Deming and his followers use a figure of 95% in at least two different contexts.
- 95% of problems are system-related (whatever that means)
- 95% of the so-called improvement initiatives are futile because people don’t know the profound knowledge or competency
I have found two sources that credit the second observation to Peter Scholtes.
- Nomination for the Deming medal (Peter Scholtes website)
- Pareto Distribution: Deming, Juran and Simon’s Rule and Worlds of Knowledge (Hanching Seminars and Consulting, 1998/11)
I tend to read this kind of claim as more rhetorical than scientific. Not only is it hard to find any empirical study that might support this kind of claim, it is difficult to see what kind of evidence might be adduced.
As I pointed out in my paper Reasoning about Systems and their Properties, to decide that one intervention is successful and another is futile is an act of interpretation, and assumes we know which system we are talking about, from whose perspective, with what timescale, and so on. In order to have any credible basis for dismissing an intervention as "futile" or "meddling" or "tampering", you actually have to do a full systems analysis on the case, and the results could still be disputed. You can't jump to conclusions - "Oh, this didn't work, so the people who attempted it obviously didn't know what they were doing" - because success and failure and their causes are extremely complex, and well-chosen interventions based on a deep and thorough systems analysis may also sometimes fail.
If this 95% were both meaningful and true, what would be the consequences for action, and what would be a reasonable target for improvement? Writing books and articles complaining about a general lack of management profundity, or the folly of "The Regime", looks suspiciously like an ineffectual meddle rather than a well-chosen intervention with a well-designed outcome. Will Deming's followers take their own medicine? See my post Easier Seddon Done.
See also Paul Hebert on Deming and Systems in Today's Business World - No Answers Just Questions, pondering whether the 95% still applies to knowledge/software or whether a 50%/50% split would fit better (via @baob)