AI through SAP BTP

The SAP Parallel Universe: BTP and BDC

BTP (Business Technology Platform) and, in the future, probably also BDC (Business Data Cloud) are positioned as the natural boundary between the ERP core (clean core) and customer-specific modifications, data management, and add-ons. Within these platforms, the SAP BTP Generative AI Hub (as a function of SAP AI Core) forms the necessary orchestration layer for the use of LLMs.

Existing SAP customers gain access to a wide range of models via the GenAI Hub – not only those developed by SAP, but also those from external hyperscalers such as OpenAI, Google, and AWS. The hub is technically essential for combining the power of LLMs with relevant ERP data from the business context, for example, using the Hana Cloud Vector Engine for efficient vectorization and semantic search in RAG (retrieval-augmented generation) applications.

SAP AI without an SAP roadmap

Despite current technological advances, existing SAP customers face significant ERP challenges that often slow down the acceptance and ROI of AI applications within a cloud ERP system.

SAP's strategy and independence (unconventionality) in the field of AI is a continuing point of criticism. Users lack a clear ERP AI master plan or an S/4 AI roadmap for the period up to 2040. Some see SAP as too small for the big topic of AI when compared directly with the billions invested by Google, Microsoft, or Meta. SAP's focus on partnerships and acquisitions instead of its own fundamental AI expertise is seen as a waiver of a unique selling point.

SAP's former CEO Bill McDermott compensated for the company's lack of cloud expertise with acquisitions worth billions, demonstrating his willingness to take risks. This courage and a long-term AI strategy are lacking at SAP under CEO Christian Klein.

SAP AI without compliance, governance, and trust

Transparency, trust, governance, and compliance are existential issues. Especially when it comes to business-critical ERP applications, existing SAP customers demand a high degree of explainable AI (XAI) and traceability to prevent AI models from acting as black boxes.

The use of autonomous AI agents in S/4 systems also raises serious questions about operational safety, 100 percent availability, and liability, as the damage caused by an uncontrolled agent is estimated to be „astronomically high.“.

SAP Cloud First versus BTP GenAI Hub with on-premises

SAP's cloud-first strategy has widened the gap in relation to AI. Innovations such as Joule and the GenAI Hub are primarily geared toward cloud solutions, which puts existing on-premises customers with their historically grown, highly customized ECC 6.0 landscapes (the so-called Z world) at a disadvantage. DSAG criticized that AI must also be accessible to on-premises customers and should not depend on existing cloud contracts.

The cost and licensing risks associated with SAP are highly complex and often opaque. DSAG criticized the fact that the costs for development, quality assurance, and general operation of BTP services (in the consumption model) are too high. The issue of indirect use is seen as particularly critical when sensitive SAP data flows into external LLM models outside the SAP ecosystem. These unclear contractual and commercial risks must be clarified at an early stage, as an incorrect assessment can quickly lead to unexpectedly high costs.

AI exit: SAP Foundation Model

SAP is considering how the global market leader in ERP can be guided out of the AI dilemma: According to its own statements, SAP has developed the first basic model that is specifically designed for structured business data (ABAP tables). It is a machine learning model that has been pre-trained to understand table-based business data for a variety of forecasting tasks. It is called SAP-RPT-1 (RPT stands for Relational Pretrained Transformer) and, according to SAP, supports context-based learning. The specific ABAP tables are to form the basis for this. Whether ABAP modifications can also be incorporated into the machine learning model has not yet been answered.

„You simply provide the model with a few labeled table rows from the respective forecasting task, and it immediately delivers a highly accurate forecast,“ explained SAP Chief AI Officer Philipp Herzig. „A single model can fulfill an enormous range of business forecasts in the areas of finance, supply chain, human resources, and more by looking at examples from a different area in the respective context each time.”

However, many experts in the SAP community believe that SAP is struggling with AI applications because no consistent roadmap was developed years ago: SAP stumbled from the cloud era into the AI era.

Successful SAP cloud history

SAP recognized the strategic importance of the cloud early on—not least thanks to its former co-CEOs Jim Hagemann Snabe and Bill McDermott. McDermott currently heads the highly successful cloud company ServiceNow. Snabe, still CEO of Siemens in Munich, has a very good knack for AI.

This year, Jim Hagemann Snabe was the keynote speaker at Europe's most important AI event: Siemens was represented at Viva Technology in Paris in June (four days and 180,000 visitors). There wasn't much to see from SAP! In 2026, Germany will be the partner country of Viva Tech (this year it was Canada). I was already in Paris this year, and E3 magazine is planning a comprehensive media partnership with and coverage of Viva Tech for 2026. (Please send written inquiries in French or German to Dr. Ulrike Godler, ulrike.godler@b4bmedia.net).