SAP's finance chief warned that the current bulk of AI usage in enterprise environments concentrates on relatively low-risk applications - and that the industry must progress into more demanding process automation before broad productivity gains become apparent.
Speaking to reporters after SAP's second-quarter results, CFO Dominik Asam said the "lion's share" of AI token consumption today is absorbed by what he described as "low-hanging fruits" - primarily coding assistants and chatbots. In those settings, he noted, the consequences of AI hallucinations are limited because the outputs tend to carry lower business risk if they fail.
By contrast, Asam said, embedding AI into core functions such as finance and supply chain is materially more difficult. Errors in those workflows can propagate through multiple steps and create compounded issues that raise compliance and reliability concerns. "If you have some hallucinations in the process, the errors will actually compound statistically over many steps," he said, referring specifically to finance workflows. That, he added, "requires much more excruciating assurance levels."
Asam framed the more valuable opportunities - the "high-hanging fruit" - not as deploying a generic plug-and-play large language model across an enterprise but as designing governed systems tailored to specific business processes. That approach, he said, hinges on companies making their own data usable and governed so that AI can operate with corporate knowledge and established controls.
"The idea that AI will solve all these problems if they are messy, legacy data silos is not true," Asam said, adding that relying on that path produces "extremely high token costs." He cautioned that simply accessing the most advanced model is not always the correct choice; instead, organizations will favor the cheapest reliable solution that achieves the required outcome safely - whether that is conventional software, an open-source model or a pricier frontier model.
The comments underline SAP's stance that returns from generative AI are more likely to emerge from governed, business-specific implementations than from general-purpose models. For sectors such as finance, supply chain and enterprise software, the emphasis will be on data readiness, reliability and cost-efficient model selection rather than model size alone.
Impacted sectors: enterprise software, finance, supply chain and logistics.