An open letter to the CEO who is not buying this yet:
You have been told for three years that AI will change everything, and in that time you have sat through the vendor demos, the board questions, and the pilot that produced a slide deck and little else.
If your instinct is to wait until the dust settles, you have reasons, and we are going to start by agreeing with most of them.
If your instinct is to wait until the dust settles, you have reasons, and we are going to start by agreeing with most of them. Then we are going to ask you to look at one thing the evidence does not say, and one thing about timing that we think is different now.
A disclosure first. We are a consultancy that sells advice on this subject, so we have a stake in your answer. Weigh what follows accordingly, and check our sources; we name them below, so you can do just that.
You Are Right About More Than You Think
Your skepticism about results is supported by the numbers. PwC’s 2026 Global CEO Survey, covering 4,454 CEOs in 95 countries and territories, fielded in the fall of 2025, found that 12 percent said AI had improved both their costs and their revenue, 33 percent saw a gain in one of the two, and 56 percent reported no significant financial benefit so far. Deloitte’s State of AI in the Enterprise, which surveyed 3,235 senior leaders in 24 countries in August and September 2025, found that 66 percent reported productivity or efficiency gains, but only 20 percent reported higher revenue, even though 74 percent hoped for it.
Your skepticism about hype is supported too. Gartner predicted in June 2025 that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing cost, unclear value, and weak risk controls. It also said many vendors are rebranding existing chatbots and automation tools as agents, and estimated that only about 130 of the thousands of vendors claiming agentic capability are genuine. That is a forecast, not a measurement; the poll behind it was 3,412 webinar attendees, and Gartner does not show how it reached the 40 percent. But it describes a market in which a careful buyer should be cautious, and you are being careful.
What the Failure Evidence Does Not Say
Here is the part we would ask you to read twice. Those surveys tell you what executives report today, across companies that went about it in very different ways. They do not separate a well-designed attempt from a shallow one. Deloitte, for example, found that 37 percent of organizations are using AI at a surface level, while 34 percent describe themselves as deeply transforming their business, and that only one in five has a mature model for governing autonomous agents. If a third of the sample is barely trying and four out of five lack a mature model for governing autonomous agents, a weak average return is what you would expect, and it tells you little about what a disciplined attempt would earn.
That is our inference, not a finding. Neither survey tests it, and we know of no published study that compares careful adopters with shallow ones at your size.
We raise it because the usual conclusion, that the returns are not there, rests on the average, and the average mixes serious efforts with unserious ones.
We raise it because the usual conclusion, that the returns are not there, rests on the average, and the average mixes serious efforts with unserious ones.
The Risk Is Real and It Can Be Bounded
If your concern is risk, you are describing a real thing. AI tools that act inside your systems can leak data, make commitments you did not authorize, and fail in ways nobody notices for weeks. Our answer is not that the risk is small. It is that you do not have to take it all at once. A first test can be read-only. It can touch one process, with a named owner, a limit on what it is allowed to see and do, and a written rule for when a person has to step in. We have written about how to build those lanes in the Governance Freeway, and about why employees are already using AI on their own in The Best of Both Worlds. If your people are already using these tools without your knowledge, as we expect many are, then your choice is not between risk and no risk. It is between risk you can see and risk you cannot.
The Case for Waiting, Stated Fairly
The best argument for waiting is that you let others pay for the mistakes. Tools are changing every few months, prices are falling, and anything you build this year may be obsolete next year. For some companies in some industries, waiting a while longer is a sound call, particularly where regulation is heavy and the tools are still thin. We said as much in our piece on readiness, where we argued that deferring in 2023 and 2024 was reasonable.
So why do we say now? Our reason is about asymmetry, and it is a judgment, not a measurement. We looked for a trustworthy public estimate of what waiting costs a mid-market company and did not find one; what we found was mostly vendor marketing, which we have not used. So we will argue from the shape of the decision instead.
Why We Say Now
Picture a company like yours with a competitor in the same market. Over two years, the competitor runs a series of small, bounded tests, keeps the two that work, and quietly lowers its cost to serve. Nothing dramatic shows up in a quarter. In the language of our four shapes of the AI dividend, this is a slow melt, and the first sign is a price you cannot match or a customer who is no longer quite as loyal as you expected.
The asymmetry is this. A small test costs little and can be stopped. A lost position costs a great deal and cannot easily be recovered. The skeptic’s real advantage is that waiting is reversible, and we think that advantage is wearing down, because the learning that competitors gain by practicing does not wait for you. Gartner’s CEO survey, reported in April 2026 and covering 469 respondents with fieldwork ending in the fourth quarter of 2025, found that 80 percent of CEOs expect AI to force a high or medium degree of change to their operational capabilities, and that 54 percent said automation was limited to specific tasks today, while only 13 percent expected to still be there by the end of 2028. Those are expectations, not results. But if most of your peers expect to change and you do not, one of you has misread the market.
Our broader view is that AI is changing faster and more powerfully than any technology in history, as a commoditization of cognition, and that its potential and its risks are greater than those of earlier economic revolutions. That is our view and not a proven comparison. We mention it because it is the reason we do not think “later” is the safe default it used to be.
What This Will Not Do
This letter will not prove that AI will pay off at your company. It may not. It will not replace a test on your own data. And we are not asking you to believe the headlines, only to notice that the evidence for skepticism and the evidence for urgency can both be true at once: average results are poor, and a careful, bounded effort may still be your best defense. If you try a disciplined test and the numbers do not hold, stop, write down what you learned, and look again in a year. That is a fine outcome.
Just a Conversation
We are not asking you to buy anything, sign anything, or commit to a plan. We are asking for a conversation. Tell us what you are seeing in your market, what has disappointed you so far, and what you would need to see before you move. We will tell you what we think, and if the honest answer is that your company should wait, we will say so. If you would like to start, you can reach us through our contact page.
We will tell you what we think, and if the honest answer is that your company should wait, we will say so.
