The 70 percent figure nobody can find

Where the number came from

In 2011 the management researcher Mark Hughes traced the 70 percent claim back through the literature. His conclusion was blunt. While a popular narrative of 70 percent failure clearly exists, "there is no valid and reliable empirical evidence to support such a narrative" (Hughes, Journal of Change Management, 2011).

He went back to it a decade later. He reported that he had searched, without success, for the evidence behind the figure. One of the best known sources, a Harvard Business Review piece by Beer and Nohria, offered neither empirical evidence nor references to literature for the claim (Hughes, Journal of Change Management, 2022).

It is not the only part of the change canon that is shakier than it looks. Kotter's eight steps, probably the most used change model in business, did not reference any outside sources when first published (Appelbaum and colleagues, Journal of Management Development, 2012). And the "unfreeze, change, refreeze" model credited to Kurt Lewin was, on close reading, never developed by Lewin in the form it is taught (Cummings, Bridgman and Brown, Human Relations, 2016).

None of this makes those ideas useless. Kotter's steps are a reasonable checklist. But a checklist is not a law, and a slogan is not a statistic.

So what is the real failure rate?

Honest answer: it depends on what you mean by failure, and when you measure it.

A 2026 review in the Journal of Engineering Management and Competitiveness found that where failure rates are grounded in actual studies, they range from roughly 30 percent to more than 65 percent, depending on how failure is defined and over what time horizon (de Waal, JEMC, 2026). That is a wide range. It is also a more useful one, because it tells you the definition is doing most of the work.

A system that went live on time but nobody uses. A restructure that hit its cost target and lost the three people who knew how anything worked. Are those successes? It depends who you ask. That is the point.

Why it matters more than it seems

A made-up number is not harmless. It does three quiet things.

It frames change as a gamble, which makes leaders reach for control and speed. It gives everyone a ready excuse when things go wrong ("well, most of them fail"). And it pulls attention away from the questions that actually predict outcomes: what is changing in people's jobs, who gets a say, and how you will know it is working.

Management scholars have been asking for exactly this shift. Stouten, Rousseau and De Cremer reviewed the popular change models alongside the research and argued that practitioners should integrate the two, keeping what the evidence supports and dropping what it does not (Academy of Management Annals, 2018).

What we suggest instead

1. Define success before you start.

   Write down what "worked" will look like in 90 days, 12 months and two years. Adoption, not launch. A leadership team that cannot agree on this has found its first real piece of work.

2. Cite a range, and say where it came from.

   If you need a benchmark for a board paper, use a sourced range and the definition behind it. "Studies put this between a third and two thirds, depending on how failure is measured" is less dramatic and much harder to argue with.

3. Measure the people side as seriously as the plan.

   Track what changes in workload, control and support for the people doing the work. Those are the things that tend to decide whether the plan survives contact with Tuesday.

4. Treat models as tools, not truth.

   Use Kotter, Lewin or anyone else where they help. Drop them where they don't fit your situation. Nobody gets marks for loyalty to a framework.

5. Ask for the footnote.

   When a consultant, vendor or keynote speaker gives you a striking statistic, ask where it came from. That includes us.