In 1993, one year before his death at 93, W. Edwards Deming published The New Economics for Industry, Government, Education. It was his final and most philosophically ambitious book.
Buried in the middle is a sentence that, in my view, is the most important thing Deming ever wrote about how organizations actually learn.
“Experience teaches nothing. In fact there is no experience to record without theory. Theory leads to prediction. Without prediction, experience and examples teach nothing.”
Read that carefully. Experience teaches nothing. Examples teach nothing. Case studies teach nothing. Best practices teach nothing.
Only theory teaches. And theory teaches only by producing predictions that can be tested against experience — and updated when the predictions fail.
Why case studies don’t work
The dominant mode of executive learning in most organizations is case-study consumption. McKinsey publishes a case study. HBR publishes a case study. A vendor presents a case study. A conference features a case study. The executive team reviews the case study and attempts to derive lessons for their organization.
Deming’s argument is that this mode of learning does not, in fact, produce learning. Not because case studies are wrong. Because the executives consuming them have no underlying theory against which to test the case study’s claims.
Without a theory, the case study is a story. Stories are memorable. They are not predictive. They do not tell you what will happen in your situation. They tell you what happened in someone else’s situation, with all the specific conditions that made that outcome possible, most of which are invisible in the case study itself.
The 2026 application
Every senior leader I work with in 2026 is consuming AI transformation case studies at unprecedented rates. Vendor case studies. Consulting firm case studies. Conference case studies. Peer-company case studies. LinkedIn case studies. Podcast case studies.
The case studies are describing real outcomes at real companies. The consumption of them is producing almost no learning inside the consuming organizations.
The reason is the Deming reason. The consuming executives have no theory of AI value creation in their specific context against which to test the case studies. They cannot ask, based on the case study, what would happen in their own organization. They can only observe that something happened somewhere else.
The case study is a story. The theory is what would turn the story into learning. Without the theory, the case study is entertainment.
What a theory produces
A theory of AI value creation produces predictions. Specifically, it produces predictions about what will happen when specific AI capabilities are applied to specific workflows in your specific organization.
If the theory is right, the predictions come true. The organization compounds learning.
If the theory is wrong, the predictions fail. The organization discovers the error, updates the theory, and moves forward with a more accurate understanding.
Either way, the organization learns. The learning comes from the theory, not from the case studies. The case studies inform the theory only to the extent that the theory is precise enough for the case studies to be relevant evidence.
Three practical questions
One: what specific predictions does your organization’s AI theory produce? Not aspirations. Predictions. If your AI initiative succeeds, what will be measurably different in six months? If the specific measurable difference cannot be named, you have no theory.
Two: which of your recent AI decisions have been informed by case studies rather than by theory? Case studies without theory are stories. Stories produce feelings, not decisions. Which of your recent AI decisions were driven by stories?
Three: when your theory’s predictions fail, what will you update? If the answer is “the strategy,” you have no theory. Strategies are outputs of theories. If the answer is “the theory itself,” you are practicing what Deming was pointing at.
The closing thought
The reason Deming’s framing is so uncomfortable is that it implies most organizational learning is not learning. Most case-study consumption is not learning. Most best-practice benchmarking is not learning. Most consulting engagements do not produce learning.
What produces learning is the disciplined construction of theory, the derivation of predictions from the theory, the testing of predictions against reality, and the updating of theory when predictions fail.
This is expensive. It requires intellectual work that consulting deliverables do not do for you. It requires patience that quarterly cycles do not accommodate. It requires humility that most executive incentive structures do not reward.
It is also the only thing that produces durable value in a moment of technological discontinuity. In 2026, the organizations that operate on theory will compound learning. The organizations that consume case studies will accumulate stories that produce no lasting improvement.
The world has changed. The leaders who notice will be the ones the next decade is built around.
