SAP board sees a problem: "Germany's AI research is world class, but ..."

For Philipp Herzig, it is clear: good AI is invisible. "Our job is to hide the technology," says SAP's AI boss in the ntv podcast "So techt Deutschland". He is enthusiastic about German research. But the difficulties begin in the next step.

SAP is relying on artificial intelligence. In the fourth quarter, half of all deals now have AI components integrated, said SAP CEO Christian Klein at the end of January. The "SAP Business Data Cloud" is intended to make data usable for AI applications across company boundaries. But Philipp Herzig, Chief AI and Chief Technology Officer (CTO) at SAP, does not see AI as an end in itself: "Our job is always to hide the technology, because most people just want to use the added value without thinking about how something works."

That's why it's clear to him: good AI is invisible - and at most one click away. "The best use cases are those that the company can simply switch on, that are tightly integrated into the workflow and the tools of the employees." In practice, this means that a manager should not have to click through complicated software, but should be able to create a performance assessment or a financial analysis with a single command.

Does Germany need its own language model?

While OpenAI is currently opening its first office in Munich and is calling Germany the most exciting AI market in Europe, the question arises: does Europe or even Germany need to develop its own large language model? Herzig is skeptical: "On the one hand, in my view, we have two language models in Europe that we can use very well - Aleph Alpha and Mistral." However, it is more important to use the technology sensibly, "but I don't think it's a good idea to just copy what already works."

Nevertheless, Germany is a world leader in AI research, stresses Herzig: "The research is world class. No matter which university, you can go to Saarbrücken, the DFKI, the TU Munich, the LMU, the Hasso Plattner Institute or the Berlin universities."

The problem begins in the next step, adds the AI ​​expert, pointing to the topic of image generation as an example. The so-called diffusion models were invented at the Ludwig Maximilian University (LMU). "Who commercialized them and made them successful? Other companies, not necessarily German ones," says Herzig.

Research top, implementation flop

His conclusion: "We are in a great position in terms of research. The difficulty we have is: once we have done great research, what happens next? How are companies and start-ups founded? Do they get the money?" This is exactly where Germany needs to be more courageous.

AI is still primarily a tool - but SAP is already thinking ahead. In the future, says Herzig, AI agents will be able to independently formulate hypotheses, evaluate data and say with a certain degree of probability: "This is where the problem lies." A finance manager will then not only be able to ask: "How were our sales last week?" Instead, the AI ​​will be able to provide clues on its own as to where something has changed - before anyone asks.

Herzig's approach is not only to make AI easier to use, but also to integrate it into existing business processes. To organize the topic of AI like a small startup within SAP in order to then "scale it across the rest of the company and SAP's 400,000 business customers."

Frauke Holzmeier and Andreas Laukat spoke with Philipp Herzig . The conversation has been shortened and smoothed for better comprehensibility. You can listen to the entire conversation in the podcast "So techt Deutschland" .