Kaizen is improvement by accumulation. Many small changes, made continuously, by the people closest to the work — as against occasional large changes designed by someone else.

The arithmetic is the familiar one: a 1% improvement each week is roughly a 68% improvement over a year, if the improvements hold. Compounding does the work, and the phrase "if the improvements hold" is carrying most of the difficulty.

Why small and continuous beats large and occasional

The people doing the work know where the problem is. Not as a matter of respect — as a matter of information. The person who runs a step forty times a week has seen the failure modes. Somebody analyzing it for a fortnight has seen a sample.

Small changes can be reversed. A change affecting one step can be tried and undone on Thursday. A redesign of the whole process cannot, which means it has to be argued about in advance, at length, by people who are guessing.

The feedback is immediate. Change one thing and the effect is attributable. Change fifteen and you have learned something about the bundle and nothing about any element of it — which is why large improvement projects so rarely produce a repeatable lesson.

It does not require a budget. The most underrated property. Anything requiring approval competes with everything else requiring approval, and most small improvements would lose that competition despite being worth more in aggregate than the thing that wins.

The condition that makes it work

One, and without it the rest is ceremony: the person doing the work must be able to change the work without asking.

Every serious kaizen implementation rests on that, and every failed one has replaced it with a suggestion scheme. A suggestion scheme routes the change through an approval, which reintroduces the delay, the competition for attention and the distance from the problem — and produces a system in which improvements are proposed by the people who know and decided by the people who do not.

There has to be a boundary. Changes within a step, reversible, not affecting other people's work: no permission. Changes crossing a boundary or hard to reverse: discussed. Drawing that line explicitly is most of the implementation.

Where it does not reach

Kaizen improves a process. It does not replace one, and it will not tell you the process should not exist.

A business incrementally improving a service nobody wants gets very good at producing it. The small-change mechanism is structurally incapable of producing the large discontinuous change, because every step of the path to a different approach is worse than the current one — which is path dependence operating on improvement itself.

So it pairs with something else: a periodic willingness to ask whether the thing being improved is the right thing. Rare, deliberate, and a different activity from the weekly one.

It also runs into diminishing returns. The tenth improvement to a step is worth less than the first, and a process that has been improved continuously for three years is mostly done. The signal is that the changes stop being obvious to the people doing them, and at that point the attention is worth moving to a step that has not had any.