CIOs must stop conflating AI in SAP operations with AI in SAP delivery

The HFS Point of View on AI in SAP transformation is for CIOs and transformation leaders distinguishing proven AI in SAP delivery from unproven AI claims in production ERP operations.

The AI narrative in SAP transformation has split into two distinct realities, and enterprise CIOs have not noticed the boundary line being drawn. Most service providers have genuinely embedded GenAI and copilots into migration tooling, code remediation, documentation, and test generation. Productivity gains in the build phase are measurable and real where providers have made them contractually visible.

While providers are deploying AI in AMS and operational ticketing with measurable improvements, enterprises should be clear-eyed that this is not the same as AI in business-critical ERP decisions. The gap between AI in SAP delivery and AI in production ERP operations is stark, and the blockers are structural, not just technical.

AI is working well in the build phase, but the impressive numbers don’t tell the whole story

Across the 25 providers HFS assessed as part of the HFS Horizons: SAP S/4HANA Transformation Services, 2026 report, AI integration in SAP delivery is genuine and shows up in the form of automated fit-gap analysis, AI-generated documentation, GenAI-led code remediation, migration copilots, and automated test script generation. The results of this integration are genuine and easily measurable. Providers report 20%–30% improvement in developer productivity, 70% in auto-generated test scripts, 20% in build effort, and 30% in testing cycle. The direction is consistent across multiple providers and shows significant improvements in the delivery machine.

However, enabling Joule in mission-critical finance, procurement, and supply chain workflows faces three persistent obstacles that no product release resolves on its own:

• Data quality deficiencies that make AI outputs unreliable for financial decisioning;
• Heightened AI governance and explainability requirements that most enterprises are not yet equipped to satisfy;
• Organizational risk aversion in ERP teams where the cost of an agent error in a financial close or procurement approval is high and visible.

Every provider has an agentic AI story, so the bar for proof should be proportionally high

The ecosystem has invested heavily in agentic AI across multi-agent workflows, orchestration layers, custom SAP agents, and Joule Studio integrations. The ambition is legitimate, as automating cross-process decision chains in finance and procurement is exactly where SAP transformation needs to go.

The reality is that most of these are in controlled pilots with limited production scope. Enterprises should welcome the R&D investment, but require specifics before it factors into vendor selection. Ask for which processes are in production, at which clients, under what governance model, and who owns the outcome when an agent makes the wrong call.

SAP Business Data Cloud readiness is the next gating factor, and most enterprises are not ready for it

The AI story in SAP is inseparable from data architecture, and the infrastructure direction is now clear. SAP Business Data Cloud (BDC), converging SAP data assets, HANA, Databricks, and Datasphere, is how SAP is driving a governed, AI-ready foundation. Providers that can help enterprises build toward BDC readiness, including clean master data, governed extensibility, and real-time data fabric, should have structural positioning in the next investment cycle. Those focused on migration velocity alone should be commoditized.

For enterprise leaders, the decision ahead is not only whether to complete the S/4HANA migration, but also whether the data architecture on the other side will support the AI use cases the board is expecting in 18–24 months.

The Bottom Line: Your provider must be able to clearly distinguish between AI capabilities that will be in production at go-live and those on the forward roadmap. They are not the same commitment.

Before signing the next SAP transformation contract, ask what data quality and governance conditions must be satisfied for the AI use cases to be activated and what will happen commercially if those conditions are not met. The ECC maintenance deadline is real pressure, but it should not compromise due diligence on AI claims with a delivery horizon of 2027 or beyond.