Concept card

Datasphere Replication Flows for AI 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 Replication Flows for AI Grounding?

An agent is only as current as the data it is grounded on. In SAP Business Data Cloud, SAP-managed data products are provisioned by SAP; everything else — custom CDS views, uncovered tables, non-SAP sources — reaches Datasphere through replication flows (initial and delta load, CDC, continuous or scheduled delta runs). Freshness, filters, target choice and monitoring of these flows decide whether Joule and custom agents answer from today's data or from a silently stale copy.

Why replication is an AI topic

A grounded answer inherits the freshness of its data. If an agent reasons over yesterday's open purchase orders, it will recommend actions that a buyer can refute in one click — and trust is gone. In BDC, two paths feed the governed layer. SAP-managed data products for SAP applications are replicated and curated by SAP itself; customers install and consume them (including sharing them to HANA Cloud, where they are read through virtual tables). Replication flows in SAP Datasphere cover the rest: custom ABAP CDS views, tables no data product covers yet, BW and ECC extractors, and non-SAP sources. For those, the replication design is your grounding design.

How replication flows work

A replication flow is a Data Builder artefact for mass data replication from one source to one target, with simple projections, row filters and column mapping, running cloud to cloud without a Data Provisioning Agent. SAP product management lists the core mechanics:

Why it matters

A failed or slow delta does not stop an agent from answering — it makes it answer confidently from old data.
SAP-managed data products are provisioned by SAP; replication flows are the customer's lever for everything they do not cover (custom CDS views, BW/ECC, non-SAP).
July 2026 changed the options: delta load types for BW and ECC sources, local table files as source, REST APIs as source (restricted, EU10).

Key points

Two feeding paths in BDC: SAP-managed data products (provisioned by SAP) and customer replication flows in Datasphere.
Load types: Initial Only, Initial and Delta; BW and ECC sources gained Initial and Delta and Delta Only in July 2026.
CDC is trigger-based with logging tables on the source, except ODP sources (delta without triggers).
ABAP source containers: CDS, CDS_EXTRACTION, ODP_BW, ODP_SAPI, SLT.
Delta runs: On Delta Interval (default, long-running) or At Scheduled Time; e-mail notifications on failure.
Non-SAP targets = premium outbound integration (capacity must be assigned).
REST API source: restricted release, EU10 only, SAP Note 3696110, no projection/filter/mapping.
Agents should consume governed models or data products on top of replicated tables, never raw tables.

Terms used on this page

Replication flow: Datasphere artefact for mass replication from a source to a target with projections, filters and initial/delta loads.
Initial and Delta: Load type that performs a full initial load, then replicates inserts, updates and deletes.
On Delta Interval / At Scheduled Time: Delta run settings: continuous long-running delta checks versus delta loads at defined times.
Premium outbound integration: Replication from Datasphere to non-SAP targets; requires assigned capacity.
SAP-managed data product: Data product for SAP application data that SAP replicates, curates and publishes in BDC.
ODP: Operational Data Provisioning: SAP extraction framework whose delta mechanism replication flows use without triggers.

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 →