Continuous Improvement vs Transformation: The Real Test

What does this article cover?

The honest version of this comparison turns on one question that is usually skipped: is the thing you want to change divisible. Continuous improvement works on anything you can do a little of. Step-change is required for anything you cannot, and there is no incremental version of closing a plant or exiting a business line. That distinction resolves most of the argument, and it leaves one genuinely open question about rate that I will put to you without pretending it is settled.

Watch the companion video on 3-A Continuous Improvement.

Watch on YouTube

What does the evidence actually say?

That increments do not accumulate. An empirical test of transformation found most happened through rapid discontinuous change across most domains at once, and that small changes in strategy, structure and power distribution did not add up to a fundamental transformation.

Romanelli and Tushman published that test in the Academy of Management Journal in 1994, examining event histories to see whether the punctuated equilibrium model held. Three findings came out of it. Most transformations happened through rapid, discontinuous change over most or all domains of organisational activity. Small changes did not accumulate into fundamental transformation. And major environmental change and chief executive succession both influenced when transformations occurred.

The model behind it, set out by Tushman and Romanelli in 1985, describes organisational life as long periods of incremental, adaptive change punctuated by brief periods of discontinuous reorientation. Note what that means for this comparison: the model does not say incremental change is wrong. It says incremental change is the normal state, and that it is not the mechanism by which organisations become different.

The replication work is worth knowing about. A later study examined event histories of 50 US bank holding companies between 1975 and 1995, testing the same assertion in a less turbulent industry than the minicomputer firms of the original. The model has been tested more than most things in this field, which is not a common thing to be able to say.

What does that finding not cover?

Operational performance. The study measured strategy, structure and power distribution, which are indivisible: there is no incremental version of closing a plant or exiting a business line. How well work gets done is divisible, and divisible things do respond to frequency.

This is the distinction that does the real work and it is missing from most versions of this argument. Sort what you are trying to change into two piles.

Divisible. Cycle time, defect rate, changeover duration, quote turnaround, on-time delivery, inventory accuracy. Every one of these can be five per cent better next month and five per cent better again the month after. Continuous improvement is not merely adequate here, it is superior, because the alternative is a large project that disrupts the work to achieve what steady pressure would have delivered anyway. For how much a trainable execution detail can decide at the top level, see how Michael Phelps won by a hundredth of a second.

Indivisible. Which businesses you are in. Who reports to whom. Where a decision right sits. Whether a plant exists. What your price list says. These are discrete acts, and there is no version of them that happens gradually. You cannot exit forty per cent of a business line as a pilot and see how it goes.

The failure mode this explains is specific and common: a company applies an excellent continuous improvement discipline to a portfolio that should have been cut, and gets steadily better at producing things it should have stopped producing. Every metric improves. Nothing that matters changes.

A company applies an excellent improvement discipline to a portfolio that should have been cut, and gets steadily better at producing things it should have stopped producing. Every metric improves. Nothing that matters changes.

Does the rate change the answer?

It might, and the research does not rule it out. Most improvement programs run a handful of projects a year. At fifty-two, with everyone participating, accumulation may behave differently, and nobody has tested increments at that frequency against the transformation question.

I want to put this fairly because it cuts against the evidence I have just presented and I think it has force.

The studies above observed organisations doing what organisations typically do, which is run improvement through a designated function at a modest rate. If the finding is that increments at four or six per year do not accumulate into transformation, that is a finding about four or six per year. It is not a demonstration that accumulation is impossible at any rate, and the difference between six and fifty-two is not a difference of degree in an organisation’s experience of itself.

The approach I run, which I call 3-A continuous improvement, is built on that premise: fifty-two improvements a year, one per week, each one fixing a single thing rather than everything, each one accepted at roughly seventy per cent right rather than held until it is complete. At that cadence a business is genuinely different in eighteen months, and I am not willing to describe that as merely operational.

What I cannot tell you is whether it produces the kind of change Romanelli and Tushman were measuring. My honest position is that it transforms performance decisively and still does not exit a business line, because nothing at any frequency does that. The indivisible pile stays indivisible. Whether high-frequency improvement eventually shifts strategy, structure and power by making a different configuration obviously correct is a question I have views on and no evidence for.

What does that rate actually require?

