Create topics

  • 8 minutes

Now that the initial structure of the agent has been created, your next step is to begin writing topics. Topics define how the customers interact with the agent, and they typically represent common issues, questions, or tasks that customers might need assistance with. For example, you might create a topic to provide customers with item return instructions.

Each topic consists of two primary elements:

  • Trigger phrases - Phrases, keywords, or questions that are entered by users and relate to a specific issue.

  • Conversation nodes - Define how an agent should respond and what it should do.

You can design topics by customizing provided templates, creating a new one, or using a topic that has been suggested from existing sites. Your agent can have up to 1,000 topics.

Get started with topics

Each created agent includes several predefined topics to help you get started. These predefined topics are separated into two types:

  • Custom – Pre-created user topics include key agent elements such as greeting, conversation start and end, as well as lesson topics that can help you understand simple and complex ways of using nodes to create agent conversations.

  • System - Pre-populated topics that represent common use cases that can occur during an agent conversation.

System topics

System topics represent scenarios that customers are likely to encounter while interacting with your agent. These scenarios might include a topic that maps out when to do when multiple topics are matched or ending a conversation or escalating a conversation to a live agent. System topics have a basic structure already in place, based on what the scenario is. For example, the fallback topic represents the topic that is presented to a user in the event that the agent is unable to identify a topic that answers their question. System topics can be modified based on your needs.

Screenshot of the system tab showing triggers and statuses.

Create topics

You define any other topics by selecting Topics in the side navigation pane and then selecting Add at the top of the page. There are two options for creating topics:

  • From blank: Opens a new blank topic. You create everything from the triggers to conversation flow, etc.

  • Create from description: Uses Copilot to assist you in creating your topic. You provide some basic details about what you want and the topic is created.

Each topic that you define should include some trigger phrases. Trigger phrases are examples of text such as questions or utterances that teach the agent when to respond with this dialog. For example, the following image contains a topic called Store Hours, which is used to provide customers with store location hours based on different scenarios.

Screenshot of creating topics process on navigation pane.

Six trigger phrases have been added, such as What are your hours? and When are you open? These phrases are used to determine when the Store Hours topic should be initiated. The more trigger phrases you add, the more likely it is that the topic is appropriately used. Trigger phrases should be unique to each topic. If you have the same trigger in multiple topics, the agent can't identify which topic to load. After you have added the initial triggers, select Save topic to add the topic to the Topics list. More triggers can be added later as needed.

Once you have defined your trigger phrases, you can define how customers are guided through their conversational interaction with the topic. The authoring canvas is a graphical dialog tree editor that allows you to define agent responses and the overall agent conversation.

Screenshot of the trigger with phrases captioned Trigger phrases will be pre-populated based on what you defined on the previous screen.

Work with conversation nodes

Conversation nodes help define the path that the conversation takes. Conversation nodes can display messages, ask questions, or run actions. You can add these nodes by selecting the plus sign (+) below the node. For example, if you want to provide store hours based on where the customer lives, you would add an Ask a question node to identify which store location they want the hours for.

The following image shows the Ask a question node being used to ask the customer, which stores location they want the hours for. In addition, the customers are provided with two multiple-choice options to choose from: Seattle and Bellevue.

[!div class="mx-imgBorder"] Screenshot of the question node with caption Different node types can be inserted based on what you want to display or capture, and the Options node with caption Options are presented as buttons in the chat conversation window.

Separate conversation paths are created based on the customer's response. In the forked conversation path, each node has automatically checked for Seattle in one path and Bellevue in the other path to take the appropriate next step. More nodes can be added for each path based on what you want it to do.

The following image shows that for each path, a Message node is added that provides the store hours for that specific location.

Screenshot of the message node on each path with store hours.

You now have a simple branching dialog tree and can create more complex versions of this tree. You can incorporate variables, entities, and Power Automate flows.

Use generative answers with search and summarize content

Once you have defined the content that you want to leverage with generative AI, that data will be used for fallback generative answers. However, you might find scenarios where generative answers would be helpful within specific topics of your agent. You can accomplish this by using the Create generative answers node. This special node allows you to specify additional sources that will be searched based on your inputs. Information sources defined in the Generative answers node override sources you specified at the agent level, which functions as a fallback.

These sources include:

  • External sources:

    • Bing Search

    • Bing Custom Search

  • Internal resources:

    • Azure OpenAI on your data

    • Documents uploaded to Dataverse

    • SharePoint

    • OneDrive

    • Custom data (internal or external): Supply your own content from any source. For example, a Power Automate Flow, a Skill, or other source.

To learn more about the different data sources available, see: Generative answers with Search and summarize content.

Generative answers can be added to a topic, by selecting the plus icon to open the new node menu, and under Advanced, selecting Generative answers. This will create a new node called Create generative answers.

You can specify and configure your new data sources. To learn more about configuring the different options, use the links below: