toggle: The new layout and features are controlled by a feature toggle. Contact your administrator to enable the 26.3 Toggle.

How Topic AI Works

  • The Topic AI enhances the existing out-of-the-box models and generates new intents and actions. You can then publish the model and start using it.

  • Topic AI helps you to:

    • View customer intent hierarchies with related examples and sentences.

    • View agent action hierarchies with associated examples and sentences.

    • Utilize pre-built vertical models optimized for specific business lines.

    • Monitor intent and action percentages based on customer data.

    • Optimize models using customer interactions.

    • Create enriched model versions while preserving originals.

    • Edit enriched models through renaming intents, merging similar ones, and customizing for specific requirements.

    • Continue enriching modified models by incorporating new data to refine performance over time.

    • Lock seasonal or critical subtopics from being removed during model updates.

Features

Key features of the Topic AI include:

  • Select Vertical Model: Choose and view different industry-specific models. You can view the intents and actions for each model and publish the model for use in production. Only one model can be published at a time, and the published model will be used in any NiCE CXone solutions. such as Interaction Analytics (CXone),

    Currently, the Interaction Analytics (CXone) application detects one primary intent and up to eight actions for any new voice or digital interactions. Note that a primary intent is the first intent uttered during the conversation.

  • View Intents: View the intent and its hierarchy, along with the examples and sentences related to the intent.

  • View Actions: View the actions and their hierarchy, along with the examples and sentences related to the action.

High-level features of Topic AI include:

  • Model overview: View the vertical model and optimize the model by adding your company details and uploading Topic AI collections.
  • Manage models: Enrich the base model, edit/modify the enriched model, update model and then publish the model.

 

Language Support

Topic AI is a global solution supporting a wide array of languages.

Language Type

Supported Languages

Whitespace

CXone: English (Default), Dutch, Spanish, Brazilian Portuguese, French, and German (CXone)

Nexidia: CXone Languages + Italian, Korean

 
Non-Whitespace

CXone: Japanese

Nexidia: CXone Languages + Chinese

Limitations

  • Non-Whitespace Models: A single model can support only one non-whitespace language (e.g., you cannot combine Japanese and Chinese in the same model).

  • Code-Switching: The system does not support interactions where the speaker switches languages mid-conversation.

  • Validation Expiry: Validation metrics are based on a data collection that expires after 30 days. If the collection expires, the system will display the last known validation numbers until a new enrichment is performed.