In naval intelligence, there is a specific failure mode that gets its own name in the tradecraft — not fabrication, something subtler. The reporting chain begins to optimize for what the decision-maker wants to hear. Data arrives shaped. Friction gets stripped out at each level, not because anyone lied, but because the system has learned what kinds of information survive the climb upward. By the time the brief reaches the commanding officer, it looks like progress. The decision made from it lands wrong. The unit pays for it in ways the briefing could not have predicted.

That pattern has a direct organizational equivalent, and it is currently running through AI adoption in enterprises large enough to have a board conversation and a floor conversation that never meet.

The Number That Deserves a Slower Read

An April 2026 piece in Harvard Business Review — "Managers and Executives Disagree on AI — and It's Costing Companies" — documented something that most executives will recognize when they sit with it long enough. Executives report strategic momentum. Managers are watching tools sit unused, workflows fragment, and teams quietly build workarounds to avoid the platforms they were told to adopt. The two groups are not describing the same organization.

The broader data makes the gap harder to dismiss with confidence. Ninety-five percent of organizations investing over a million dollars annually in generative AI are seeing zero return on that investment. Last year, 42% of companies abandoned most of their AI initiatives. The average organization scrapped 46% of its AI proof-of-concept projects before reaching production. These organizations are not technically incompetent. Most of them have capable engineers, executive sponsorship, and real budget behind the work.

What they built — without intending to — is an organizational architecture that reports up one reality while operating another.

The Diagnosis Most Leaders Reach For

When adoption stalls, the default explanation is resistance. People do not like change. Training is insufficient. Incentive structures need adjustment. More pressure from the top is the solution.

That diagnosis is not entirely wrong, but it is convenient in a specific way: it locates the failure below the executive level. The strategy was sound. The execution failed. The people on the floor did not follow through.

The harder read — the one most executive teams will not do unprompted — is that the AI strategy was built from boardroom conversations, analyst briefings, vendor demonstrations, and media coverage. None of those sources carry honest signal from the people who actually run workflows. None of them are designed to surface what does not work. By the time an executive signs off on an AI rollout, they have consumed an enormous volume of information that was curated for persuasion, not accuracy.

The floor report on any major initiative is almost never the report that reaches the boardroom. Not because people are dishonest — because the system has been trained to reward the news that survives the climb.

Middle managers, according to the HBR research, are significantly more likely than executives to flag readiness problems with AI. In some data sets, three times more likely. That signal is being generated somewhere in the organization. It is simply not reaching the people who need it — because raising it is career risk, and the system has taught people what career risk looks like.

What the Workaround Is Actually Telling You

There is a specific moment in almost every failed AI rollout where the floor has already solved the problem the executive team is still trying to mandate. The team has built a workaround. They found a way to complete the work without using the platform handed to them. They are doing it not out of obstinacy but because the platform does not fit how the work actually gets done.

That workaround is intelligence. It shows you exactly where the gap between the boardroom's model of work and the floor's actual experience of work is widest. If you knew where every workaround in your organization lived, you would have a better map of your AI adoption failure than any progress report has given you.

The workaround almost never makes it into the weekly briefing. The manager who built it does not raise it. The adoption metric continues to look acceptable. The executive never learns it exists.

What Changes First

The executives who have broken through this pattern do not start with a new AI platform or a revised change management framework. They start by changing what they use as information sources.

They go find the person who built the workaround. They sit next to someone using the tool and watch for an hour. They ask managers to bring the worst-performing use case to the next meeting, not the best one. They reward the person who tells them what is broken over the person who tells them things are on track. These are not complicated interventions. They are the kind of moves that require deciding, at the leadership level, that you actually want to know what is happening rather than what the reporting chain has learned to deliver.

The AI initiative that has stalled in your organization probably did not fail on the floor. It failed the moment the briefing that launched it was built entirely from sources that could not tell the truth about operational constraints. Senior leaders have significant leverage over what gets reported upward. The question worth sitting with is whether that leverage has been used to build a system that surfaces reality or one that has simply learned to deliver a story that moves fast through the chain.

AI is making that question more expensive to leave unanswered. If you are getting clean reports and zero ROI, the two facts are connected — and the connection starts somewhere above the floor.