A conversation profile configures a set of parameters that control the suggestions made to an agent. These parameters control the suggestions that are surfaced during runtime. Each profile configures either a Dialogflow virtual agent or a human agent for a conversation. If a human agent is in the conversation, you can also configure which category of suggestions are surfaced (for example, FAQ documents or Article Suggestions.You must create a conversation profile in order to create a conversation with an end-user.
There are two ways to create a conversation profile: Using the Console tutorial workflows, or manually creating a conversation profile in the Console using the Conversation profiles tab. We recommend that you use the Console tutorials as a first option. To use the Console tutorials, navigate to the Agent Assist console and click the Get started button under the feature you'd like to test.
This page demonstrates how to create a conversation profile manually.
Before you begin
- To implement FAQ Assist or Article Suggestion you must create a knowledge base with documents in it.
- To implement Smart Reply you must train a Smart Reply model and manage its allowlist. Agent Assist provides a demo model and allowlist if you want to see how Smart Reply works or test your integration before uploading your own data. See the Smart Reply Console tutorial for more information.
- To implement Summarization you must train a Summarization model. Agent Assist provides a demo model and allowlist if you want to see how Summarization works or test your integration before uploading your own data. See the Summarization Console tutorial for more information.
Create and edit a conversation profile
Navigate to the Agent Assist console.
Select your project from the list and click Conversation profiles in the left sidebar menu.
Click +Create new, on the top right of the page.
In the menu that appears, enter a unique name for your conversation profile in the Display name box.
Choose one or more Suggestion types from the list of available options. For each feature you select, you can select either a model (for Smart Reply and Summarization suggestions) or a knowledge base (for FAQ Assist or Article Suggestion). You are also asked to input Confidence threshold and Maximum suggestions values. Maximum suggestions is the number of response suggestions or knowledge suggestions returned, and the confidence threshold refers to the model's level of confidence that each knowledge suggestion or response suggestion is relevant to the agent's request. A higher confidence value increases the likelihood of relevant responses being returned, but can result in fewer or no responses returned if no available option meets the high threshold value. We suggest that you set the confidence threshold to the following values, depending in which features you're using: Smart Reply=0.01, Summarization=0.01 Article Suggestion=0.44, FAQ Assist=0.4. We recommend that you start with a maximum suggestions value of 3 in all cases. You can experiment with the results using the simulator and make adjustments to the confidence values until you achieve the performance you're looking for.
Choose your Retrieval method (inline or Pub/Sub). Inline enabled by default. Optionally, you can enable Cloud Pub/Sub notifications.
(Optional) Enable Sentiment analysis. This feature analyzes messages from both the agent and end-user to determine emotional intent.
(Optional) Enable handoff to a Dialogflow virtual agent. Virtual agents are automated agents that attempt to resolve customer issues before escalating to a human agent. This toggle enables a pre-existing virtual agent to hand off to a human agent when you run the simulator. You must have already created a Dialogflow virtual agent using the Dialogflow ES Console, and associated that agent with the same project you are using with the Agent Assist Console, in order to simulate this feature.
Click Create. It might take several minutes until the conversation profile is ready to use.
What's next
Test your conversation profile's performance using the Agent Assist simulator.