Indirect Use of SAP Data and AI

I must confess that my wife has often reprimanded me in this regard: when it comes to contracts, legal issues and stumbling blocks, general terms and conditions and unfair competition, I am naive, clueless, disoriented—but also disinterested. I have my wife, who is extremely conscientious and meticulous, and I have our chief legal advisor. However, in the dawning age of AI, and with an SAP-like history of indirect use, I will have to change my defensive attitude toward all things legal.

Anyone who has been paying attention to the AI discourse over the past few months can already see the greed for data and the challenges of sustainable learning. LLMs (Large Language Models) are a computer science construct that is capable of learning in the sense of statistical functions and algorithms, but of course not in the sense of understanding. This leads to an immense hunger for data; the more data a system receives, the smarter it becomes.

Aleph Alpha co-founder Jonas Andrulis raved about the potential of his Large Language Model and Signavio's process mining results at an AI event organized by the German newspaper Handelsblatt. This combination sounds tempting—but what about privacy and intellectual property? The business processes of an ECC or S/4 were created by customizing and modifying SAP customers. Do I want an LLM trained in our business processes to give clever answers to a competitor?

Of course, this selfish approach flies in the face of any open source idea, but the processes contain not only our IP, but also algorithms and data from partners, customers, and suppliers. Salesforce added powerful AI capabilities to its successful CRM a few years ago. How did they do it? Anyone can read about it in Salesforce's T&Cs. There, the cloud provider secures the right to access all data in an anonymized (!?) manner for the purpose of training an AI model. This is not an isolated case: the HCM provider Workday functions in a similar way. The “stupid” computer has to learn somehow, right?

In addition to GDPR, IP, and open source, the classic SAP customer is also confronted with the issue of indirect use. Jonas Andrulis' vision can only succeed if data and algorithms are freely exchanged between ECC or S/4, Signavio and the Aleph Alpha LLM. If SAP would keep the indirect usage fee scheme, every ERP LLM would be unaffordable. If SAP were to suspend indirect usage for LLMs, it would set a precedent for other AI applications. The fact is that SAP has shot itself in the foot on the issue of indirect use, which could have disastrous consequences in the age of AI.

According to media reports from the Technology Days conference in Hamburg, held by the DSAG (German Speaking SAP User Group), SAP is aware of this dilemma. There is an AI vision at SAP, but how it will be realized is written in the stars. Is SAP too small for the mammoth topic of AI? At the moment, it does not look very promising: SAP's investments in artificial intelligence, directly or indirectly in Aleph Alpha, are marginal compared to the sums at Microsoft, Meta, Amazon, and Google. From a European perspective, we often misjudge the balance of power. I took the liberty of copying a chart from Manager Magazine on different companies’ estimated value.

The most valuable companies in the world—Europe and Germany by market capitalization—in trillions of dollars.
As of January 30, 2024. Source: Manager Magazine.

In view of this chart, SAP urgently needs an activist leader like former Deloitte boss Punit Renjen—however, this vision will now amount to nothing. Professor Hasso Plattner damaged his own life's work by firing the designated chairman of the supervisory board last year, but that's another story for another time.