Stay connected with KayaToday, follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
There is a peculiar tension sitting at the heart of enterprise AI right now. Most large organisations have moved past the pilot stage, they are deploying AI at meaningful scale, and a clear majority say they are satisfied with the returns. Yet in the same breath, two-thirds of those same leaders admit the technology is not delivering anywhere near its full potential. That contradiction is not confusion. It is, in fact, a fairly precise diagnosis of where the industry stands in mid-2025.
The SAP Value of AI Report 2026, produced with Oxford Economics and drawing on a survey of 2,600 business leaders across 13 countries, puts hard numbers to that tension. AI now supports roughly 30% of all tasks in the average organisation, up from 25% the previous year. General AI return on investment has climbed from 16% to 21% year-on-year, and is projected to reach an average of USD 15.9 million per organisation within two years. Meanwhile, ROI expectations specifically for agentic AI have jumped from 10% to 17% in a single year. The direction of travel is clear. The gap between current performance and what is theoretically achievable, however, remains wide.
Fragmented Adoption Is Capping Returns Before They Begin
The most structurally important finding in the report is also the least surprising to anyone who has watched AI roll out inside large corporations. More than half of organisations still invest in AI in an ad hoc or piecemeal fashion. Only 17% report a strategic, holistic approach to prioritisation, though that figure has nearly doubled from 9% a year ago, which suggests the lesson is slowly landing.
The fragmentation tends to originate at one of two failure points. In some organisations, board-level pressure to adopt AI arrives without a coherent strategy or adequate AI literacy to back it up, producing scattered skunkworks efforts that each solve a narrow problem in isolation. In others, a lack of board-level attention leaves employees importing their own tools and experimenting quietly, with no coordination across teams or functions.
“You end up with a lot of organic, disjointed AI initiatives that pop up, and they struggled sometimes just because of data quality,” said Sean Kask, chief AI strategy officer at SAP. “But even the initiatives taking a strategic approach are still working in silos, where they may have consistent data that works in that one use case, but they’re still not at the level where they’re transforming an entire business process.”
This explains the report’s most counterintuitive data point. The 69% of businesses that say they are satisfied with AI ROI are not wrong, exactly. They have proven the technology can generate returns. But the same learning process has made them aware of how much more value is theoretically available, and how far their current infrastructure falls short of capturing it. Satisfaction and ambition are coexisting, which is a healthier state than either complacency or disillusionment.
Agents Raise the Stakes on Data Quality
The next frontier in enterprise AI is agentic systems, and the economics here are genuinely significant. Unlike a conventional AI tool that responds to a single prompt, an agent can plan across multiple steps, call on various tools, and work iteratively toward a defined objective. SAP has shipped more than 400 AI use cases across its portfolio and is now extending into agents as the next layer.
Kask offered a concrete illustration. An accruals accounting task that typically takes a human accountant around 12 hours a month for a mid-sized company can be reduced to two or three hours with an AI agent currently in beta at SAP. Scaled across dozens of comparable processes, the cumulative time and cost savings become substantial. Yet only 3% of organisations in the survey say they are fully prepared for agentic AI, which means the gap between the technology’s readiness and organisational readiness is still enormous.
That unpreparedness connects directly to data. According to 73% of respondents, data quality and availability are now the primary reason organisations are not extracting more value from AI. Some 79% report that low-quality outputs cause rework, delays, or backlogs at least occasionally. The nature of the data problem has also shifted with the move to foundation models. Classic deep learning required organisations to find, extract, clean, and train on bespoke datasets. Large language models largely eliminate that burden, but they make preserving business context far more critical in its place.
“As soon as you extract data from an ERP system, you break all the contextual information, all of the semantics, and for generative AI, that’s the most useful part,” Kask said. SAP’s response is a knowledge graph embedded in its cloud ERP that maps 452,000 ABAP tables and 7.3 million data fields, allowing data products to carry their business meaning intact across both SAP and non-SAP systems.
Governance Is the Problem Most Organisations Have Not Noticed Yet
If data quality is the barrier companies know about, AI governance is the one most have not yet fully confronted. Only 12% of businesses say they are fully prepared to govern AI, while 69% acknowledge at least occasional use of unapproved shadow AI tools. As agentic systems proliferate, the governance gap becomes more consequential because agents can take actions, not just generate text.
“As companies roll out their AI initiatives, they often discover shadow agents, agents that can access data they shouldn’t or take actions they shouldn’t. The question then becomes: How do we audit these things?” Kask said. SAP’s AI Agent Hub addresses this by building an inventory of agents, large language models, and MCP servers across a customer’s technology landscape. Organisations using it have already surfaced thousands of SAP and non-SAP agents they did not previously know existed within their own systems. The platform then applies lifecycle management, identity and access controls, and performance monitoring on top of that inventory. Kask compares the discipline to standard employee onboarding, noting that most companies would never bring on a new hire without defining what data and systems that person is permitted to access.
Governance also has a human dimension that the report takes seriously. Almost 80% of respondents agree that maximising AI value requires more than technical upskilling, and 75% are already planning to reskill employees. The conversation in boardrooms is shifting away from which roles AI will eliminate and toward how people and AI systems can collaborate most effectively, given that agents still require human oversight, redesigned workflows, and sharper organisational judgment about when to intervene.
For businesses in Malaysia and Singapore, where digital transformation investment has accelerated sharply across sectors from financial services to manufacturing, the report’s core message carries direct relevance. The constraint on AI value is rarely access to the latest model. It is the unglamorous work of connecting AI to clean, contextually rich data, governing it properly, and redesigning processes around it rather than simply layering it on top of existing ones. The organisations that treat AI as primarily a technology procurement decision will keep hitting the ceiling that 67% of survey respondents are already bumping against. Those that treat it as an organisational redesign challenge, supported by technology, are the ones most likely to reach the returns the projections promise.
Read More: Instacart’s CTO Says AI Has Made Tech Debt Irrelevant. Here’s What That Actually Means
