Exercise 6 (Optional): Ground the Agent with Unstructured Data

This exercise goes beyond the core Merchant Account Manager build. The agent you built answers from structured CRM data, so no knowledge base is required. This optional exercise shows how you'd add Retrieval Augmented Generation (RAG) if your account managers also needed answers from unstructured documents such as partner playbooks, onboarding guides, or policy PDFs, the kind of question that lives in documents rather than records.

Step 1: Create a Data Library

You create the Data Library in Setup first, then attach it to the agent in the next step.

1. In Setup, search for Data Libraries and select Data Libraries.

2. Click New Data Library.

3. Enter the following values:

   Field	Value
   Name	Account Manager Playbook
   API Name	Keep default
   Description	Partner management playbooks, SLAs, and onboarding guides for Pronto account managers.

4. Click Save.

5. For Data type, select Files.

   TIP

   Data Libraries support multiple source types:

   • Files: Ingest documents like PDFs and index them for retrieval (best for playbooks, policies, FAQs).
   • Knowledge: Index Salesforce Knowledge articles so retrieval can ground answers in your Knowledge base.
   • Web: Crawl and index website pages (for example, a help center) to ground answers in published web content.

6. Click Upload Files and select the partner-management documents you want to ground on (playbooks, SLAs, onboarding guides). If you completed the RAG chapter, you can reuse the same Pronto support PDFs:

   • pronto-customer-support-playbook.pdf
   • pronto-delivery-and-order-lifecycle-faq.pdf
   • pronto-promotions-membership-and-gift-certificates-terms.pdf
   • pronto-merchant-and-storefront-standards.pdf

7. Click Save to confirm the file upload.

8. Once the files have uploaded, click Done.

   TIP

   This screen can occasionally throw an error after uploading files. If it does, refresh the page to confirm whether your file upload was successful.

Step 2: Add the Data Library to your agent

1. Back in Agentforce Builder, in the Explorer, select Data, then Data Library.

2. Select your created Account Manager Playbook Data Library.

Step 3: Add a knowledge subagent

1. In Agentforce Builder, click the + sign besides Subagents in the explorer.

2. Select Create new subagent.

3. Set the following values:

   Field	Value
   Name	Partner Playbook Q&A
   Description	Answer questions about internal support processes and policies (case creation, escalation, refunds and credits, storefront standards) using the Account Manager Playbook data library, and defer to the analytical topics for anything data-driven.

4. Click Create and Open.

5. Paste the following into the Instructions section:

   Answer questions about internal support processes and policies (case creation, escalation rules, refunds and credits, customer verification, and storefront standards) using the Account Manager Playbook data library. Ground every answer in the retrieved documents and cite them when available. Defer to the analytical topics for anything data-driven (account health, reviews, marketing events, cases), and route process and policy questions here rather than to those data topics.

6. Click Save.

Step 4: Add the knowledge action

1. In the Explorer, select the Partner Playbook Q&A subagent.

2. Under Actions Available For Reasoning, open the Select action dropdown and choose Add from Asset Library.

3. Search for and add Answer Questions with Knowledge (or the equivalent RAG action), and point it at the Account Manager Playbook data library.

4. Click Add to Agent.

5. Click Save at the top level of the agent.

Step 5: Preview

1. Click Preview and ask a question that should be answered from the documents:

   Which scenarios require immediate escalation?

   When should I create a support case?

2. Confirm the response cites content from the uploaded documents rather than hallucinating.

Summary

• Learned when RAG complements structured queries: process and policy questions vs. data questions.
• Created a Data Library and a knowledge subagent, and grounded responses with a search action.

Next (optional), you'll surface unified Data 360 data in the agent.