Here's the uncomfortable truth buried in SAP and Oxford Economics' just-released Value of AI Report 2026: most enterprises are already losing control of their AI agents — and they know it.
The study surveyed 2,600 business leaders across 13 countries. The headline numbers look impressive. ROI from AI is growing. Investment is accelerating. Optimism is high.
But dig one layer deeper, and a troubling picture emerges. 69% of enterprises say they are either unsure or believe they are deploying AI agents faster than they can govern them. That's not a technology problem. That's a governance crisis hiding inside a growth story.
The ROI Numbers Are Real — And Getting Bigger
Let's start with the good news, because it's genuinely significant.
The average global enterprise spent $28 million on AI this year. That investment is generating a 21% ROI — roughly $6.3 million in returns. That's up from 16% last year. Not explosive, but a meaningful year-over-year improvement.
Two years from now? The same businesses expect that ROI to hit 38%, translating to roughly $15.9 million in returns on the same base investment. That's a 2.5x improvement in returns from where we are today.
But the real growth story is agentic AI. Expected ROI from agentic AI alone is projected to jump from $4.3 million today to $17.6 million within two years. That's a 4x increase. In enterprise terms, that's the difference between a pilot program and a strategic transformation.
For CFOs running AI investment cases, this data matters. The payoff window isn't 5-10 years out. The benchmarks from this study suggest meaningful returns within 18-36 months for well-executed deployments — and faster still for organizations that deploy agentic AI at scale.
So Why Can't Enterprises Capture It?
Here's where the data gets sobering.
Only 3% of businesses say they are fully prepared for agentic AI. Three percent. The majority describe themselves as either partially prepared or not prepared at all — yet 83% say agentic AI has moderate to very high potential to transform their organization.
That gap between belief and readiness is where billions in expected ROI will evaporate.
The SAP/Oxford Economics report identifies three structural barriers that are holding enterprises back right now.
Barrier 1: Governance Is an Afterthought
The numbers here are stark.
- 38% of companies have no human-in-the-loop process for agentic workflows. Agents are making decisions, taking actions, and executing transactions without a human review step.
- 37% have no permission or access controls for agents. There is no mechanism to define what an agent can and cannot do.
- Only 44% maintain a registry of the agents operating across their business. More than half of enterprises don't even know what agents they have running.
Only 12% of businesses say their skills, processes, and frameworks are fully ready to govern AI effectively.
In conversations with CIOs managing large-scale AI deployments, the pattern I hear most often mirrors this data. Agents get stood up quickly — often by individual teams, sometimes without central IT involvement — and governance gets deferred to "once we've proven the value." The problem is that by the time value is proven, you have dozens or hundreds of agents operating without oversight, and retrofitting governance is exponentially harder than building it in from day one.
BCG's recent research on agentic AI strategy echoes this directly: governance chaos — unclear ownership over who manages what agent — is one of the top three killers of agentic AI programs. Not the technology. Not the cost. Unclear ownership.
Barrier 2: Shadow AI Is Already Inside Your Organization
The report doesn't soft-pedal this one. 69% of businesses say shadow AI use is happening at least occasionally. That number is up year-over-year.
Shadow AI — employees using AI tools that haven't been approved, reviewed, or integrated into enterprise systems — isn't just a compliance risk. It's a data risk. It's a quality risk. And in an agentic world, it's becoming an operational risk.
The distinction matters. Shadow use of a generative AI chatbot means someone is getting answers from an unapproved source. Shadow use of an agentic tool means someone is giving an AI system the ability to take autonomous actions — book meetings, send emails, execute transactions — without any enterprise oversight or guardrails.
Gartner has flagged this in their own projections: 40% of CIOs will demand "Guardian Agents" by 2028 — dedicated systems whose sole job is to track and contain the actions of other AI agents. That's not a product feature. That's a response to governance failure at scale.
For CISOs and compliance leaders reading this: the question isn't whether shadow AI is happening in your organization. The question is whether you have visibility into what those agents are doing.
Barrier 3: Your Data Isn't Ready
AI agents are only as good as the data they operate on. And right now, most enterprise data isn't ready.
73% of companies report challenges with incomplete data. That's a drop from last year — meaning the data readiness problem is getting worse, not better, even as AI investment increases.
The downstream effect is visible: 79% of businesses are experiencing rework, delays, or backlogs caused by low-quality AI outputs. Nearly eight in ten. That's not an AI problem — that's a data infrastructure problem manifesting as an AI problem.
When agents operate on incomplete or inconsistent data, they don't just produce wrong answers. They take wrong actions. At scale. With downstream consequences that can ripple across systems.
