W. Edwards Deming, the American statistician who taught Japan how to build cars and appliances after World War II and then spent the last decades of his life trying to teach the same lessons back to American industry, wrote a single sentence in 1982 that has been quietly explaining organizational failure for the four decades since.
The sentence appears on page 18 of Out of the Crisis. In compressed form:
“Best efforts are essential. Unfortunately, best efforts, people charging this way and that way without guidance of principles, can do a lot of damage.”
Read that twice. The first sentence is what everyone expects Deming to say. The second sentence is what almost no one is prepared to hear.
Best efforts, applied without theory, do damage. Not neutral. Not merely suboptimal. Damage.
The specific mechanism
Deming’s framing rests on a specific observation from his decades of consulting to manufacturers. When workers on a production line saw a defect and, without any theory of what was causing the defect, adjusted the machine to compensate — they made the defect worse. When managers saw a bad quarter and, without any theory of what was causing the bad quarter, changed the strategy — they made subsequent quarters worse.
The intuition was: something is wrong, we must do something. The intervention was: change whatever seems adjustable.
The result was: worse outcomes than if the workers and managers had done nothing at all.
Deming named this pattern tampering. The tampering was well-intentioned. The tamperers were working hard. Their best efforts were, in the specific technical sense, best. The problem was that the efforts were unguided by any theory of what was actually producing the observed variation. So the interventions were amplifying the variation rather than reducing it.
The 2026 application
Almost every enterprise AI transformation I have visibility into in 2026 is a tampering operation at scale.
The pattern: senior leadership observes that their organization is falling behind on AI. Something is wrong. They must do something. They authorize investment. Vendor selections are made without a coherent theory of what the AI investments are supposed to produce. Pilots are launched without a theory of what would constitute success. Change management programs are initiated without a theory of what specific change is being managed. Consultants are hired without a theory of what specific gap the consultants are supposed to close.
Everyone is working hard. Best efforts are being applied everywhere. The tampering, in Deming’s specific technical sense, is severe. The outcomes are worse than if the organization had done less, more slowly, guided by a coherent theory.
What theory looks like
A theory of AI transformation is not a strategy document. Strategy documents are what tamperers produce to justify the tampering. A theory is a specific, testable claim about the mechanism by which AI investments are supposed to produce value in this specific organization.
A theory answers questions like: what specific human workflow is going to be transformed, and by what mechanism? What is the specific bottleneck that AI capability removes, and why is that bottleneck currently binding? What is the specific unit of value creation that will change, and by how much?
An organization that can answer these questions has a theory. It can act on the theory. If the theory is wrong, it will discover the error quickly and update. If the theory is right, it will compound value over time.
An organization that cannot answer these questions has no theory. It is tampering. Its best efforts will produce damage.
Three practical questions
One: what is your organization’s theory of AI value creation? Not the strategy document. The specific, testable claim about the mechanism.
Two: which of your current AI activities is guided by that theory, and which is tampering? Every activity that cannot trace back to the theory is, by Deming’s definition, tampering. Stop it.
Three: what would falsify the theory? If nothing would falsify it, it is not a theory. It is a slogan. Slogans do not guide effort. They only justify it after the fact.
The closing thought
Deming’s warning about best efforts is uncomfortable because it undermines the standard executive response to falling behind. The standard response is to work harder, invest more, and demand more effort from the organization. Deming says: without theory, this response makes things worse.
The specific damage done by tamperers is invisible to them because it looks like activity. Meetings are held. Vendors are selected. Pilots are launched. The organization looks like it is transforming. What it is actually doing is amplifying the variation, degrading the underlying processes, and consuming resources that could have been deployed productively if guided by a coherent theory.
In 2026, most enterprise AI transformations are best-efforts operations without theory. Most will do damage. The small number that develop and act on a coherent theory of AI value creation will produce durable transformation. The distinction will be visible in three years.
The world has changed. The leaders who notice will be the ones the next decade is built around.
