Building a practical AI roadmap for SAP landscapes

SAP & Enterprise Technology · Intlex Technologies · October 2026 · 4 minute read

For organizations running SAP, the most valuable AI opportunities sit inside the processes SAP already supports: the financial close, procurement, planning, maintenance. A practical roadmap starts from those processes, not from a list of technologies.

Start with the process

Record to report, procure to pay, plan to produce and asset maintenance are common starting points. They combine high transaction volumes, well-understood workflows and outcomes that finance already measures: days to close, cost per invoice, forecast accuracy, unplanned downtime. That makes value easy to baseline and hard to dispute.

Know what the platform provides

SAP continues to embed AI into its applications and into the SAP Business Technology Platform, including SAP Business AI and the Joule assistant. A roadmap should separate three categories clearly:

• Capabilities already available in your SAP release, which need configuration, adoption and governance rather than development.
• Capabilities that need extension on SAP BTP, which need design and integration standards.
• Needs better served by external AI services or data platforms, integrated back into SAP processes.

Getting this wrong is expensive in both directions: building outside SAP what SAP now delivers, or forcing into SAP what another platform does better.

Sequence around readiness

The right order depends less on ambition than on readiness. For each candidate process, assess:

• Master data quality and ownership in that process.
• Alignment with the S/4HANA migration or upgrade timeline. Investment in an ECC process that is about to be redesigned may not carry forward.
• Integration requirements with non-SAP systems.
• The controls the process is subject to, and the risk tier the AI use case would fall into.

What the first twelve months can look like

• Months 1–2: confirm priorities with finance and operations leadership, inventory AI capabilities already licensed in the SAP landscape, and baseline the target process measures.
• Months 2–4: switch on and govern the capabilities that need configuration rather than development, typically in document processing and assistance, and fix the master data issues they expose.
• Months 4–9: design and deliver one or two extensions on SAP BTP for processes where SAP does not yet meet the requirement, with integration standards written as you go.
• Months 9–12: measure results against the baseline, retire what did not work, and re-sequence the roadmap around the S/4HANA plan for the following year.

The point is not the exact timing, which depends on your landscape, but the order: use what you already have, prove value, then build.

Keep the core clean

SAP’s own guidance for S/4HANA, and particularly for RISE with SAP, encourages keeping the core close to standard and building extensions outside it, typically on SAP BTP. The principle matters for AI too. Custom AI logic embedded in modified core objects is expensive to carry through upgrades; the same logic built as a side-by-side extension, calling SAP through released APIs, is far easier to maintain.

A practical test for each proposed AI extension:

• Does it read or write SAP data only through released APIs and events?
• Can it be upgraded, or replaced, without touching the SAP core?
• Is its data flow documented, including what leaves the SAP environment?
• Does it follow the same integration standards as every other extension in the landscape?

If the answer to any of these is no, the design should be revisited before build starts.

A worked example: invoice processing

Consider accounts payable, a common first candidate. The roadmap question is not “should we use AI for invoices?” but a sequence of narrower decisions:

1. What does the current release already offer? Document capture and matching capabilities may already be available in the landscape and only need configuration, testing and adoption.
2. Where does it fall short? Perhaps non-standard supplier formats, or invoices in languages and layouts the standard capability handles poorly.
3. Is the gap worth closing? Compare the volume of exceptions with the cost of an extension, using the current cost per invoice and the exception rate as the baseline.
4. How should it be built? If an extension is justified, build it on SAP BTP or an external AI service, integrated through released interfaces, so the core stays standard.
5. How will it be governed? Invoice processing is usually a lower-risk use, but it touches financial controls, so the controls owner signs off on accuracy thresholds and on the human review of exceptions.

Each step produces a decision that can be explained to finance leadership, which is what makes the roadmap credible.

Plan it with the rest of the enterprise

SAP AI decisions should not be made in a separate program office from the rest of the AI portfolio. The same governance, risk tiers and data platform decisions apply. A single roadmap, with SAP and non-SAP initiatives side by side, lets leadership fund the best opportunities regardless of which platform they run on.

Related service: SAP Business AI & Joule