The relationship between data quality and agentic AI ROI is direct: better data governance → more reliable agent behavior → higher actual ROI versus projected ROI.
The Leadership Gap Underneath Everything
One data point from the SAP report deserves its own paragraph because it explains almost everything else.
Only 46% of companies have a dedicated AI leader responsible for AI adoption.
In nearly half of all enterprises surveyed, there is no single person or role accountable for how AI is governed, deployed, and scaled. AI strategy is either distributed across functions — where it gets fragmented — or assigned as a secondary responsibility to an existing role, where it competes with other priorities and rarely wins.
Only 17% describe their AI approach as strategic. The majority — 41% — are still operating with piecemeal, use-case-by-use-case deployments without an overarching framework.
This explains why the ROI gap between early movers and laggards is widening. Organizations with formal AI strategies, dedicated AI leadership, and governance frameworks embedded from the start are capturing disproportionate returns. Those without them are spending more, getting less, and dealing with growing compliance exposure.
What the 3% Are Actually Doing
The 3% who describe themselves as fully prepared for agentic AI aren't waiting for governance to catch up with deployment. They're building governance into the deployment architecture from day one.
From conversations with enterprise AI leaders and cross-referencing the SAP data, the patterns that distinguish this cohort are consistent:
Dedicated AI leadership with real accountability. Not a committee. Not a rotating responsibility. A single accountable executive — Chief AI Officer, VP of AI, or equivalent — with a mandate that includes governance, not just adoption.
An agent registry maintained in real time. Every agent deployed in the enterprise has a documented owner, defined scope, access controls, and a review cadence. This isn't bureaucratic overhead — it's the minimum viable governance layer for an agentic environment.
Human-in-the-loop checkpoints for high-stakes actions. Agents are authorized for low-stakes, high-frequency tasks autonomously. High-stakes, low-frequency decisions have human review gates. The risk framework determines the boundary, not blanket automation.
Data readiness as a prerequisite, not an afterthought. Before an agent is deployed to a new workflow, the data it will operate on is audited for completeness and consistency. Agents aren't used to paper over data quality problems — they're deployed into workflows where the data is already trustworthy.
The Business Leader's Frame
For CFOs, COOs, and business-side executives evaluating AI investments: the SAP/Oxford Economics data gives you a useful benchmark.
Your peers are spending $28 million on average. They expect 21% ROI this year and 38% in two years. If your organization is below those thresholds on returns, the gap likely traces back to one of three places: data quality, governance maturity, or leadership accountability.
The agentic AI ROI projection — $4.3M today to $17.6M in two years — represents a significant opportunity. But that projection assumes organizations will actually be able to govern the agents they deploy. The same study reveals that most are already struggling to do so.
83% of business leaders believe agentic AI will transform their organization. Only 3% are fully prepared to govern it. That 80-point gap is where the real strategic risk lives.
The companies that win in an agentic AI era won't necessarily be the ones who deploy the most agents. They'll be the ones who can govern them — reliably, at scale, across the enterprise.
What To Do Now
Three specific actions that don't require a multi-year transformation program:
1. Audit your agent landscape this quarter. Build the registry you don't have. Identify every AI agent operating in your organization — vendor-deployed, internally built, or team-initiated. Document owner, scope, and access permissions. If you can't build this list in 30 days, your governance problem is already significant.
2. Define your risk tiers before deploying new agents. Classify actions by risk level before automating them. Customer-facing communications, financial transactions, and data access decisions need human review gates. Internal scheduling and data aggregation can run autonomously. Define the line explicitly, not case-by-case.
3. Assign a single accountable executive for AI governance. Not a committee. Not a shared responsibility. One person whose job includes knowing what agents are running, what they're doing, and what the escalation path looks like when something goes wrong.
The SAP data is a useful mirror. Globally, AI ROI is growing and the potential for agentic AI is enormous. But the study is also clear that most enterprises are building on an unstable governance foundation. Fixing that foundation is the highest-leverage investment a business leader can make in AI right now.
Source: SAP Value of AI Report 2026, conducted with Oxford Economics, surveying 2,600 business leaders across 13 countries.
Continue Reading
- The 3x ROI Gap: Why Your AI Strategy Isn't Working
- 71% of Your 'AI Agents' Are Just Chatbots in Disguise
- 88% of Companies Use AI. 95% See Zero ROI. Here's Why.
- Agentic AI Just Crossed the Enterprise Adoption Threshold: What Q2 2026 Data Tells Us
- 95% Have AI Strategy But Only 8% See ROI: What Winners Do
