SAP’s chief financial officer is pushing back on the idea that today’s most popular artificial intelligence tools are delivering the kind of sweeping business improvements many companies are hoping for — and he says the real payoff will only come when AI moves into more demanding territory.
CFO Dominik Asam made the remarks to reporters following the release of SAP’s second-quarter financial results. He said the vast majority of AI usage right now is concentrated in what he called “low-hanging fruits” — things like coding assistants and chatbots — where mistakes by the AI carry relatively little consequence if something goes wrong.
The bigger challenge, Asam explained, is putting AI to work inside critical business operations such as finance and supply chain management. In those environments, a single error doesn’t just disappear — it can ripple through many steps in a process, creating compounding problems and raising serious compliance concerns.
“If you have some hallucinations in the process, the errors will actually compound statistically over many steps,” Asam said, referring specifically to finance workflows. “It requires much more excruciating assurance levels.”
He described this more demanding application of AI as the “high-hanging fruit” — and said achieving it is less about dropping a generic, off-the-shelf AI model into a company’s operations and more about carefully building systems tailored to that specific business.
A key part of that effort, Asam said, is getting a company’s own data in order. AI systems need clean, well-organized, and properly managed data to function effectively within a business context. “The idea that AI will solve all these problems if they are messy, legacy data silos is not true,” he said, adding that poorly structured data also leads to “extremely high token costs.”
Asam also pushed back against the assumption that companies should always reach for the most powerful AI model available. In reality, he said, businesses will gravitate toward whichever tool — whether simple software, an open-source model, or a high-end frontier model — can get the job done safely and at the lowest cost.
