From PoC to Production — the SAP AI Delivery Lifecycle

What you will learn

• Name SAP's own six-phase AI Golden Path (Explore/Discover, Design, Deliver, Run & Scale, Evaluate & Improve, Extend) and where SAP says most engineering effort belongs
• Make the ML vs LLM vs agentic design decision explicitly, naming rejected alternatives and the SAP service each option would use
• Distinguish deployment as SAP AI Core resource groups/scenarios/executables from Joule's package-and-feature activation flow
• Plan Run & Scale as a ramp with evaluation-gated checkpoints instead of a single go-live date
• Build a production evaluation cadence (Evaluate & Improve) tied explicitly to SAP's documented model-deprecation schedule
• Explain what the Extend phase requires (documented interface, authorisation model, reusable configuration) for a solution to become a pattern rather than a one-off

Module overview

You have a business case and a sponsor who said yes to a pilot. This module is the route from that yes to a production system that survives its first model retirement, its first audit and its first busy Monday. The backbone is SAP's published delivery lifecycle, the AI Golden Path on SAP Architecture Center; for each phase you get the SAP object that carries it, the artefact you must produce, and the gate that decides whether you move on. It assumes fundamentals and feeds AI Launchpad operations, evaluation harness, LLMOps and security for SAP AI.

Prerequisites

• AI & LLM Fundamentals for SAP Consultants and Building the SAP AI Business Case
• Working knowledge of at least one of SAP AI Core, the generative AI hub, or Joule Studio
• Comfort reading a delivery plan or project phase gate

Outcomes

• Map a real SAP AI initiative onto SAP's six AI Golden Path phases and identify which phases lack an owner or a deliverable.
• Produce a written Design-phase decision record naming the chosen approach, the rejected alternatives, and the SAP service for each.
• Design a Run & Scale ramp with evaluation-gated checkpoints instead of a single go-live milestone.
• Specify an Evaluate & Improve cadence (test-set size, metrics, re-run frequency, alerting owner) tied to SAP's documented model-deprecation practice.