Where the 70% transformation failure statistic actually came from

You have seen it this month. It opens the second slide of every change deck: seventy per cent of transformations fail. It gets nods around the table, it justifies the budget ask, and nobody asks where it came from.

I went and looked. The answer is worth knowing, because at some point you will be in a room with someone who has also looked.

1993: an estimate the authors called unscientific

The trail starts with Michael Hammer and James Champy’s Reengineering the Corporation. They wrote that “somewhere between 50 and 70 percent” of reengineering efforts did not achieve the results intended.

Two things about that sentence. It is a range, not a figure, and it is about business process reengineering, which is not digital transformation and is certainly not AI adoption. The book itself describes the estimate as unscientific. By 1995 Hammer had rejected the idea that it represented a fixed failure rate at all.

So the number was never a finding. It was two authors giving an impression of what they had seen, and saying so.

2011: somebody checked

Mark Hughes published a paper in the Journal of Change Management asking whether 70 per cent of organisational change initiatives really fail. It is peer reviewed, which already puts it in a different category from most of what gets quoted on this subject.

He traced the claim across five published sources. He found no empirical basis for any of them. He came back to the same conclusion in 2022.

That paper has been available for fifteen years. The statistic has kept spreading anyway, which tells you something about how business evidence travels.

2015 onwards: the consultancies

McKinsey, 2015. States that seventy per cent of change programmes fail to achieve their goals. No source given, no method, no sample size. It is also about organisational change in general, not digital transformation, though it now gets quoted as though it were about both.

McKinsey, 2018. A digital transformation survey finds that 16 per cent of respondents said their transformation both improved performance and sustained the improvement. This is a genuine measurement, and it is frequently conflated with the 2015 assertion even though the two are measuring different things.

Harvard Business Review, 2019. Cites seventy per cent missing their goals, and 900 billion dollars wasted out of 1.3 trillion spent. Follow that one back and you land on a 2018 Forbes column by the chief technology officer of a software company, who applied the seventy per cent figure to global spending projections. Not a study. A column, doing arithmetic on a number that was already unsourced.

BCG, 2020. Reports that seventy per cent of transformations fall short of their objectives. This one is the most interesting, because BCG publishes the breakdown: 30 per cent fully succeeded, 44 per cent created some value while missing their targets, and 26 per cent created little value.

Read that again. A “fall short” band that includes projects which created real value but missed a target is not a failure rate. If you ran a hiring process where 30 per cent of offers were accepted outright and 44 per cent led to a candidate joining on adjusted terms, you would not describe that as a 74 per cent failure rate.

Why any of this matters to a People team

Three reasons, and the third is the one that costs money.

Numbers you cannot trace make you fragile. If your case for a programme rests on a statistic, and someone in the room has read the BCG breakdown, you lose the room on a detail that has nothing to do with your actual argument.

Vague failure rates make people fatalistic in a way that is quietly damaging. If seven in ten of these things fail regardless, the rational move is to keep your head down and let someone else volunteer. That is the opposite of the behaviour a transformation needs from a line manager.

And the number does no work. It tells you that transformations are hard, which everybody already believes. It tells you nothing about which ones fail, at what stage, or what the people who succeeded did differently. A statistic that changes no decision is decoration.

What to say instead

If you need a figure, use McKinsey’s 2018 finding: 16 per cent both improved performance and sustained it. Say it is self-reported, say it is from a survey, and check the sample size against the original before you use it. A number offered with its limits attached is more persuasive than a big round one offered without.

If you do not need a figure, do not use one. “Most of these programmes disappoint the people who paid for them, and here is what we are doing differently” is a stronger opening than any percentage, because it invites the next question instead of closing it down.

And if someone else quotes the seventy per cent at you, be gentle about it. The point is not to win a fact-check. The point is that the way you handle evidence in a slide is the way you will handle evidence in a benefits case, and everyone in the room is quietly making that inference about you already.