Article

The CEO Who Came Back from the Conference

A few months ago I sat across from a CEO who had just gotten back from an industry conference. He was genuinely excited. He had seen demos, sat through keynotes, talked with peers at dinner. The word he used most often was “transformative.” He was convinced AI was going to change his business, and he wanted to know how fast we could move.

I asked him what he had in mind.

He said, “We’ll use AI.”

I waited. He waited. I asked again, gently, what specifically he wanted to use it for. Where in the business. What problem. What outcome.

He did not have an answer. He just knew he wanted to.

This is not a story about a CEO who was out of his depth. He was a capable executive running a healthy business. He had read the articles. He had talked to his peers. The problem was that the information available to him, mostly from vendors, conferences, and case studies, was not the kind of information that turns into a strategy for his business.

The operating model that keeps breaking

What made this conversation different from productive strategy conversations I have been part of was what came next. The AI strategy was going to get worked out in leadership meetings I was not going to be invited to. Once the strategy was ready, it would be handed to the technology organization and they would be expected to produce it.

I have seen this operating model before in a lot of organizations. Business leadership builds a strategy in one room. Technology is handed the strategy and told to make it happen. It rarely works in mainstream enterprise software. It almost never works with AI.

I am not the only one seeing this pattern. McKinsey’s research on the companies pulling ahead with AI found that what separates them is not better technology. It is executive sponsorship and cross-functional collaboration, and the senior leaders it interviewed named collaboration across the teams that have to deliver the work as the hardest barrier to clear. That barrier is exactly what gets created when strategy happens in a room that technology is not in.

The reason this breaks for AI specifically is data.

Why AI exposes the alignment problem faster

Traditional enterprise software can tolerate a certain amount of disconnect between the strategy room and the technology organization. You can define a goal like “improve customer service response times,” hand it to IT, and they can evaluate tools, pick a platform, and roll it out. The data in the process is mostly shape-of-task data. It lives in the system you are buying.

AI is different because AI runs on your data. Models and agents produce value by operating on what already exists inside your organization. That means someone has to understand what data exists, where it lives, how clean it is, how it connects across systems, what is sensitive, what is regulated, and what it would take to make any of it actually ready for a model to use. That knowledge sits with technology, not with leadership. You cannot build an AI strategy without it.

MIT research cited in industry reporting found that 95 percent of enterprise AI pilots fail to reach production, often because of unfit or untrusted data. Ninety-five percent. Not for lack of ambition. Not for lack of budget. They fail because the data foundation was not in place, and the strategy was built without anyone checking whether it could be.

Google Health saw a version of this in Thailand. Its researchers built a deep learning system that could read a retinal photo and flag diabetic retinopathy, a leading cause of blindness, with accuracy on par with specialists in the lab. Thailand has far more people with diabetes than it has eye specialists to screen them, so the promise was real. Then they put it into eleven clinics. The model had been trained on clean, high-quality images and was built to reject anything below that bar. In a real clinic, nurses were photographing dozens of patients an hour in whatever lighting they had, and the system rejected more than a fifth of the images as unusable. Those patients were told to come back another day. Slow internet added a minute or more to every upload. The model was good. The conditions it needed to do its job were not there.

That is the whole problem in one example. The model worked and the goal was sound, and it still created delays and frustration because no one had checked it against the conditions of a real clinic: the lighting, the connectivity, the pace of a nurse screening patients all day. The people who knew those conditions, the nurses and the local staff, were not part of designing the rollout. They met it after it had already arrived.

The questions the CEO could not answer alone

If he had invited me into the strategy conversation, here are the four questions I would have asked him. They are not technical. They are business questions, but they cannot be answered without technology at the table.

Where in the business are we losing time or money today that a model could actually address? Start with the friction. Document review. Customer support triage. Claims processing. Report generation. Sales lead scoring. These are the places where AI might help, because today’s process already has a problem worth solving.

What data would a model need to do that well, and do we have it clean enough to use? This is where most AI projects die. Your business data lives across a dozen systems, is inconsistent, is locked in PDFs and emails, or is only partially captured. The preparation work, sometimes called data readiness, often takes longer and costs more than the AI work itself. Technology has to tell you what is actually possible on your current foundation.

What happens to the humans doing that work now, and are we prepared to manage the transition? AI rarely eliminates roles cleanly. It shifts the work. The people leading the impacted functions have to help design what the new workflow looks like. That conversation has to include operations, HR, and the managers of the affected teams.

Who owns governance when the model makes a bad call? Models hallucinate, encode biases, and drift over time. Someone in your organization needs to own what happens when that occurs. Policy, review, escalation, sign-off. In regulated industries this is not optional. In unregulated industries it is still the difference between a controlled rollout and a reputational crisis.

If the strategy room is answering those questions without technology present, the answers are guesses. Confident guesses, maybe. Well-intentioned guesses. But guesses.

What the CEO did next

I offered to sit in the strategy conversations he was already planning, to bring a technology perspective to the four questions above alongside the business perspective he and his team were bringing. He declined. Not rudely. He just preferred to keep the strategy work closed, produce a plan, and then bring it to technology for execution.

We parted professionally. I do not know what they produced. I do know what they were working with, which was a set of assumptions about the business’s readiness for AI that had not been tested against the data, the workflows, or the people who would have to adopt it.

The pattern I see when companies try to scale AI is the same one. The hard parts are not technical. They are the constant churn of changing tools and processes, ongoing disagreement about what actually counts as value, and people inside the company reacting to the change at very different speeds. Every one of those is a cross-functional problem. None of them gets solved in a room with only business leadership in it.

What to do before your next executive discussion about AI

If you are an executive thinking about AI for your business, and you want to avoid being the CEO in this story, here is what I would do before the next conversation.

Write down three specific business problems where you would want AI to help. Not “improve customer experience” or “drive efficiency.” Specific enough to measure. Specific enough that your team could tell you whether AI is a fit.

For each one, write down what data the system would need, whether you have that data today, and how clean it is. If you cannot answer this, that is your sign to bring technology into the conversation.

For each one, write down who is doing that work now and what happens to them if AI does part of it. Bring operations and HR into the conversation.

For each one, write down who in your organization owns the outcome if the model is wrong. Legal, compliance, and the accountable business owner all need to weigh in.

If you cannot finish this exercise in 30 minutes, you are not ready to buy AI. You are ready to have a different conversation with your leadership team, and technology needs to be in that room from the start.

One last thought

AI has a place in your business. That place is specific, bounded, and earned. It is not revealed by a keynote. It is not handed to you by a vendor. It is discovered by the people who understand the work, the data, and the customer, sitting together and being honest about what is working and what is not.

The CEOs getting value from AI right now are the ones who brought their technology leadership into the strategy conversation on day one. The ones still chasing the conference high are mostly producing expensive pilots that do not reach production.

The difference is not the strategy. The difference is who is in the room when the strategy gets built.

Written by Jon McAnnis, Principal Advisor at Groundwork Technology Advisors.