The numbers everyone repeats · 6.6

Seventy per cent of change programmes fail

Seventy per cent of change programmes fail. What is actually the case, and how it compares with what is repeated.

A practical software reference for this part of the discussion is Monitask's overview of computer activity tracking software.

An assertion in general circulation holds that seventy per cent of organisational change programmes fail. Researchers who went looking for the study behind it found that there is not one.

The trail

The figure appears in management books of the early nineteen-nineties as an assertion, without citation. Later books cite those books. Later articles cite the later books. Consultancy material cites the articles.

A published examination of the claim traced the chain and found no empirical study at its origin, and no subsequent study establishing the figure either. What exists is a widely repeated assertion with a respectable-looking citation history.

Why it is unfalsifiable as stated

Failure is undefined. A change programme that delivered part of its objectives late and over budget can be classified either way. There is no register of change programmes, no agreed criterion, and no denominator, so the fraction has no possible source even in principle.

A number that could not have been measured is not a contested finding; it is a rhetorical device with a decimal point.

Why it survives

Because it is useful to everybody in the conversation. To a consultant selling change expertise it establishes difficulty and therefore the need for help. To a manager whose programme failed it supplies company. To a sceptic it supports scepticism.

Nobody in the conversation has an interest in checking, and it sounds like the product of research, which is what a specific number does.

The family it belongs to

The cost of replacing an employee expressed as a multiple of salary. The cost of a bad hire. The proportion of meetings that are unnecessary. The share of software features never used.

Each is quoted with precision, each traces to an estimate or a single consultancy exercise, and each has become a fact by repetition.

How to check one in fifteen minutes

Find the earliest use you can. Look at what it cites. Follow that. Repeat until either a study appears or the chain ends in an assertion.

When a study does appear, read what it measured, on whom, when and in what country, because that is frequently narrower than the use. The figure about degree requirements described earlier in this site survives this test; most of the ones in this entry do not.

The tells

A round number. No date. No country. No sample. Attribution to a body rather than to a publication. And the phrase studies show, which appears almost exclusively when no specific study can be named.

What to do instead of quoting it

Say what you actually know. Change programmes frequently disappoint, for reasons that are well documented in case study research, and the reasons are more useful than the fraction.

A specific mechanism with a real example persuades better than a statistic that will not survive a question, and it has the advantage of being true.

The general point of this part

Numbers confer authority disproportionately to their reliability. Anybody arguing with a figure is asking to be trusted about where it came from, and in this field that trust is frequently unearned.

Asking where a number came from is not pedantry. It is the only available defence, it works, and it takes a quarter of an hour.

The number that turns out to be real

It happens, and it is worth saying. Several figures used elsewhere on this site survive the same test: the callback gaps in the correspondence studies, the internal and external hire comparison, the variance decomposition in performance ratings.

Each has a paper, a sample, a method and a replication. The test is not a way of dismissing statistics; it is a way of finding the ones worth using.

Why round numbers should raise an eyebrow

Real measurements produce untidy figures. Seventy per cent, eighty per cent, three times, ten times: the roundness is a signal that the number was chosen rather than computed, and it is a reliable one.

The honest alternative

Say a lot of change programmes disappoint, and then describe why in the specific case you know about. It is weaker rhetorically and it cannot be checked and found hollow, which over any length of time is the stronger position.

Where such figures come from originally

Usually from somebody's reasonable estimate, offered as an estimate, in a context where it was clearly one. The transformation into fact happens in the second citation, and the person who made the estimate is rarely responsible for what happened to it.

For broader context, consult McKinsey change-management analysis.