Concept card

Real-Time Operational Analytics

As of 2026-10-06

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 Real-Time Operational Analytics?

Sub-minute analytics on streaming data — shop floor, logistics, trading desks — requires CDC plus streaming engines plus HANA live federation, a fundamentally different stack from batch BI.

What it is

Real-time operational analytics is reporting whose value decays in minutes — shop-floor status, order exceptions, service-level breaches — as distinct from analytical reporting whose value is stable over days.

Why it matters

The decay rate is the design constraint, and it inverts the usual priorities. Latency beats completeness: a partial answer now is worth more than a complete one after the shift ends. That single inversion changes the architecture, and it is the thing most teams get wrong by reusing a warehouse pattern.

It also changes what "correct" means. An operational figure that is revised as late data lands is normal; the same behaviour in a financial report is a defect. Saying which regime a number lives in prevents an argument later.

How it works

The practical question is where the query runs. Embedded analytics against the transactional system gives you the freshest possible answer at the cost of load on the system of record. A replicated store gives you isolation at the cost of latency. Federation sits between and is defensible for low-cardinality, low-frequency operational lookups.

In an SAP estate the honest framing is that operational immediacy usually argues for staying close to the source, while cross-domain analysis argues for the analytic layer — and most real requirements are a mix, which is why one pattern for everything fails.

Why it matters in practice

Traditional batch BI architecture cannot be retrofitted to sub-minute latency — this is a different stack decision, not a performance-tuning exercise on the existing one.
Shop floor, logistics and trading-desk use cases share the same requirement: CDC plus streaming, not periodic extracts — the pattern generalizes across very different industries.
HANA live federation is the piece that lets streaming data join with existing modeled data without a full re-platform — the practical bridge most clients actually need.

Key points

Sub-minute analytics on streaming operational data — shop floor, logistics, trading desks.
Requires CDC + streaming engines + HANA live federation, not traditional batch BI.
State the staleness budget explicitly (how stale may an answer be before it stops being actionable) and design to that number, not to the word 'real time'.
Latency beats completeness in this regime — a partial answer now often outranks a complete one after the decision window closes.
Adding a Joule conversational layer introduces a new latency component (orchestration-service grounding + LLM call) that must be priced into the staleness budget if the AI answer gates an action.
Decide whether the AI layer sits in the decision loop (its latency counts toward the budget) or beside it (a human asks a follow-up after the direct-path alert already fired) — the second is the safer default for genuinely decay-sensitive use cases.
HANA Cloud's Vector Engine plus Knowledge Graph Engine (GeoSPARQL added QRC3 2026) support hybrid RAG for control-tower-style questions blending real-time position data with a routes/facilities graph — but that hybrid query has its own latency profile too.
Embedded reporting that competes with transaction processing degrades the system the business actually runs on — isolation (a replicated store) trades latency for exactly that protection.

Terms used on this page

Staleness budget: The explicit, stated maximum age an answer may have before it stops being actionable for a given decision — the design constraint this card centres, expressed as a number, not the word 'real time'.
CDC (Change Data Capture): A technique that streams only the changed rows from a source system rather than re-extracting full tables, the usual mechanism feeding a sub-minute operational analytics pipeline.
HANA live federation: Querying data in place across systems without full replication — defensible for low-cardinality, low-frequency operational lookups, but not a substitute for isolation when query volume is high.
Embedded analytics (operational): Running an operational query directly against the transactional system of record for maximum freshness, at the cost of adding load to that system.
Orchestration-service latency: The added round-trip time of grounding retrieval, content filtering and the LLM call itself in SAP's generative-AI-hub orchestration service — a new line item in an operational staleness budget once a Joule or LLM-based answer sits in a decision path.
AI beside vs. inside the decision loop: The design choice this card's aiSection introduces: an AI answer that merely informs a human after a direct-path alert already fired ('beside') versus one whose output gates the action itself ('inside', which inherits the AI call's latency into the staleness budget).
Hybrid RAG (HANA Cloud): Combining vector similarity search and graph-relationship queries (via HANA Cloud's Vector Engine and Knowledge Graph Engine, the latter with GeoSPARQL support added QRC3 2026) in one query layer — relevant to control-tower questions spanning live position data and a structural graph.

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 →

Guides that answer with this page

These guides cite this page as one of the sources their answer rests on.

Real-time operational analytics with AI in SAP