Overview
In order to make suggestions, Agent Assist relies on data in the form of either documents or conversation transcripts. You must upload your data before you can use Agent Assist. The linked tutorials at the bottom of this page walk you through the steps required to upload your data using the Agent Assist console. You can use the console to configure Agent Assist features and test out how they function.
Data types
Agent Assist uses two types of data to make suggestions to human agents: Conversation datasets, which are collections of conversation transcripts, and knowledge bases, which are collections of knowledge documents (articles or FAQ documents). Agent Assist features analyze a conversation in real time and make suggestions to human agents based on either conversation datasets or knowledge bases.
Smart Reply and Summarization surface suggestions trained on conversation datasets. Smart Reply suggests text responses to agents as they converse with an end-user, and Summarization suggests conversation summaries after an exchange with an end-user has completed. Each model is custom by definition because each conversation dataset is made up of your own conversation transcript data.
The FAQ Assist and Article Suggestion features draw on knowledge bases to make recommendations instead of conversation datsets. Article Suggestion suggests knowledge documents (such as articles) to agents during a conversation. FAQ Assist makes suggestions based on FAQ pairs (an FAQ question and its associated answer) rather than entire articles. You do not need to train a custom model in order to use these features: Agent Assist uses default baseline suggestion models to make suggestions from your knowledge base. If you want to upload your own conversation data to train a custom suggestion model for Article Suggestion, please contact your Google representative. Custom suggestion models are not available for FAQ Assist.
What's next
Create a conversation dataset or a knowledge base.