Artificial Intelligence and SAP ECC—No Need to Wait

The discussion about artificial intelligence in the SAP ecosystem has a fundamental problem: it almost always assumes that S/4 Hana is already in use, that BTP has been configured, or that a modernization project is well under way.

Reality is different. Thousands of companies around the world continue to work with SAP ECC 6.0. These are stable, proven systems that support business-critical processes. The migration to S/4 Hana is costly, complex and not an immediate priority for many companies. Some have decided to extend support beyond 2027. Others are looking at alternatives. And in the meantime, solutions with applied AI don't seem to be for them.

This scenario is not an exception, but the rule—even for large companies. In December 2025, Catherine Jestin, Executive Vice President Digital at Airbus, publicly stated that most CIOs in Europe will not have completed their migration from SAP ECC 6.0 by 2030. If companies with the resources of Airbus follow such timelines, waiting for a "tidy house” in S/4 Hana to start using AI means—in practice—forgoing innovation for half a decade.

However, this comes at a price. While other technology industries are integrating generative AI capabilities into their daily workflows, the traditional SAP world seems to merely be watching. SAP Joule and other solutions from SAP partners are already being deployed in companies, to be used immediately in day-to-day business. Is this divide inevitable? Not necessarily.

The barrier is not technical

There is a general perception that the integration of artificial intelligence into SAP ECC systems requires a modern architecture. That you first have to migrate, then modernize the interfaces with RAP and finally connect everything via BTP. Only then, supposedly, can the talk around AI begin.

This idea seems to be the official statement, but is not entirely accurate. ECC 6.0 systems have native functions that enable communication with external systems: RFC function modules, HTTP services via ICF, standard exchange formats. These interfaces have been around for years and work reliably.

What has changed is not SAP's technical performance, but what is on the other side of the connection. Current language models can understand Abap code, analyze data structures, suggest technical solutions, and generate functional code. The question is no longer whether AI can understand SAP, but how we establish communication.

This consideration led to the development of a different approach: connecting AI models directly to ECC systems via RFC interfaces and native HTTP services, without intermediate layers in the cloud, and without dependencies on additional platforms.

Governance and control are the real challenges

When you introduce the idea of linking artificial intelligence directly to an SAP system, the first reaction is usually a mixture of interest and concern. And this is completely justified. The closer you are to business-critical information, the more important questions about control, audit, and responsibility become.

Who approves the changes proposed by the AI? How is it ensured that data or code is not changed without supervision? What about authorizations and the separation of functions?

These questions are not obstacles, but design requirements. Any serious integration of AI into productive SAP systems must answer these questions from the outset.

The basic principle is simple: the AI supports, the human decides. Any process that involves changing system objects (programs, classes, function modules) must be explicitly validated by a human and may only be carried out in development environments or sandboxes. The AI can analyze, suggest and prepare, but the final activation remains in the hands of the developer.

The proposed changes are managed via the existing transport system. They will go through the same quality controls, approvals and audit processes as any other development. The AI will be integrated into the existing governance framework, but will not replace it.

There is already a security model for read operations (table queries, log checks, analyses of existing code): the standard SAP authorizations of the logged-in user. If a user does not have authorization to access certain information manually, they do not have it via the AI. There are no shortcuts that allow a user who previously did not have access to certain data to now obtain it.

When reading data, you can opt for a "translator" from natural language to SQL, which translates a query like "I want to see the last orders of supplier 'X' over 1000 Euro" into a "select" for the tables EKKO/EKPO and LFA1. This translator is invaluable for business users who don't need to know what the SAP tables are called.

A bridge, not a stopgap

It is important to clarify what this approach is not. It is not an alternative to modernization. It is not an argument for staying with ECC indefinitely. The reasons for migrating to S/4 Hana are still valid: better performance, simplified data model, new business functions.

Rather, this approach offers an answer to a practical question: what do we do in the meantime? For companies whose migration is planned for 2026, 2027 or later, "in the meantime" can mean years of operation. Years in which Abap development continues, technical teams continue to maintain and evolve systems, and the gap with other technologies widens.

The introduction of AI functions into these systems does not delay modernization today. In fact, it can actually make it easier. Teams that use AI to better understand their legacy code, document implicit business logic, or identify hidden dependencies are better equipped for a successful migration.

Furthermore, the fact that this solution enables the connection of SAP ECC systems with AI does not mean that it will no longer work when migrating to S/4 Hana or the cloud. It will probably require updating the objects that manage actions performed by agents within SAP, but the idea itself should continue to work.

Looking Ahead

The SAP ecosystem is changing. Some customers will quickly switch to S/4 Hana. Others will take a longer route. Some will explore alternatives outside of SAP. What they all have in common is that artificial intelligence will be part of their future, regardless of which platform they choose.

The relevant question is not whether classic SAP systems can benefit from AI. The question is whether we are prepared to build the necessary bridges with the care and governance that these critical systems require.

The technology is ready. The interfaces exist. AI models understand Abap and SAP data. Now we just have to decide whether to wait for the perfect future or start with what we have today.