Now accepting guest editorial from SAP, Oracle, Workday and ServiceNow practitioners.Pitch a guide
The Sign-Off Table

Lesson 6 of 6

Data foundations for Joule agents

10 min

Listen to this lesson

0:00 / 6:40

At SAP Sapphire 2026 SAP described its AI architecture in layers: Joule as the user experience, agents in the process layer, and data and AI as the foundation, with BDC and the SAP Knowledge Graph at its center. The message to customers was blunt: an agent is only as good as the data and business meaning it can reach. This lesson explains what that means in practice and what a tester should ask for.

Why agents need governed data

A Joule agent that answers "Which fleet customers are at risk of churn?" or proposes a collections priority list has to read data, understand what it means and combine it correctly. Transaction APIs give it current records; data products give it history and cross-system context; the SAP Knowledge Graph gives it relationships between business entities; and master data governance gives it one trusted version of a customer or product. If those layers disagree, the agent will be confidently wrong.

Joule and Joule WorkThe user asks a business question
Joule agentsPlan steps and call tools: APIs, data products, models
SemanticsSAP Knowledge Graph and Datasphere business layer: what entities and measures mean
Governed data productsSAP-managed and customer data products with owners and lifecycles
Trusted master dataSAP Master Data Governance (named a core BDC component in May 2026)
The data foundation an agent stands on. A weak layer anywhere shows up as a wrong answer at the top.
The Sign-Off Table · thesignofftable.com/academy/sap-dap-101

What SAP announced for agents in BDC

Announcement (May 2026)What it meansStatus to check
Joule agents in BDCSearch and create data products from prompts; generate analytical and planning models; generate SAC storiesNo dates given at announcement; check SAP Help and the Discovery Center
SAP HANA Cloud as a core componentOne engine for analytical, graph and vector workloads that agents can useNamed as part of BDC
SAP Master Data Governance in BDCGoverned master data and policies for agentsNamed as part of BDC
Reltio (acquisition completed)Multi-domain master data management with MCP support for agent workflowsPackaging and dates to be confirmed
SAP AI Core integrationBatch inference inside data products to enrich data used by JouleNo dates given

Sapphire 2026 BDC announcements relevant to agents. Announced is not the same as generally available.

An agent answer is a composition of data products and semantics; each input must be correct for the answer to be correct.
The Sign-Off Table · thesignofftable.com/academy/sap-dap-101

Contoso: is our data ready for an agent?

Dan wants a finance agent to explain monthly margin movements for CMC-ALT-100. Noor and Priya write down what the agent will need and test each input before anyone builds the agent: billing items with revenue (for example invoice 90001877: 50 × 240.00 = 12,000.00 USD), the standard cost of 150.00 USD in Detroit and 156.00 USD in Chicago, the profit center logic (Chicago Assembly carries CMC-AFTMKT), and customer master data for Great Lakes Fleet Services (17100001). They find that the product cost data product does not yet carry the Chicago cost component split. That gap would have produced a plausible but wrong explanation of Chicago margin.

  • Every data product the agent reads has an owner, a refresh time and a reconciliation to the ERP.
  • Business definitions (revenue, margin, cost) are agreed and documented in the semantic layer.
  • Master data for the entities in scope is governed: duplicates resolved, key attributes complete.
  • Access controls in the data layer match the business roles of the people who will ask the agent.
  • There is a gold set of questions with known correct answers, checked against the source systems.
A well-designed agent reports its sources and data dates, which gives testers something to verify.
The Sign-Off Table · thesignofftable.com/academy/sap-dap-101

What would you do?

A pilot agent tells Dan that Chicago margin fell because of higher labor rates. Priya checks and finds that the product cost data product last refreshed 10 days ago, before a standard cost update. What should happen next?

  1. InventoryList the data products, models and APIs the agent will use
  2. ReconcileProve each one against the ERP for a fixed period
  3. DefineAgree measure definitions in the semantic layer
  4. SecureAlign data access controls with business roles
  5. EvaluateRun gold questions; record answers, sources and freshness
Data readiness steps before an agent touches real decisions.
The Sign-Off Table · thesignofftable.com/academy/sap-dap-101
Key points
  • SAP positions BDC and the SAP Knowledge Graph as the data foundation for Joule agents.
  • Agents compose answers from data products, semantics and master data; any weak input produces a wrong answer.
  • Sapphire 2026 announced Joule agents in BDC, SAP HANA Cloud and SAP MDG as core components, and Reltio; check statuses before planning.
  • Test data freshness, reconciliation, definitions and access as part of every agent evaluation.

Check your understanding · not graded

Lesson check

  1. 1. What role did SAP give BDC in its Sapphire 2026 AI architecture?

  2. 2. An agent's answer is well cited but wrong because a data product stopped refreshing. What does this show?

  3. 3. Which component did SAP name as a core part of BDC in May 2026 to govern master data?

  4. 4. What is the purpose of a gold set of questions for an agent?