Knowledge documents are articles (for use with Article Suggestion) or FAQ documents (for use with FAQ Assist) that make up a knowledge base. Agent Assist analyzes an ongoing conversation between a human agent and an end-user and suggests relevant knowledge documents or FAQ answers to the agent. For more information on creating a knowledge base, see the knowledge base tutorial. This document highlights best practices that help to optimize the quality of suggestions.
We recommend that you remove irrelevant content from your knowledge documents, particularly at the beginning of documents. Article Suggestion shows the first few sententes of a document to human agents as snippets to help the agents understand what the document is about. If the beginning of a document contains with irrelevant information, it could mislead the human agents. Common irrelevant content includes: A navigation bar, last-modified dates, feedback forms, and so on.
Exclude documents with content that is mostly audio-, video- or image-based. Article Suggestion and FAQ Assist process text content only.
If some of your documents are very long (more than 1000 words), we suggest that you break them down into multiple short documents. This helps document suggestion quality and also makes it easier for agents to find answers in the suggested documents.
We also recommend that you include only your most useful and frequently-visited documents in the knowledge base. Which documents are most useful depends on who will be seeing suggestions based on that knowledge base. For example, if you intend to have tier-2 technical support agents seeing suggestions, you should make sure to include documents that contain technical information and are often viewed by the agents receiving the suggestions.
In addition to including useful docs, we also recommend that you exclude documents that are inactive, out of date or very rarely visited. Such documents are unlikely to be useful to agents.