SAP Abap Machine Learning

AI training material from SAP Abap tables

The whole drama of the AI roadmap at SAP can be reduced to a simple formula: There is no training data! This unfortunate circumstance also seems to be the reason why SAP never considered an LLM (Large Language Model) for ERP. The power of an LLM is mainly based on existing training data, see also the numerous copyright processes in the USA, Germany and the UK.

Workday has done it better: as a native cloud provider for people management and finance, Workday has asked its existing customers for permission to use anonymized data for training purposes from the outset. Initially, the data was analyzed using statistical methods and machine learning, but Workday is now working on LLMs and AI agents. Every IT company is working on and with AI agents, but without training data, it's like learning to swim on land.

SAP Predictive Analysis Library

SAP has a long tradition of statistical analysis of ERP, PLM and SCM. Many problems in logistics and production can be elegantly solved using sophisticated statistical methods and rules. Over time, a library of function modules for ERP analysis has been developed.

Naturally, machine learning is also a suitable means of analyzing ERP Abap tables. However, an AI system for ERP based on reinforcement learning would be even more exciting. This would produce clear results similar to chess and Go: the SAP customer has won or not.

However, reinforcement learning for ERP would be a completely different challenge and would require significantly more AI commitment from SAP. According to SAP insiders and well-informed sources, the global ERP market leader is currently making do with simple machine learning for „Simple Finance“ and „Simple Logistics“ based on Abap tables.

Abap tables, SAP BTP and Steampunk

SAP's decision to intertwine the two domains of AI and Abap is a good move: Abap has not only emancipated SAP R/3, but will also live on in the Business Technology Platform (SAP BTP) for many years to come. Many SAP partners are currently working on IT projects to transfer Abap knowledge from older ERP installations to the new S/4 world. A small language model or machine learning based on this domain should deliver significant added AI value.

According to SAP experts, however, this AI Abap development is at an early stage: whether machine learning with Abap tables will become part of SAP BTP has not yet been decided, similar to the GenAI hub on BTP. SAP also has no idea yet about a pricing model. Ultimately, however, it will benefit Steampunk, embedded Abap and all other efforts to achieve an orderly ERP modification (after Clean Core) and the expansion of ERP add-ons.

The AI reputation of SAP S/4 Hana is dwindling

Ultimately, SAP has realized that it will not work without its own AI, its own small or large language model, machine and reinforcement learning as well as analytics, see also SAP PAL (Predictive Analysis Library). For a long time, SAP believed that it could benefit from the research results and investments of other IT companies - now SAP has to put its own money where its mouth is and carry out AI innovation work.

SAP is thinking late about its own small or large language model based on its own Abap tables. It will probably be more of a „machine learning“ model based on the SAP Predictive Analysis Library. AI with Abap tables will remain a generic model because SAP was far too late in providing itself with the opportunity to access ERP training data from existing customers. The global ERP market leader has not even dared to publish a roadmap for Abap machine learning.

AI still seems to be an unknown domain for SAP. SAP CEO Christian Klein seems to have recognized that ERP also needs something with AI, similar to cloud computing, but the ERP world market leader lacks a clear AI roadmap over several years. SAP Predictive Analysis Library (PAL) is not enough, Abap Machine Learning is too simple and an ERP LLM is too big a deal for the small SAP (compared to Microsoft, Meta, Oracle or Amazon).