Concept card

Datasphere Catalog Hygiene for Joule Grounding

As of 2026-09-27

Edited by Cédric Mary, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is Datasphere Catalog Hygiene for Joule Grounding?

Joule and agents choose data by its metadata. In SAP Datasphere and BDC that metadata lives in the Catalog & Marketplace: descriptions, business terms, KPIs, hierarchical tags, lineage and classifications. SAP now sells AI help for it — content generation for catalog, AI-assisted search and semantic generation for non-SAP sources are premium AI features, while Joule in Datasphere is base — and since July 2026 personal-data columns from BDC data products are tagged automatically. Hygiene is still a steward's job: AI drafts, humans approve.

Why the catalog is part of the AI stack

An LLM cannot inspect millions of columns to guess which one holds "billing value". It relies on names, descriptions, business terms, classifications and relationships — metadata. SAP's own positioning says so: at TechEd 2024 SAP described SAP Knowledge Graph as the connection between natural language and SAP's metadata; in 2026 BDC is marketed as a knowledge core that connects data with semantics so agents understand the business. Community practitioners building agents on BDC data products make the same point from the other side: business vocabulary such as SalesOrder or Supplier, carried in data product metadata, is recognised by LLMs far more reliably than technical names like EKKO or KNA1. A catalog with empty descriptions, unlinked terms and missing classifications therefore degrades AI answers even when the data itself is complete and fresh — and the failure is quiet: vague answers, dropped filters or "I could not find that".

Why it matters

LLMs select and label data from metadata; empty descriptions and unlinked terms degrade AI answers even when data is complete and fresh.
SAP now offers AI help for catalog work (content generation, AI-assisted search, semantic generation) as premium AI features — scope and entitlement must match.
Since July 2026 personal-data columns from BDC data products are tagged automatically; custom and replicated objects still need human classification.

Key points

Catalog & Marketplace unified in Datasphere release 2025.05: assets, data providers, terms, KPIs, data products in one search.
Business semantics: glossary terms, KPIs, hierarchical tags; impact and lineage analysis.
July 2026: automatic personal/sensitive-data tags on columns from BDC data products, inherited downstream.
Premium AI features: content generation for catalog (descriptions, terms, KPIs, auto-tagging), AI-assisted search, semantic generation for non-SAP sources.
Joule with SAP Datasphere is a base feature (informational, navigational, transactional).
AI drafts, stewards approve: unreviewed generated descriptions are a governance gap.
Measure hygiene monthly: assets without description, term or classification in the consumption layer.

Terms used on this page

Catalog & Marketplace: Unified Datasphere tab (since 2025.05) for discovering assets, terms, KPIs and data products, with governance and lineage.
Glossary term: Business concept with definition and synonyms, linked to assets or to individual measures and dimensions.
Hierarchical tags: Tag taxonomy used to classify catalog assets; the source list for AI auto-tagging.
Content generation for catalog: Premium Datasphere AI feature that drafts business descriptions, terms and KPIs and auto-tags assets.
Semantic generation: Premium Datasphere AI feature that generates business semantics for non-SAP sources.
Data steward: Role accountable for definitions, classifications and quality of governed data, including review of AI-generated metadata.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →