Concept card

From proof of concept to production — the SAP AI delivery lifecycle

As of 2026-09-25

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

What is From proof of concept to production?

SAP AI Core's own object model — scenario, executable, configuration, deployment — is a production-promotion gate disguised as a technical container hierarchy: nothing runs until a configuration is explicitly turned into a deployment, and nothing scales across environments without a named resource group and a promotion decision. Treating a PoC as "basically done" once the demo works skips the two gates SAP built the lifecycle around.

The object model is the gate

SAP AI Core structures every AI asset through a fixed chain of objects, and that chain is where a delivery team should be pricing the PoC-to-production gap, not treating it as implementation detail to sort out later. A scenario is the container that groups everything related to a specific AI use case — models, configurations, executables — under one identity. An executable is the packaged, runnable form of a model or workflow, provided as a serving template a team selects for the model they want to run. A configuration binds a specific executable, version and parameter set together, and nothing actually runs until that configuration is turned into a deployment: you select the scenario, version, serving executable and configuration, then explicitly create the deployment. This is not incidental plumbing — it is the point in the lifecycle where a PoC's ad hoc model call becomes an operated, versioned, monitorable production object, and it is also the point where cost, resource-group placement and access control get decided, not inherited by accident from whatever notebook the PoC ran in.

Why it matters

AI Core's scenario -> executable -> configuration -> deployment chain is the object model that forces the production-promotion decisions (resource group, access control, cost tagging) a PoC usually skips by design.
A solution spanning both AI Core and an S/4HANA-embedded scenario has two separate production-promotion gates (AI Core's deployment step and ISLM), not one — scoping only the AI Core side under-prices the engagement.
The PoC-to-production gap is priced work, not go-live support: exception handling, monitoring and the approval/evaluation gate typically run 30-50% of the original PoC effort again, and a proposal that hides this in a vague go-live line will re-litigate scope mid-delivery.

Key points

AI Core object chain: scenario (groups models/configs/executables for a use case) -> executable (packaged, runnable model/workflow) -> configuration (binds executable+version+params) -> deployment (the only thing that actually runs).
Resource groups are the isolation boundary for tenancy, cost allocation and access control; the AI-Resource-Group header routes every orchestration call to one.
A PoC usually runs in one default resource group under one developer's credentials — production needs a deliberate per-environment resource-group decision before go-live.
ISLM (C346) is the separate, ABAP-side production-promotion gate for AI scenarios embedded in S/4HANA, distinct from AI Core's deployment step.
C114 documents SAP's full agentic delivery lifecycle (Plan and Build, Discover and Provision, Observe and Analyze, Secure and Govern, Optimize and Decommission) plus Joule Studio's generated-test/AI-evaluation/GitHub-approval gate.
PoC-to-production readiness checklist: named resource group, passed evaluation/approval gate, exception handling built (not just demo paths), monitoring stood up, Ideation risk classification (C372) completed.
The PoC-to-production gap typically costs 30-50% of the original PoC effort again — price it as its own SOW line, not folded into "go-live support."
A solution spanning AI Core and an embedded S/4HANA scenario has two separate promotion gates to price, not one.

Terms used on this page

Scenario (AI Core): The container grouping all models, configurations and executables related to one specific AI use case.
Executable: The packaged, runnable form of a model or workflow, selected as a serving template.
Configuration: The binding of a specific executable, version and parameter set — the object that gets turned into a deployment.
Deployment: The only AI Core object that actually runs; created explicitly from a configuration.
Resource group: SAP AI Core's isolation boundary for tenancy, cost allocation and access control, selected via the AI-Resource-Group header.
ISLM: Intelligent Scenario Lifecycle Management — the ABAP-side production-promotion gate for AI scenarios embedded in S/4HANA.

Sources

Full card available to members. What the full card adds: the full decision framework · the common pitfalls and their fix · the cheat sheet · the code blocks · the facts worth quoting.

Open in the app →