For years, "AI in SAP migration" was a slideware promise. That changed. In SAPinsider's 2026 benchmark, SAP's AI announcements were the #1 external factor shaping ERP strategy at 43% — ahead of the 2027 deadline itself at 39%. AI is no longer a side story; it's the lever that decides how fast, how cheaply, and how well a migration gets done. This article explains, concretely, what agentic AI does to a migration, how much effort it removes, and — just as importantly — the 20% it can't touch.
For context, see our complete migration guide. Here we focus on the delivery revolution.
The Numbers: Significantly Less Effort
The headline figures are real and worth stating precisely. SAP is targeting around Significantly less transformation effort using coordinated autonomous agents — and the key word is coordinated: custom-code, configuration, and testing agents hand context to one another, so the saving compounds rather than living in one tool. An independent study of roughly 180 migration activities put potential AI savings closer to 60%. The honest framing that keeps recurring is "AI does the 80% solution, experts finish the rest" — which is exactly the right mental model.
Where AI Actually Helps
AI earns its keep on the slowest, most repetitive phases:
- Assessment — agents scan your custom code and data to size the work in days, not weeks, turning a vague estimate into an evidence-based plan.
- Custom-code (ABAP) remediation — the worst bottleneck. Agents find, document, and fix custom developments for S/4HANA; SAP's first ABAP custom-code migration agent for ECC→S/4HANA is already shipping. See migrating custom ABAP code.
- Data migration and cleansing — AI validates, maps, and cleans master data at scale. See data and clean core.
- Testing — AI generates and runs test cases, compressing one of the longest phases of any migration.
"AI absorbs the mechanical 60-80% of a migration. That doesn't make expertise less valuable — it makes it more valuable, because the margin moves to the judgment that agents can't supply."
— ESS ENN Associates AI Engineering Team
What AI Can't Do — And Why That Matters
Beware anyone selling full automation. The reality is more nuanced and more useful: automated quick-fixes cover roughly 40-60% of typical S/4HANA simplification cases; the rest remains manual. The scarce asset in this new world isn't hands — it's judgment. Business-process decisions, gnarly custom logic, and hard data-quality calls still need experienced humans. As agents absorb the mechanical majority, value concentrates in the remaining slice: custom-code intelligence, data-quality remediation, industry process templates, and agent orchestration. "We do SAP migrations" is commoditizing fast; owning that judgment layer is not.
An Open, Multi-Vendor AI Stack
One underappreciated fact: SAP's autonomous-enterprise stack is deliberately open, not a closed SAP-only AI. Reasoning models from partners — including Anthropic's Claude, alongside NVIDIA and Palantir — connect to SAP Joule agents via the Model Context Protocol (MCP). For anyone building accelerators on SAP BTP, that openness is strategically important: the orchestration layer isn't locked to one AI provider, so specialist partners can plug in best-of-breed reasoning and tooling. It's the same agent-engineering discipline we apply everywhere — see our work on deploying custom AI agents to production.
AI Reallocates the Market — It Doesn't Shrink It
A natural worry: if AI cuts effort by half, does the work dry up? No. AI reallocates the market rather than shrinking it. Fewer billable hours per project, yes — but more projects made viable (mid-market and late starters who couldn't previously afford or fit a migration), plus a durable second wave of clean-core and agent work after go-live. The services pool stays multi-year and large, just distributed differently. That's good news for late-starting organizations: the thing that makes 2027 reachable from a standing start is precisely this collapse in effort. See S/4HANA for mid-market and late starters.
The Practical Play for 2026-2027
The highest-leverage use of AI right now is to compress the front of the funnel: assessment → code/data remediation → test, then hand the business a clean core it can actually run agents on. Start with an AI-accelerated assessment — it's fast, cheap, and turns the deadline into a concrete, sized plan. From there, AI-assisted remediation and testing keep the timeline honest, while your experts spend their scarce hours on the decisions that matter.