A VP of strategy at a mid-size insurance company walked me through her company's AI roadmap last winter. Forty slides. A five-tier maturity model, a center of excellence, a governance council, three workstreams with owners listed by title, and a color-coded timeline that ran to the end of the following year. It was a polished document. I asked her one question. What would be different on the floor the Monday after this deck existed? She went still for a moment, and then she gave me the honest answer. Nothing yet. The deck was for the board.

I have seen versions of that deck in a dozen rooms since, and the survey data suggests it is closer to the rule than the exception. WRITER, working with the research firm Workplace Intelligence, surveyed 1,200 executives and 1,200 employees this year, and three-quarters of the executives admitted their company's AI strategy is "more for show" than actual internal guidance.1 Almost six in ten of those companies are spending more than a million dollars a year on it.1 Nearly half of the leaders called their own adoption a massive disappointment, up from a third the year before.1 Thirty-nine percent had no formal plan to turn any of it into revenue.1 Hold those numbers next to each other and a strange picture forms. A great deal of money is moving, a great many documents are being produced, and the people producing them will tell you, if you ask plainly, that the documents are not meant to direct the work.

The deck is doing a job it does not admit to

A strategy is supposed to be a set of instructions for choosing under uncertainty. It tells people what to do when two reasonable options compete for the same week. The insurance roadmap did none of that. It described a destination and an org chart, and it left every actual decision exactly where it had been before. The document was very good at the job it was actually built for, which was to demonstrate that leadership was awake to the most watched technology shift in a generation.

The tell is the missing Monday. When you ask what changes for a specific team on a specific day, and the honest answer is nothing, the thing in front of you was never an instruction. It was a signal. Signals are aimed at an audience, and the audience here is almost never the workforce.

Who the signal is for

It is the board. It is the analyst on the next earnings call. It is the peer at the industry dinner who already claims to have agents running in production. WRITER found that 73 percent of CEOs report stress or anxiety about their AI strategy, and 64 percent fear they could lose their job if they fail to lead the transition.1 That fear is the engine. A leader who believes the board is grading them on visible AI fluency will manufacture visible AI fluency, and a forty-slide maturity model is an efficient way to manufacture it. The roadmap ends up protecting the person who commissioned it well before it does anything for the company that paid for it.

I want to be careful about where that figure comes from. WRITER sells an enterprise AI platform, and the report's prescription is, predictably, to buy one. That interest is real and worth saying out loud. The descriptive numbers still hold, and they match what leaders tell me in private once the deck is closed and the room is smaller.

A strategy written to be seen will always read better than a strategy written to be used, because the audience for the first one is sitting in the room, and the audience for the second one is the work itself.

What a performed strategy costs downstream

The bill comes due below the executive who never has to read it. MIT's NANDA initiative looked at the wider field and found that ninety-five percent of organizations were seeing no measurable return on their generative AI spending, against thirty to forty billion dollars committed.2 They called the cause the learning gap, the failure to fold the tools into the real workflows, structures, and habits of the place.2 A document built to be looked at cannot close that gap, because closing it means someone has to change how a particular job gets done, and the document was careful never to name a particular job. So the mandate reaches the floor with its meaning drained out. Teams are told to adopt AI, handed no definition of done, and left to invent priorities that shift again at the next quarterly refresh. I spent years as an agile coach watching strategies dissolve in exactly this stretch of road, the space between the boardroom and the desk. AI has not redrawn the geography. It has raised the stakes and the budget.

The unglamorous alternative

The fix is smaller than the problem, which is part of why it keeps getting skipped. Name one real thing. Pick a single workflow that somebody in the building does slowly or badly, put a named owner on it, give that owner the authority to change how the work is done, and measure whether the work got better. One claims adjuster's queue. One underwriting handoff. The version of that VP's roadmap that would have moved something on Monday could have fit on a single line: one team, one owner, permission to act, and a number to watch.

That move is harder than the deck, and not because it is technically difficult. It is harder because it is legible. A named owner working on one real workflow can fail where everyone can see it, and a maturity model never can. The leader who chooses the smaller, riskier version is agreeing to be judged on whether the work actually improved instead of on whether the strategy looked current. Most of the performance we are talking about is an attempt to avoid exactly that judgment.

So here is the question worth sitting with. If you closed the deck and walked out onto your own floor, could the people there tell you what they are supposed to do differently this week, without looking at it? If the answer is no, the document is doing some other job than guiding them, and it is worth asking honestly who you wrote it for.