The boardroom debate about whether AI delivers real returns is settling. SAP and Oxford Economics just dropped the 2026 edition of their Value of AI report, and the headline result is hard to argue with: AI now supports 30% of all tasks in a typical business, up from 25% last year. ROI expectations are rising in parallel, but one number keeps standing out: only 3% of organisations say they are fully prepared for what agentic AI is about to do to their operations.

The survey covered 2,600 business leaders across 13 countries, including Australia, Brazil, Canada, China, France, Germany, India, Italy, Japan, Singapore, Thailand, the UK, and the US. At that sample size, across that many markets, the findings are hard to dismiss as anecdote.

What the Numbers Actually Say

Average AI spending across the surveyed businesses rose to $28 million this year, up from $26.7 million in 2025. That is notable less for the dollar amount and more for the trajectory. Businesses are not pulling back after initial experiments. They are leaning in.

On the return side, companies expect to generate an AI ROI of 21% this year, up from 16% last year. In absolute terms, that means roughly $6.3 million in expected return on the average $28 million invested. Two years from now, that ROI is projected to reach 38%, translating to $15.9 million.

The agentic AI numbers are more striking. Last year, businesses estimated they would get about $4.3 million in two-year returns from agentic AI. This year, the same two-year estimate has jumped to $17.6 million, more than four times higher. Expectations have not just grown. They have accelerated.

The Gap Between Expectation and Readiness

Here is where the report becomes a mirror. 83% of respondents said agentic AI has moderate to very high potential to change their organisation. And yet only 3% say they are fully prepared for that change.

That is not a rounding error. That is a gap between what business leaders believe AI will do and how much they have actually done to get ready for it.

The underlying reasons are predictable but still worth naming. 73% of companies reported challenges with incomplete or siloed data, the kind of data problem that does not show up until you try to run agents against real workflows. 69% acknowledged occasional to frequent use of unapproved AI tools across their teams. And only 12% said they are fully prepared to govern their AI systems.

Shadow AI is not a new problem, but it is a governance problem with compounding consequences. Every unapproved tool in use is a system that is outside the risk and compliance framework businesses think they have.

What This Means for Business

The ROI numbers are real. If you are running any material amount of AI in your business and you are not capturing returns that look something like these industry averages, the problem is almost certainly not the technology. It is the infrastructure around it: data quality, governance, integration, and the ability to actually connect AI to the workflows that matter.

The 3% readiness figure is the one worth sitting with. Businesses that are not prepared for agentic AI are not in a holding pattern. The market is moving, competitors are deploying, and the gap between ready and not ready is widening with every quarter.

For data-driven organisations, the task support metric tells a useful story on its own. If AI is now handling 30% of work tasks on average, and the projection is 48% within two years, then the question is not whether your people will be working alongside AI agents. The question is whether your organisation will have built the data and governance foundations to make that productive rather than chaotic.

The businesses that close the 2026 readiness gap fastest, getting their data in order, bringing shadow AI under governance, and building repeatable frameworks for deploying agents, are the ones the 2028 version of this report will show winning on ROI.

For those still figuring out where to start, the answer is almost always the same: good data before good tools.

---

The full SAP and Oxford Economics Value of AI 2026 report is available at the Oxford Economics website.