Indu Learning Platform
Course/Unit 6 · AI inside MDG
Lesson 6-1

The SAP Business AI landscape

Where AI already shows up across SAP and MDG, where the model access lives, and when you build your own on SAP BTP.

Watch firstAI-Assisted Features in SAP Master Data Governance: S/4HANA 2023 FPS02 | Demo · SAP Help and LearningOpen on YouTube ↗

The stack in one picture

The first AI interview question is usually: where would the AI run, and how would it talk to MDG? SAP Business AI covers four things to keep apart: the assistant (Joule), AI features built into apps, the platform for building your own (generative AI hub, Joule Studio), and the data foundation. Names change often, so check current SAP pages before a client conversation.

Joule and Joule Assistantsuser front door, agentsEmbedded AI features in appsS/4HANA, MDG, Cloud ALMSAP Business AI PlatformJoule Studio, generative AI hubData foundationBusiness Data Cloud, HANA CloudSAP BTP servicesintegration, security, destinations
Top-down. For MDG, most value today sits in the middle two layers.

What SAP has documented for MDG

Stay inside this sourced list in interviews. AI-assisted change requests (central governance, S/4HANA Cloud Private Edition, from 2023 FPS02): describe a change in natural language, and get a generated summary of a CR's updates. Rule mining: ML proposes data quality rules from existing data (lesson 6-2). SAP's Q3 2026 highlights add Transition Data Quality agents (phased GA) and BP document extraction in Public Edition.

CapabilityWhat it doesWhere / whenAI-assisted CR creationNatural language to create orchange a CRMDG central gov., S/4HANA 2023FPS02+CR summaryGenerated digest of all updates ina CRSame as aboveRule miningML proposes DQ rules from existingdataMDG on S/4HANA, DQ managementTransition DQ agentsFind, rank, fix DQ issues withapprovalQ3 2026, phased GABP document extractionFill partner data from documentsS/4HANA Cloud Public Edition
Re-check each row against the client's release and edition.

Try it: who does what?

Try it · Match

Match each piece of the SAP AI landscape to its job.

Pick an item on the left, then its match on the right.

Generative AI hub: the governed model call

For client-specific use cases you call a model through the generative AI hub in SAP AI Core: one SAP-managed endpoint to models from several providers. Its orchestration service chains modules around the call: templating, grounding, data masking, content filtering and translation. Know where they sit, not the API.

Step through · 1 of 7
Your BTP appCAP / PythonTemplatingprompt + placeholdersData maskingpseudonymize PIIInput filterFoundation modelchosen in hubOutput filterUnmask and return
Your side-by-side app on SAP BTP collects the input, for example two supplier records flagged as possible duplicates.
Click through: your answer to 'how is the LLM call governed?'

Try it: name the module

Try it · Fill the blanks

Pick the orchestration module for each job.

Supplier contact names are replaced by placeholders by . A versioned prompt with field placeholders comes from . A note field saying 'approve this' is blocked by . Pulling in the client's payment term policy so answers are specific is .

Joule, agents and Business Data Cloud

In 2026 SAP positions Joule as the front door to Joule Assistants, which coordinate Joule Agents. Joule Studio (Sapphire 2026) is for building your own agents; SAP says the agent builder is free until December 2026, then premium. SAP Business Data Cloud (Feb 2025) is the data foundation, and SAP now places MDG under it as the layer that keeps data trusted for AI. Whatever the agent finds, master data still changes through change requests.

Joule Assistantsdomain levelJoule Agentsspecialized tasksJoule Studiobuild your ownBusiness Data Clouddata contextSAP Knowledge Graphbusiness semanticsGenerative AI hubmodel accessJoule (front door)
How SAP describes the pieces in 2026. Verify names before using them on a slide.

A pace check

Dec 2023Generative AI hubavailableS/4HANA 2023 FPS02MDG AI-assisted CRsFeb 2025Business Data CloudannouncedMay 2026Joule Studio atSapphireQ3 2026Data ManagementAssistant, phased GA
Not a roadmap. Expect at least one rename per year.

Embedded, side-by-side, or agent

Use the simplest option. Embedded if SAP ships it on the client's release. Side-by-side on BTP if it needs client policies or data. An agent only for genuinely multi-step, tool-using work. In every case the output enters MDG as a change request, mass change or consolidation, with a steward approval.

New AI use case for MDGSAP ships it embeddedCheck release + editionActivate and configureTrain stewardsNeeds client-specific logicBTP app + generative AI hubMask data firstReturn as change requestMulti-step, tool-usingJoule Studio agentHuman approval stepAudit every action
Embedded first, side-by-side second, agents when really needed.

Try it: pick the option

Try it · What would you do?

A client on S/4HANA 2023 FPS02, private cloud, wants approvers to get a quick summary of what changed in each supplier change request. What do you propose first?

A one-slide reference architecture

A BTP app reads records, calls the hub with masking on, validates the JSON answer, and creates a change request with the AI's reason as a note. The steward decides in the normal workflow.

StewardMDGBTP appGen AI hubRecords to review (OData / event)Masked prompt from templateJSON proposal + reason (then validated)Create change request + noteWorkflow itemApprove / rejectDecision feedback for metrics
The AI never writes active data. It proposes; the change request decides.
JSON
{
  "use_case": "Explain supplier duplicate candidates",
  "layer": "Side-by-side on SAP BTP",
  "model_access": "Generative AI hub (orchestration with data masking)",
  "inputs": ["BP name", "address", "tax number (masked)", "bank country"],
  "output_into_mdg": "Note on consolidation / change request for steward",
  "human_step": "Steward approves every merge",
  "metrics": ["acceptance rate", "minutes per decision", "post-merge reversals"],
  "verify_on_client_release": true
}

Keep a 'use case card' like this for every AI idea.

Remember

  • SAP Business AI spans Joule, embedded features, the AI platform (Joule Studio, generative AI hub) and the data foundation; verify availability for the client's release.
  • The generative AI hub's orchestration modules (templating, masking, filtering, grounding) are your governance story for LLM calls.
  • Embedded first, side-by-side second, agents last; results always return to MDG as change requests that stewards approve.

Check yourself

You need an LLM to explain duplicate candidates using your client's naming conventions. Where do you call the model from?
Which MDG AI-assisted capability does SAP document for central governance from S/4HANA 2023 FPS02?
Which orchestration module keeps supplier contact names away from the model?
An agent built in Joule Studio finds a wrong payment term on a supplier. What should it do?