Guide · SAP Business AI
Grounding SAP AI in BDC Data Products and Knowledge Graph
How to ground Joule and custom SAP AI agents in BDC data products and the SAP Knowledge Graph, what is available in November 2026, and what to test before sign-off.


To ground an SAP AI scenario in Business Data Cloud (BDC), you connect the agent to governed data products first, then add Knowledge Graph context only where your release and tenant actually support it. As of today, November 15, 2026, the data product path is the one to build on. The Knowledge Graph layer is still mostly SAP reference content, and the customer-facing extensions are either in beta or dated for 2027.
What "grounding" means in this stack
Grounding means the model answers from your business data and context, not only from what it learned in training. In practice, SAP's stack gives you three different grounding sources, and they behave differently under test.
- Data products. BDC's data products are the governed, reusable access layer for SAP data, and SAP's own architecture guidance positions them as the main path for analytics, data science and AI agents to consume that data (SAP Learning architecture guidance).
- SAP Knowledge Graph. This is the semantic layer. It maps entities, relationships and business rules so an agent can disambiguate terms and follow relationships between objects, such as a purchase order to a supplier to a plant (SAP, What is SAP Business Data Cloud?).
- Document grounding. This is a different mechanism. It retrieves text from unstructured documents for Joule Q&A, and it is not a substitute for data products. Out of the box, SharePoint is the primary supported connector for Joule's shared document grounding service (SAP Community architecture post).
Most failed "AI gave a wrong answer" tickets in pilots come from mixing these three up. Test each one on its own before you test the agent end to end.
Where each piece stands in November 2026
Dates matter here, because TechEd Berlin in late October 2026 announced a lot of items with different availability. Check each row against the SAP Help Portal or your SAP account team before you put it in a design.
| Capability | Status as of today | What to verify |
|---|---|---|
| BDC data products feeding Joule and Generative AI Hub | In use. SAP describes Joule Agents as grounded in the Knowledge Graph and BDC together. | Which data products are shared into your tenant, and whether the product is in your entitlement. |
| Ontology Manager | SAP's news describes it as available as beta. Its beta was planned for November 2026, and SAP has not given a GA date. | Beta access in writing. Keep it out of production designs. |
| Customer and SAP-created data products in the Knowledge Graph | Targeted to enter early adopter care in Q4 2026, per an SAP Community post. | Whether your tenant is in the early adopter group. |
| Joule Studio access to BDC data products | General availability planned for Q1 2027. | Build against the current method, not the planned one. |
| Customer-managed graph in HANA Cloud | Available today through the Knowledge Graph Engine, using RDF and SPARQL. | HANA Cloud license and triple store enablement. |
Sources disagree on one point. One third-party write-up says the Ontology Manager was introduced in beta at TechEd, while a second summary says SAP did not state an availability stage. SAP's own TechEd news says "available as beta," so treat the beta as real but unsupported for production.
Step by step: build the grounding path
Use this sequence for any Joule agent, Joule Studio scenario or custom agent that reads SAP business data.
- Source systemS/4HANA tables and CDS views, with their type columns and keys
- Data productCurated, governed product in BDC with a named owner
- Knowledge GraphEntities and relationships; SAP reference content today
- Model and orchestrationGenerative AI Hub model called through the agent
- Answer with evidenceResponse, data product used, and the query or lineage behind it
Step 1: Confirm entitlement and provisioning
BDC is not something you turn on because you have S/4HANA. Provisioning starts in SAP for Me with an S-user that is linked to the customer account holding the BDC entitlement on its order details (SAP Community provisioning guide). On RISE, confirm with your SAP contact which BDC scope is in your contract before the design review. The RISE scope itself is covered in our RISE testing ownership guide.
Step 2: Choose the data products, not the tables
Pick the data products the agent needs for one scenario, such as open purchase orders with supplier master data. Do not point an agent at raw S/4HANA tables. One SAP-linked practitioner write-up shows that S/4HANA tables often store several concepts in one table through type columns, so a column-level read can mix up object types (SAP Community, Knowledge Graph Agent series). A curated data product removes that ambiguity before the model sees it.
Step 3: Map relationships that cross data products
This is the step most teams skip. The same write-up notes that a graph built from a single data product only reports relationships inside that product. Links that span products, such as a work center belonging to a plant in another product, have to be bound by a domain expert. Write those links down as a design artifact, with an owner, before you test answers that depend on them.
Step 4: Add the Knowledge Graph only where it is live
SAP's Knowledge Graph today primarily holds SAP reference knowledge such as standard products, processes and interfaces. SAP has said that extending it to SAP-created and customer-created data products is targeted for early adopter care in Q4 2026 (SAP Community, What SAP Knowledge Graph Adds to BDC). If you need custom relationships before that is generally available, the documented option is the customer-managed Knowledge Graph Engine in SAP HANA Cloud. That needs a HANA Cloud instance with the triple store enabled and an AI Core connection for the orchestration model (SAP Developer Center tutorial). Our AI Core resource group and orchestration 404 fix covers the most common failure in that connection.
Step 5: Lock the identity and tenant setup
Joule and every SAP product integrated with it have to use the same SAP Cloud Identity Services tenant (SAP Joule integration guide). Mismatched tenants show up as access errors that look like missing data. Check this before you blame the data product.
What to test before you sign off a grounded answer
Grounded AI is still non-deterministic, so the test is about evidence, not exact wording. For each scenario, collect the following:
- The data product name, version and last refresh time that the answer used.
- Results compared against the same report run in S/4HANA for three to five known records, including one with a cross-product relationship.
- A test user for each role. Confirm the agent does not reveal rows or fields that role cannot see in the application.
- A negative case, such as a supplier that is blocked or a plant that was deleted, to check that the answer does not invent a current status.
- Whether the answer cites the data product or relationship it used, and whether that citation matches the record you checked.
- Results from a second run of the same prompt, logged, with a tolerance you agreed before testing.
Re-run the same test set after any model change or Joule update. Our guide to testing SAP AI features after a model or Joule update covers that cycle. The SAP Joule and embedded AI feature acceptance checklist gives you a starting form for the evidence above.
Cost and access checks
Grounded calls consume AI capacity like any other Joule scenario, so estimate the volume with the test set before go-live. Our SAP AI units cost estimate guide shows how to do that. For agents that act, not only answer, check the authorization model too, using our Joule agent authorization testing guide.
To go deeper, the free SAP Academy course RAG for SAP: Document Grounding, SAP HANA Cloud Vector Engine and Answers Over SAP Documentation covers the document grounding side. For evaluation sets and sign-off, take Testing Joule Agents and AI Features in SAP.
Frequently asked questions
Does an S/4HANA system ground Joule automatically in BDC data products?
No. You need BDC provisioned with the right entitlement, the data products shared into your tenant, and Joule configured to use them. Confirm each step with SAP for Me and your BDC cockpit before you test the answers.
Can I use the SAP Knowledge Graph for my own custom relationships today?
Not in the SAP-managed graph yet. SAP has targeted customer data products for early adopter care in Q4 2026 and lists the broader Knowledge Graph capabilities for Q1 2027. For custom relationships now, use the customer-managed Knowledge Graph Engine in SAP HANA Cloud, which uses RDF and SPARQL.
Is document grounding the same as grounding on data products?
No. Document grounding retrieves text from unstructured documents, such as policies and guidelines, and uses that text for Q&A. Data product grounding uses curated business data and relationships. Test them separately, because they fail in different ways.
What is the single most important test for a grounded answer?
Compare the answer to the source record for a known case that includes a cross-product relationship and a role-restricted field. If the answer is right for the easy case and wrong for the relationship case, the grounding is not ready for sign-off.
Sources
- SAP Empowers AI Developers to Build What Matters and Turn Intent into Autonomous Execution, SAP News Center, October 2026
- SAP TechEd in 2026: SAP Tells Developers "You Can Build the Autonomous Enterprise", SAP News Center, October 28, 2026
- SAP TechEd 2026 Innovation News Guide, SAP, 2026
- SAP TechEd 2026: What's Rolling Out and What's Planned Through 2027, SAPinsider, 2026
- Foundation Layer: AI, Data & Knowledge Graph, SAP Learning architecture, 2026
- What is SAP Business Data Cloud?, SAP product page, undated
- Joule Agents and Joule Assistants, SAP product page, undated
- What SAP Knowledge Graph Adds to SAP Business Data Cloud, SAP Community, November 2026
- Joule is Not Magic - Here is the Architecture Behind It, SAP Community, 2026
- Knowledge Graph Agent on SAP HANA Cloud Series, Part 2-2, SAP Community, 2026
- Introductory Guide to Provisioning of SAP Business Data Cloud, SAP Community, 2026
- Set up and use the Data Exploration Tools, SAP Developer Center, 2026
- Integration Guide: Integrating Joule with SAP Solutions, SAP Help Portal, 2026
Free S/4HANA testing templates
Master test strategy, migration test plan, cutover runbook, go/no-go checklist and more.
The RISE Testing Playbook
One practical move a week for testing your ECC to S/4HANA transformation.
More on SAP Business AI
- Cloud ALM AI Requirement Generation for Fit-to-Standard
- EU AI Act and SAP Joule: What to Document Before Go-Live
- Joule for Developers Not in ADT? S/4HANA Setup Steps
- Test SAP AI Features After a Model or Joule Update
- SAP AI Units: How to Estimate Business AI Cost
- Joule in SAP Ariba: Setup, Entitlement and Test Checks
- SAP AI Core Resource Group & Orchestration 404 Fix
- Accounting Accruals Agent: How to Test AI Accrual Proposals
How this guide is made: researched with AI-assisted search across SAP documentation and practitioner sources, with every fact linked to its source. Found an error or have field experience to add? Tell us.