Dear CIO,

David Linthicum is a respected enterprise technology leader, researcher, and industry commentator known for his work in cloud computing, enterprise architecture, and AI strategy. His analysis is widely followed by technology executives and delivery leaders because he focuses on the practical realities of implementation rather than the promise of emerging tools. In his recent survey, his perspective is especially useful because it frames AI ROI as an execution and operating-model challenge rather than a simple technology adoption problem. In today’s newsletter, we are going to dive into this report.

Best Regards,
John, Your Enterprise AI Advisor

A bad system will beat a good agent every time

The headline number gets the most attention: 84% of respondents said they haven’t seen measurable ROI from AI. However, that does not mean AI is not delivering value. It’s more accurate to say most organizations are stuck in the messy middle from past experimentation, but not yet at consistent, repeatable results. Out of 456 IT professionals surveyed, 72 reported ROI, and almost all of those wins came from tightly scoped, well-defined use cases. This distinction matters. AI can absolutely produce ROI, but it tends to show up where the problem is clear, the scope is narrow, and the work is grounded in real operations with the right data and ownership behind it. The survey points less to a failure of the technology and more to a gap in execution.

The biggest issue presented in this report? Weak business cases. About 11% of respondents pointed to this, which is not surprising. A lot of AI projects start with “Where can we use AI?” instead of “What problem are we actually trying to solve?” That shift sounds small, but it means everything in organizations. A flashy use case will not go far if there is no baseline, no clear owner, no adoption plan, and no shared definition of success.

Right behind that is data quality, with about 10.5% citing fragmented or unreliable data. This is the classic problem: organizations try to automate processes that aren’t working in the first place. AI depends on clean, consistent, accessible data. If your systems are scattered or incomplete, AI won’t fix that. It will just make the gaps more obvious.

Integration is another sticking point. Nearly 9% pointed to slow connections with legacy systems. This is where a lot of pilots stall out. AI does not create value sitting off to the side. It needs to be embedded in the tools people already use. If it is not connected to systems like ERP, CRM, or finance platforms, it never really becomes part of the workflow.

Then there is ownership, or the lack of it. About 8.3% said there is no clear product owner or accountable model. That is both a governance issue and a value problem. AI initiatives tend to span multiple teams, and without someone clearly responsible for outcomes, adoption, measurement, cost, and improvement, they drift. When everyone owns something, no one really does.

Cost is another reality check. Around 7.2% said expenses outpaced value at scale. AI can look cheap in a pilot, but production is a different story. Infrastructure, inference, monitoring, compliance, and talent add up fast. Without a clear handle on those costs, something that looks promising early on can become hard to justify later.

The rest of the list keeps reinforcing the same pattern. Security and compliance delays (6.8%), low user trust (6.4%), and weak change management (6.1%) all show up as blockers. None of these is about the model itself. They are about the environment the model is dropped into.

This is where Deming’s idea really hits: a bad system will beat a good person every time. The same applies here: a bad system will beat a good model. If the workflow is unclear, the data is shaky, ownership is vague, and users don’t trust the output, AI will not fix the problem. It will just move it faster. Even the smaller factors tell a familiar story. Tool sprawl (4.8%) creates unnecessary complexity. Talent gaps (4.6%) slow progress. Unrealistic expectations (3.9%) don’t help either.

Put it all together, and there is no single root cause holding back AI ROI. It is a chain of very normal, very fixable issues: unclear priorities, weak data foundations, poor integration, fuzzy accountability, and not enough focus on change management or cost discipline.

So the takeaway isn’t “AI doesn’t work.” Organizations must not rush to automate things that are not working yet. Before introducing AI, it’s worth asking a few basic questions: Is this process worth improving? Is the data ready? Do we understand the workflow? Will people actually use this? Can we measure the outcome? And most importantly—who owns it?

AI can create real leverage, but leverage cuts both ways. In a strong system, it accelerates value. In a weak one, it amplifies confusion and cost. That might be the most useful lesson from the survey: getting to AI ROI starts well before the model ever shows up. It starts with fixing the system around it.