System intents

• Default Fallback Intent—An AI agent's capabilities are inherently limited by the intents that are designed to recognize and respond to. While an enterprise can't anticipate every possible question you might ask, the default fallback intent can help conversations to be on track.

  By implementing a default fallback intent, AI agent developers can ensure that the AI agent gracefully handles unexpected or out of scope queries, redirecting the conversation back to known intents.

  AI agent developers need not add specific utterances to the fallback intent. The agent can be trained to automatically trigger the fallback intent when it encounters known out of scope questions that might otherwise be incorrectly categorized into other intents.

  For example, in a banking AI agent, customers might attempt to inquire about loans. If the AI agent is not configured to handle loan-related inquiries, these queries can be incorporated as training phrases within the default fallback intent. When a customer queries about loans at any point in the conversation, the AI agent recognizes the query as falling outside of its defined intents and triggers the fallback response. This ensures a more appropriate response.

  The fallback intent:

  • shouldn’t have any slots associated with it.
  • must use the default fallback template key for its response.

• Help—This intent is designed to address customer inquiries about the AI agent's capabilities. When customers are unsure of what they can accomplish or encounter difficulties during a conversation, they often seek assistance by asking for help.

  By default, the response for the help intent is mapped to the Help message template key. However, AI agent developers can customize the response or change the associated template key to provide more tailored and informative guidance.

  It’s recommended to convey the AI agent's capabilities at a high level, providing customers with a clear understanding of what they can do next.

• Talk to an agent—This intent enables customers to request assistance from a human agent at any stage of their interaction with the AI agent. When this intent is invoked, the system automatically initiates a transfer to a human agent. The default response template for this intent is agent handover. While there are no UI restrictions on changing the response template key, altering it won’t affect the outcome of the human handover.

Small talk intents

All newly created AI agents include four predefined small talk intents to handle common customer greetings, expressions of gratitude, negative feedback, and farewells:

• Greetings
• Thank you
• The AI agent wasn’t helpful
• Goodbye

Entity types

Webex AI Agent Studio offers 11 prebuilt entity types to capture various types of user data. You can also create any of the following custom entities.

Custom Entities

These entities are configurable and allow developers to capture use-case specific information.

• Custom list—Define lists of expected strings to capture specific data points not covered by prebuilt entities. You can add multiple synonyms against each string. For example, a custom pizza size entity.

• Regex—Use regular expressions to identify specific patterns and extract corresponding data. For example, a phone number regex, as in 123-123-8789

• Digits—Capture fixed-length numerical inputs with high accuracy, especially in voice interactions. We use this as an alternative to Custom and Regex entity types in nonvoice interactions. For example, define a length of five to detect a five-digit account number.

• Alphanumeric—Capture combinations of letters and numbers, providing accurate recognition for both voice and nonvoice inputs.

• Free form—Capture flexible data points that are difficult to define or validate.

• Map location (WhatsApp)—Extract location data shared by you on the WhatsApp channel.

System Entities

You can edit created entities from the entities tab. Linking entities to an intent annotates your utterances with detected entities as you add them.

Entity roles

Entity roles are crucial when you need to collect the same entity multiple times within a single intent. By assigning distinct roles to the same entity, you can help the AI agent understand and process user input more accurately.

For example, to book a flight with a layover, you can create an Airport entity with three roles: origin,destination, and layover. By annotating training utterances with these roles, the AI Agent can learn expected patterns and seamlessly handle complex booking requests.

Only Mindmeld (custom and system entities) and Rasa (custom entities only) support entity roles. Administrators must select the Entity roles check box under the advanced settings of the NLU engine selector dialog box to enable entity roles.

System responses

The following pre-configured system responses are available for the scripted AI agent. You can customize the messages for the default system responses. However, you cannot delete these responses.

• Welcome message

• Response suggestion

• Partial message

• Fallback message

• Entity suggestion

• Agent handover

Small-talk responses

You can customize and delete the following small-talk responses:

• Goodbye

• Greetings

• Help message

• Not helpful

• Thank you

The supported channels for which you can configure the responses are Web chat (default), Apple Messages for Business, Messenger, RCS, SMS, Voice, WhatsApp.

Supported response types for channels 

The following table outlines the supported response types for different channels and their configuration details: