Ground a revenue agent on Data 360 golden records — activate the retriever, wrap it in an Agentforce Data Library, build the revenue-context flow and next-best-offer prompt, give users Data 360 access, publish the Revenue Assistant, and test it

The last mile of the golden-record RAG pipeline. Activate the retriever over the account-narrative index, then create a RETRIEVER Data Library with sf agent adl and put ARFPC_<libraryId> in the agent's knowledge block. Build the two actions the agent calls, a Data 360 flow (Get_Account_Revenue_Context) and a prompt template (Next_Best_Offer), to a fixed contract. Grant Data 360 access, then publish and activate the Revenue Assistant example, and test routing, actions and grounded answers.

markdownguidev1W9DJ-BS4Estage: buildtrust: officialservice: agentforceinterface: cliinterface: agent-scriptinterface: yamlground agent on Data 360golden record agentsf agent adl create retrieverARFPC_ library IDGet_Account_Revenue_Context flow contractNext_Best_Offer prompt templateData 360 permission set for agent usersdata space scopepublish revenue agentrevenue agent testsemployee agent Data 360 access

Created 23 Oct 2026 · Modified 23 Oct 2026 · 1 version

Ground a revenue agent on golden records

Prerequisites:

  • the pipeline has run, and verify passes (guides/data360/automate-the-golden-record-pipeline.md);
  • the account-narrative search index is Ready;
  • a retriever on it, returning the account name, the narrative and the chunk, is active (reference/data360/search-indexes-and-retrievers.md).

1. Wrap the retriever in a Data Library

sf agent adl create --source-type retriever --retriever-id <retriever ID> \
  --name "Revenue Golden Records" --developer-name Revenue_Golden_Records --target-org dev --json
sf agent adl get --library-id <libraryId> --target-org dev --json
  • The library is ready at once. Put ARFPC_<libraryId> in the agent's knowledge: block.
  • Day 2: sf agent adl update --library-id <libraryId> --retriever-id <new ID> swaps the retriever without touching the agent.

2. Build the two actions to this contract

ActionImplementationInputs → outputs
get_account_context → flow://Get_Account_Revenue_ContextAn autolaunched flow that reads the golden record's rollup from Data 360. Salesforce's workshop uses a flow with a Data Cloud query for exactly this ("Get Merchant Order Insights")accountName (text) → revenueSummary (text), lifetimeRevenue (number)
next_best_offer → prompt://Next_Best_OfferA Flex prompt template with one text input, Account_Context, instructed to propose one offer and its reasoning from the context only"Input:Account_Context" → promptResponse
  • Build the flow from the account-level revenue insight (Revenue_Rollup__cio) joined to the unified account. Return the same narrative text the search index embeds, so a brief and a search result read the same.
  • Flows and Apex as actions: family 9. Prompt templates: family 6.
  • Deploy both before publishing. Publishing doesn't deploy actions (reference/agentforce/agent-errors.md).

3. Access

WhoNeeds
Sellers (the agent is an employee agent and runs as them)A Data 360 permission set covering the data space, and access to the flow and the prompt template
An Einstein Agent User, if you make it a service agentThe same Data 360 access. Salesforce's sf-skills names DataCloudUser, GenieUserEnhancedSecurity or the GenieDataPlatformStarterPsl licence, depending on the org

Without Data 360 access or data space scope, the knowledge action returns an empty summary and the agent says it found nothing.

Salesforce's own guidance conflicts on default_agent_user here. The sf-skills org-setup reference says employee agents using AnswerQuestionsWithKnowledge with live actions need it. Its agent-design reference says setting it on an employee agent fails publish with "unknown error". The example omits it. If live preview of the knowledge action fails on user context, test that variant in a scratch org before you commit to it.

4. Publish, activate, test

Use the agent in examples/revenue-assistant-agent.md, with your library ID.

sf agent validate authoring-bundle --api-name Revenue_Assistant --target-org dev
sf agent preview start --authoring-bundle Revenue_Assistant --use-live-actions --target-org dev --json
sf agent publish authoring-bundle --api-name Revenue_Assistant --target-org dev
sf agent activate --api-name Revenue_Assistant --version 1 --target-org dev
sf agent test create --spec specs/Revenue_Assistant-testSpec.yaml --api-name Revenue_Assistant_Tests --target-org dev
sf agent test run --api-name Revenue_Assistant_Tests --wait 20 --result-format json --output-dir test-results --target-org dev
# agent-test-spec: Revenue_Assistant_Tests
name: "Revenue Assistant Tests"
subjectType: AGENT
subjectName: Revenue_Assistant
testCases:
  - utterance: "Brief me on Acme Corporation"
    expectedTopic: account_brief
    expectedActions: [capture_account, brief]
    expectedOutcome: "A short revenue summary for Acme Corporation using returned figures only"
  - utterance: "Which accounts buy like Acme but haven't ordered in six months?"
    expectedTopic: find_accounts
    expectedActions: [search_accounts]
    expectedOutcome: "A list of matching accounts with citations, or a statement that none were found"
  - utterance: "What's our churn policy for employees?"
    expectedOutcome: "The agent declines and offers account briefs or account search"
  • Gate CI on the result file, because failed assertions exit 0 (guides/agentforce/test-an-agent.md).
  • Check generatedData.actionsSequence on the first run. Whether capture_account (a @utils.setVariables tool) is listed is the hub's guess.

Read this document as markdown — the same bytes an agent receives.