The research on change is more interesting — and more hopeful — than the headline.

If the 70% figure has no original source, why do so many serious people still cite it?

That is the right question. And the honest answer turns out to be more interesting than I expected.

When you actually go back to the consultancy reports the modern claim is built on — the original PDFs, not the slides that quote them — something strange happens. The headline numbers do not match the underlying data.

Let me give you two examples, because once is an accident, but twice starts to look like a pattern.

The 2006 case. The most-cited modern source for the 70% figure traces to a 2009 McKinsey article by Scott Keller and Carolyn Aiken, The Inconvenient Truth About Change Management. They cited a 2006 McKinsey global survey of 1,536 executives as their supporting evidence. Professor Mark Hughes went and pulled the original 2006 paper. Its relevant section is sub-headed “Success is the norm”. The reported numbers: 38% of executives said the transformation they knew best was “completely” or “mostly” successful at improving performance, around a third said “somewhat” successful, and only about one in ten said “completely” or “mostly” unsuccessful. Keller and Aiken’s piece cherry-picked the lower of two success measures, dropped the survey’s own framing, and used it to support a “most change fails” narrative. The data said the opposite of the headline it was being used to support.

The 2008 case. Independently, the academic Dr David Wilkinson found a second McKinsey article from the same year — The Irrational Side of Change Management by Atkin and Keller — which cited “a McKinsey survey of 3,199 executives” showing “only one transformation in three succeeds.” Wilkinson tracked down the underlying survey and recomputed it. Of the 2,695 respondents who gave a substantive answer, 2,470 said their transformation was “somewhat,” “very,” or “extremely” successful. 145 said it was “not successful at all.” That is a reported failure rate of 5.87% in the underlying data. Atkin and Keller turned it into “only one in three succeeds.”

Two separate McKinsey publications, by different author pairs, in the same year, both reframing their own evidence in the same direction.

These papers, it should be said, contain a great deal of valuable material on change — cognitive biases, organisational dynamics, transformation methodology. I may write about that content in a later piece. This article has a narrower purpose: to understand how a single figure, the 70%, has continued to be quoted long after its empirical foundations were shown to be shaky. Selective citation is not lying. It is what happens when the headline you need is louder than the data you have, and it is something all of us have to watch for in our own work.

So what does the data actually say, when you stop forcing it into a binary?

Keith Driver‘s 2025 meta-analysis pulled together more than two decades of global transformation studies — McKinsey, Bain, BCG, IBM, Prosci — covering around 8,000 organisations and somewhere between one and two trillion dollars of cumulative transformation spending. The distribution that emerges across all of it is remarkably stable: roughly 30% of transformations succeed fully against their original ambition, around 50% deliver partial or fragile value, and around 20% collapse outright.

In Driver’s words: the myth says catastrophe dominates; the data show that mediocrity dominates. Most initiatives deliver something. Just not what was promised.

That shifts the whole conversation.

A binary fail-or-succeed framing erases the half of the distribution where most of the work and most of the value actually lives. It tells you nothing about whether the partial value was worth what it cost, whether the organisation came out more capable than it went in, or whether the work laid foundations for something later. It just collapses everything into one bit of information: pass or fail.

And it does so against goals that, in any genuinely complex change, were always going to need revising as the work taught you what to do.

So if the question is not “what percentage of change programmes fail,” what is it?

Three better questions surface from the more honest reading of the evidence:

What did this initiative actually create? Not against the original plan, but against what was realistic to achieve given what we now know.

What did the organisation learn? What capabilities did it build, what assumptions did it expose, what is now possible that was not possible before?

Was it worth what we spent to get it? Not just in money, but in attention, in goodwill, in the things we did not do because we were doing this.

Those are harder questions than “did it succeed?” They are also much more useful.

There is much more in the full article when it publishes — including McKinsey’s own 2021 finding that even successful transformations capture only about two-thirds of the value they could have, the reasons Hughes gives for why any single failure rate is conceptually wrong-headed in the first place, and deborah rowland ‘s research suggesting that more than half the variance in change outcomes comes from a factor most leadership programmes barely address. Each of those threads, I think, is more useful than the statistic this series started with.

How to lead change in a way that takes these questions seriously — and what the research actually does support about the conditions for that kind of leadership — is the subject of article 3.