Skip to main content

Use Knowledge in Agentforce for Service

Learning Objectives

After completing this unit, you’ll be able to:

  • Understand how Agentforce for Service uses the Agentforce Data Library.
  • Explain key differences between Agentforce Knowledge vs Einstein Bots knowledge.

Trailcast

If you'd like to listen to an audio recording of this module, please use the player below. When you’re finished listening to this recording, remember to come back to each unit, check out the resources, and complete the associated assessments.

Agentforce, You’re Grounded! (in Data)

Imagine equipping an agent with the ability to answer questions in real-time by retrieving information from your Knowledge base. That’s exactly what Agentforce does, effortlessly.

Agentforce Data Library vs. Article Answers (Einstein Bots)

Adding an Agentforce Data Library enhances accuracy and builds customer trust in the responses provided. The Data Library supports both Knowledge articles (records and fields) uploaded files, web search, and custom retrievers. Why is this such a significant improvement?

Previously, Einstein Bots relied on highly structured data using article answers, requiring natural language processing (NLP) to train the bots to be able to understand intent and retrieve information from Knowledge articles. Manual mapping was needed to link questions (which also had to be created with various variations) to specific fields within each article. Only published articles with populated fields could be indexed and searched, which limited the answers to those mapped fields. As a result, the bots could only respond to specific, predefined questions and couldn’t go beyond that scope. Depending on the Knowledge base size, a build could take up to 30 minutes or more. It was akin to conversing with someone who could only answer questions they were specifically trained for and had to be phrased in very particular ways.

The Power of Unstructured Data

Agentforce harnesses the power of large language models (LLMs) to process unstructured data, supported by the Agentforce Data Library. Unstructured data often includes chat transcripts, PDFs, audio and video files, legal documents, and large text files like books. In Salesforce, this type of data can be sourced from Knowledge articles or PDF attachments when integrating an Agentforce Data Library.

While unstructured data lacks a consistent format, LLMs, like the one used by Agentforce, can effortlessly process and search this type of information. This means that an article or PDF can be uploaded to the Agentforce Data Library. There’s no need to map specific fields for data retrieval, refresh Knowledge articles, or wait for a build. Most importantly, Agentforce doesn’t need to be trained on how to answer specific questions, as it’s built to naturally understand human interactions.

The data library uses grounding with retrieval augmented generation (RAG) to index Knowledge articles and attachments, ensuring it retrieves the most relevant and up-to-date information. When a data library is added in Agentforce, Salesforce automatically builds a RAG-powered solution to retrieve unstructured (and structured data) from Knowledge articles and PDFs. With this indexed data, Einstein can quickly fact-check responses against specific policies and information. So it’s not simply “guessing” its way to the right answer. All this to say, when a customer asks a question, Agentforce accesses the relevant information from the data library and delivers a response that feels natural and conversational.

A data library in Agentforce.

In Agentforce Builder, users can choose to use an existing Agentforce Data Library or create a new one. Just a heads-up: Data Cloud needs to be activated before setting up a Data Library. Once that’s done, users can upload attachments or link to existing Knowledge articles—and that’s all it takes for Agentforce to start working its magic. It’s that easy.

In Summary

Agentforce for Service aims to improve the productivity and effectiveness of service agents by using AI to manage and interact with unstructured data more efficiently. By adding topics, actions, and instructions, tasks are clearly defined with guidelines that direct agents on when and how to engage with customers—and when not to. Now that you understand how Agentforce for Service works, check out Quick Start: Build a Service Agent with Agentforce to get hands on!

Resources

Share your Trailhead feedback over on Salesforce Help.

We'd love to hear about your experience with Trailhead - you can now access the new feedback form anytime from the Salesforce Help site.

Learn More Continue to Share Feedback