Insights

AI on legacy SAP before migration: five options compared

Companies still running SAP R/3 or ECC can add conversational AI before they migrate to S/4HANA. The realistic choices are five: SAP's Joule through SAP BTP, an accounts payable automation platform, robotic process automation with a chat front end, Microsoft 365 Copilot Studio, or doing nothing for now.

This article summarises AI For SAP 4.7: A CIO's Field Guide to Conversational AI on Legacy SAP Before the Migration by Anupam Shrivastav (e2i research & analysis, May 2026). The book works through one use case, invoice exception handling in accounts payable. Figures and product details are the book's, as of May 2026, so check current vendor products and pricing before deciding.

Joule does not run natively on old SAP

The book is direct on this point. Joule through SAP BTP is not natively supported in an on-premise SAP R/3 system. What works is a side-by-side bridge: Joule runs in SAP's cloud platform, the data stays in the on-premise system, and the two are connected through SAP Cloud Connector and OData services. Anyone who describes Joule as natively available on R/3 4.7 is mistaken, or not being straight with you.

The five options

1. Joule via SAP BTP. Build the AI on SAP's own cloud. The most complex option and the highest in strategic value, and the work carries over to S/4HANA.
2. AP automation platforms. Buy a specialist invoice AI tool. The fastest return for accounts payable alone, but it does not carry over to S/4HANA.
3. RPA with a conversational interface. A software bot that navigates SAP screens. The fastest to deploy, but not really AI and brittle when screens change.
4. Microsoft 365 Copilot Studio. Use AI the finance team already has through Microsoft 365. Familiar to users, but shallow in its SAP depth.
5. Do nothing. Improve the reports. A legitimate choice if migration is within about two years, otherwise the competitive gap compounds.

Which option fits

The right option depends on conditions, not on a universal ranking. The book's quick filter for a CIO:

• Regulated industry and a migration within five years: Joule via SAP BTP is the clear signal.
• A board mandate for visible AI progress within 90 days and a budget under USD 150,000: RPA is the tactical answer.
• Intent to build three or more AI use cases on SAP over time: Joule via SAP BTP is favoured.
• Microsoft 365 as the main working platform: Copilot Studio and RPA fit better.

A bounded first project

The book sets out a 26-week proof-of-concept with four go or no-go gates. Scope is kept deliberately small: one use case, one business unit, 10 to 25 pilot users, no more than three business functions called in SAP, and informational queries with a basic release or escalate action. Migration, non-SAP integration and multi-country rollout are out of scope.

• Weeks 1 to 4, discovery. Check technical prerequisites, identify the SAP functions to call, run workshops with pilot users and sign off scope and success measures.
• Weeks 5 to 10, platform foundation. Connect the cloud platform to the on-premise system, set up identity so users see only what they are allowed to see, and complete a security review. No AI development starts until that gate passes.
• Weeks 11 to 20, development. Build the AI skill, put it in front of the pilot group, and test with real users.
• Weeks 21 to 26, rollout and handover. Expand to 25 users, train administrators, and compare actual results with the baseline set in week 4 before recommending the next step.

The book is available on Amazon. For why so many companies are in this position, see why companies stay on legacy SAP, and for access and audit controls on anything touching financial data, see the pre-IPO governance gap.