Using the generative replies block in conversation flows for advanced AI agents

What's my plan?

Add-on

AI agents - Advanced

Zero-training AI agents and agents with agentic AI can use generative replies in conversation flows. They are activated through the generative replies block in the dialogue builder. The input for a generative reply consists of the latest customer message along with various configuration settings.

Depending on the specific scenario triggered by the generative replies block, the conversation will progress either with a generative reply or by following the predefined dialogues associated with other scenarios.

The article contains the following sections:

• About the generative replies block
• Showing which help center articles were used to generate a reply
• Defining a custom question for the generative replies block
• Additional considerations for generative replies for email AI agents

About the generative replies block

Every advanced AI agent comes with a preconfigured generative replies block within the generative replies system reply. Depending on when your AI agent was created, this system reply might be called uGPT reply.

Additionally, you can add a generative replies block to any other reply by selecting it from the block menu.

The generative replies block serves to respond with informational content generated from an imported knowledge source. It has configuration options to specify the context that generative replies should consider when generating a response. You can also use actions and custom resolution states in a generative replies block.

When you add a generative reply block in a dialogue, the following scenarios are automatically added as children of the block and must be configured:

• Response generated
• Escalation needed

Response generated

In the response generated scenario, an AI agent successfully generates a response to a customer message by considering both the information from the knowledge source and the specific query.

Information is retained in two session parameters:

• ugptResponse: Retains the response from the AI agent, which can be used immediately in the conversation or saved for later reference.
• dataSources: Retains up to five of the knowledge base articles used as context for the generated response.

With each new generative reply, the content of these session parameters is overwritten. This ensures that the response captured is always the most recent.

You can choose from two options for integrating a generative reply into the conversation:

• Share with the customer: The AI agent immediately sends the generated response to the conversation.

  

• Save response only: The AI agent does not send the response directly to the conversation, and only retains it in the session parameter ugptResponse. The dialogue continues with other blocks and logic.

  

  The response can be used for testing or evaluation purposes. It can also be shared with the customer later using an AI agent message block along with the session parameter using the reference {{ugptResponse}}.

  

Escalation needed

The escalation needed scenario is activated when the AI agent identifies that the customer wants to speak with a human. In this scenario, information is not saved to the ugptResponse and dataSources session parameters.

To handle this scenario, continue the dialogue by incorporating further blocks, such as another reply or an escalation block.

Configuring the response fallback behavior

By default, both of the scenarios of the generative reply block (Response generated and Escalation needed) are directly linked to their respective system replies. You have the option of building a dialogue branch for either or both of them by setting the Response fallback behavior toggle to off in the Details pane.

• Not understood: The AI agent is unable to find relevant articles or data to generate a response. By default, this scenario links to the default reply.
• An error occurred: The generative replies service is either unreachable, returns an error, or encounters another issue within the system. By default, this scenario links to the Technical error system reply

Showing which help center articles were used to generate a reply

The dataSources session parameter, which is returned with the generative reply, can be used to provide the customer with access to the articles that a messaging AI agent’s response was based on. Using carousels to provide this content can improve the customer experience by adding a visual element.

Note: Generative procedures used by AI agents with agentic AI don’t currently support rich-text formatting such as buttons or carousels.

To create a carousel of data sources

1. In the AI agents - Advanced add-on, in the top-right corner, use the AI agent drop-down field to select the AI agent you want to set up the carousel for.
2. In the left sidebar, click Content > Use cases.
3. Click the use case where you want to put the carousel.
4. Select the Replies tab.
5. Select the reply that contains the dialogue you want to add a carousel to.

   The dialogue builder opens.

6. Click the plus (+) icon to add a conditional block in the appropriate location.
7. In the conditional block, set the variable ugptClassifiedTask.
8. As a condition, set the operator to is and include the value question answering.
9. Click the plus (+) icon to add an AI agent message to the fallback of the conditional block.

   

10. Click the plus (+) icon to add a carousel block to the is question answering condition block.

    The carousel block appears along with the first card block and a Details pane.

11. In the Details pane, click Convert to dynamic carousel and enter dataSources as the array type parameter.
12. In the template card block, enter text or parameters such as %imageurl, %title and %description in the respective fields.
13. In the template card button block, select Button with external link and enter text or a parameter such as %buttontext.
14. In the Details pane, deactivate the fallback block by setting the Fallback toggle to off.

    Omitting the fallback allows the conversation to restart if the customer asks another question.

    

Defining a custom question for the generative replies block

When editing the generative replies block, you can define a custom question which overwrites the customer's last message, such as a button push, allowing you to provide contextually relevant answers. The field also allows you to use parameters to include information about the customer to make a question more relevant, like their subscription status or tier. For more, see Creating a custom question in the generative replies block in conversation flows for advanced AI agents.

Additional considerations for generative replies for email AI agents

The generative replies functionality is the same for messaging and email AI agents, except for following differences:

• AI-generated responses require additional language for email AI agents. For email AI agents, you must write additional response text to frame the AI agent’s reply. For example, including the text “Our AI agent has suggested the following answer:” as a lead in to the text of the generative reply.
• No escalation options for email AI agents. For email AI agents, generative replies blocks don’t offer an escalation option. Instead, use a fallback block to address scenarios in which the AI agent can't answer the customer’s question with a generative reply.