The numbers in SAP's latest report carry a quiet kind of tension. On one hand, AI now supports nearly a third of all tasks in the average organization, up from a quarter last year, and general ROI expectations have climbed to 21%. On the other, two-thirds of business leaders say they are satisfied with the returns they've seen, and yet the same share remains convinced the technology isn't delivering its full potential. That isn't a contradiction so much as a dawning awareness: the first wave of AI value was the easy part, and the gap between where companies are and where they could be has less to do with model capability than with the unglamorous work of strategy, data, and governance.
What stands out is how much of the friction is self-inflicted. More than half of organizations still invest in AI in an ad hoc, piecemeal way, and while the share taking a holistic approach has nearly doubled to 17%, that still means most companies are running multiple disjointed experiments that never cohere into business transformation. The report points to board-level pressure to adopt AI without the accompanying literacy or roadmap, which produces scattered skunkworks and shadow tools. This connects to a broader pattern we've covered in Clean Data Starts With Catching AI Slop Before It Skews Your Model, where the quality of what flows into your systems determines what you can trust coming out of them. Similarly, Transparency in AI Voice: ElevenLabs CEO on Disclosure and the Future shows how quickly the novelty of AI capabilities gives way to the harder question of accountability. The lesson is consistent: the technology isn't the bottleneck; the surrounding discipline is.
The most useful reframing here is around agentic AI, where expectations for ROI jumped from 10% to 17% in a single year. That's not hype; it's arithmetic. SAP's beta agent for accruals accounting cuts a task that typically takes 12 hours a month down to two or three. Scale that across finance, supply chain, and HR, and the economics become obvious. But the report is equally clear that agents only deliver when they're connected to contextually rich data and governed at scale. Data quality is now the top barrier to AI value, with 73% of respondents naming it, and 79% report rework or delays from low-quality outputs. The deeper issue is that extracting data from an ERP system breaks the semantics that make generative AI useful. That's why SAP's knowledge graph, mapping 452,000 ABAP tables and 7.3 million data fields, matters more than any single model release. The value isn't in the algorithm; it's in the meaning you can preserve around the data.
For our readers, the practical takeaway is blunt: if you're not investing in data context and governance alongside your AI pilots, you're not behind on technology, you're behind on the conditions that make it work. The report's finding that only 12% of businesses feel fully prepared to govern AI, while 69% admit to shadow AI use, should be the detail that focuses your next board meeting. The question isn't whether agents will reshape workflows; it's whether you'll be auditing them from day one or discovering them after they've already started acting on unapproved data. SAP's own AI Agent Hub is already surfacing thousands of agents inside customer landscapes that no one knew existed. That's the moment to watch: not the next model release, but how quickly companies move from celebrating ROI to building the governance that makes it sustainable. The companies that close that gap won't just see better returns; they'll be the ones defining what the next decade of work looks like.