SAP relies on its own AI solutions instead of expensive language models
Quick Look
- SAP boss Philipp Herzig reports that three years ago the company rejected the idea of its own large language model and instead relied on cheaper solutions such as table models from Prior Labs.
- The new AI strategy under CEO Christian Klein is intended to improve the customer experience and justify the high prices.
AI-generated summary
Why It Matters
SAP has long struggled with its response to the AI boom, as American language models have spread rapidly since 2023 and threatened its own business model. The implementation of our own AI announcements was slow.
San Francisco, Palo Alto, Düsseldorf. San Francisco, mid-September. Philipp Herzig stands on stage at the Terra Gallery in the hip South of Market district and talks about a no that once almost cost him his job. Around three years ago, as SAP's head of technology describes it, someone suggested at a meeting of the supervisory board that we build our own large language model for artificial intelligence (AI) - like OpenAI, Anthropic or Google, whose inventions spread like wildfire at the time.
Herzig thought that was nonsense: it was “the stupidest idea on the planet,” he said at the time. “Some people wanted to fire me for that,” Herzig says on stage.
In the end, things went well for him. SAP now buys language models from the competition; the head of technology, who was still the head of AI at the time, disparagingly calls them “mass-produced goods”. Instead, SAP's money flows to where the company can keep up with less money and where GPT or Claude have so far disappointed: in adapting the models to specific business processes or so-called table models, which process business data particularly quickly and cheaply. This evening, Herzig will present the latest model from the Freiburg laboratory Prior Labs, which SAP took over in the summer.
The appearance in California is part of the departure into a new phase for the software company, which was once the most valuable company in Germany. SAP struggled with its response to the AI boom for a long time: The language models from America would spread with force from 2023 at the latest and threatened to pose a threat to its own business model. At the same time, the implementation of the announcements, such as that of our own digital assistant, was slower than expected. Customers and advisors became increasingly impatient.
But the counteroffensive by SAP boss Christian Klein is now starting to have an impact: table models from Prior Labs, a task force based on the model of the founding years and a completely revamped AI assistant Joule should finally bring the experience for customers back to the top level that SAP's high prices lead one to expect. The staff should be trained for the future of AI in innovative workshops.
Handelsblatt researched the company, met top managers in California and interviewed critical customers. When it comes to AI, one thing is certain: Klein's leadership doesn't want to waste any more time.
What to Watch
AI outlook — possibilities, not facts
SAP will further reduce its spending on its own language models and instead rely on specialized business process models.
Likely · Within months
The new AI assistant Joule will receive a major update next quarter that improves integration into business processes.
Possible · Within months
Open Questions
- How much does SAP actually invest in adapting AI models?
- When will the new AI assistant Joule be widely available?
- How do customers react to the new strategy compared to pure language models?
