SAP AI Dystopia
Indirect use prevents success with AI
To reduce self-inflicted ERP complexity, SAP introduced a new pricing model in 2018. This is intended to help customers determine the need for additional licenses for indirect use of their own data. The legal basis is controversial.
An example: an SAP customer uses third-party billing software that automatically transfers data to an SAP solution. This data connection may be subject to licensing by SAP. From time to time, SAP sends out lists of programs that SAP considers to be indirect data consumers. SAP customers typically have to search their IT infrastructure for data connections between their own SAP systems and other software. This is tedious and time-consuming, but absolutely necessary. SAP is on the lookout and can charge exceptionally high additional license fees for indirect use.
Ultimately, the sleight of hand is easy to see through: SAP does not want its own existing customers to use ERP, CRM, HCM, SCM, etc. software from other providers. However, because SAP ERP is usually the leading system, all third-party systems are indirect users of the data generated and available in SAP. What this legally controversial view means for future generative AI has yet to be discussed.
Generative AI thrives on external data
A characteristic of generative AI is that the systems are empty and stupid at the beginning. With a good amount of training data, these AI systems finally wake up. In the end, it is quite complex statistics that classify and feed back into the data. The fact is, without training data, there is no generative AI.
But the same is true for a Salesforce CRM system connected to an SAP ERP system. If that third-party CRM was not fed data from the SAP system, there would be no relationship management with end customers. ERP and SCM and CRM and HCM are interdependent—they help each other indirectly, but only SAP wants to charge licenses for this lifeblood data transfer.
While the exchange of data between SAP and Salesforce can still be largely controlled and minimalist, generative AI is much more chaotic and voluminous. The more, the better.
Abap fails due to lack of data
The more training data an AI is fed, the more intelligent it can become. One example: SAP is still failing to develop a co-pilot for Abap, SAP's own programming language. There is far too little Abap code in the world to train a generative AI like OpenAI's ChatGPT. SAP itself does not have a large language model that could be used to create an Abap co-pilot with the functionalities of generative AI.
However, the official lack of an Abap co-pilot only says something about SAP's own lack of drive and limited AI resources. Where there's a will, there's a way! According to unofficial sources, SAP partner Microsoft already has an experimental Abap co-pilot, and Microsoft surely does not have as much access to Abap code as SAP does, now does it?
Indirect use vs. AI training data
The SAP construct for indirect use of data generated and stored in SAP systems is in all likelihood the death knell for any generative ERP AI. Without sufficient training data, SAP customers face an AI dystopia and financial ruin. Licensing all the data in an SAP ERP system for indirect use would likely overwhelm any customer. With the concept of indirect use, SAP is the enemy of innovation for its own customer base.
Unlocking the treasure trove of ERP data
The current situation is ambivalent because SAP is making great efforts to provide comprehensive ERP data. With the acquisitions of Signavio and LeanIX, SAP ERP systems can be analyzed at a much higher level, which will ultimately produce optimal training data. Aleph-Alpha co-founder Jonas Andrulis raved about the combination of process mining with Signavio and a large language model (LLM) from his company at an AI conference organized by the German newspaper Handelsblatt.
The SAP company Signavio could use ERP process mining to provide the training data for Jonas Andrulis' Aleph Alpha LLM if the sword of Damocles of indirect use wasn't hanging over it.