Three things most companies cannot supply. Everyone participating rather than an improvement team. Projects that fix one thing rather than everything. And a tolerance for getting it seventy per cent right, repeatedly, which most cultures quietly punish rather than reward.

These are hard preconditions rather than success factors, and failing any one of them collapses the rate back to the handful-per-year pattern the research describes.

Everyone. Fifty-two a year is not deliverable by a central function. It requires improvement to be part of what everybody does, which means an organisation where a machine operator initiates a change without asking a black belt for permission. Most companies have professionalised improvement into a department, and that department is the rate limit.

One thing. The discipline is in the scope. An improvement that fixes one specific problem can be done in a week. The same improvement expanded to address every adjacent problem becomes a project, and a project takes a quarter. Organisations drift toward the second version because it looks more serious, and the drift is what kills the cadence.

Seventy per cent. This is the one most companies genuinely cannot do. A change that is seventy per cent right gets deployed, produces a visible imperfection, and somebody gets asked why it was not finished properly. Once that has happened twice, people stop shipping at seventy and start shipping at ninety-five, which takes four times as long. The culture has to protect the seventy per cent explicitly, and very few do.

So the rate argument above is real and it does not rescue the average programme. If you cannot supply all three, you are running improvement at the frequency the research examined, and its findings apply to you.

Sort the change first. Divisibility decides the method.Divisible: can be five per cent better next monthCycle time, defect rate, changeover, quote turnaround, on-time delivery.Continuous improvement wins. A big project here disrupts work for nothing.Indivisible: no gradual version existsWhich businesses you are in, who reports to whom, whether a plant exists.Step-change required. You cannot exit a business line as a pilot.The common failureExcellent improvement discipline applied to a portfolio that should have been cut.Every metric improves. Nothing that matters changes.

If you need both, which runs first?

Step-change first, then improvement inside the new structure. Running them the other way means spending a year perfecting processes that the reorientation will then delete, and the people who built those processes will fight the reorientation for entirely understandable reasons.

Most real situations have items in both piles, which makes sequencing the practical question rather than selection.

The argument for structure first is that improvement is configuration-specific. A changeover procedure optimised over forty iterations belongs to a plant layout, a product mix and a scheduling approach. Change any of those and much of that optimisation is discarded, along with the credibility of whoever led it. Doing the structural work first costs you the improvements you did not make in the meantime. Doing it second costs you improvements you did make and then destroyed, which is considerably more expensive in goodwill.

There is one exception worth naming. Where the structural decision itself depends on information only improvement work can produce, improvement goes first and is scoped as diagnosis. If you do not know whether a plant is uncompetitive or merely badly run, a focused improvement effort answers that question and the answer determines the structural call. What matters is that everybody understands the work is being done to inform a decision rather than to fix the plant.

The failure to avoid in either order is running both simultaneously with the same people. Reorientation consumes the management attention that improvement cadence depends on, and a business attempting both usually discovers around month four that the improvement programme has quietly stopped while everybody was in structure meetings. Better to pause it deliberately and say so than to let it decay and pretend otherwise.

Why do companies default to continuous improvement?

Because it is safer, and the cost of that safety is documented. Successful firms reorient before competitive pressure forces them. Mediocre firms reorient only after a performance crisis. Failing firms never do. Waiting for the crisis is the middle pattern.

That three-tier finding comes from Tushman, Newman and Romanelli’s work on convergence and upheaval, and it is the most uncomfortable thing in this article, because reorienting before a crisis means proposing disruptive change to a business that is currently fine. There is no burning platform to point at. The proposal has to win on argument alone, against colleagues with legitimate reasons to prefer the current arrangement.

Continuous improvement requires none of that. Nobody has to say the structure is wrong. Nobody’s previous decision is implicated. Credit distributes, blame diffuses, and there is no discrete moment anybody can later be held to. It is not that people are cowardly. It is that one of the two options carries personal risk and the other does not, and that asymmetry decides a great many corporate agendas without ever being discussed.

The tell that this is happening: a leadership team that agrees privately the structure is wrong and keeps commissioning improvement work inside it. Everybody knows. Nobody will be the one to say it in the room where it would count.

Who should this article send away?

Companies whose structure is actually fine. Most are. An article arguing that increments will not get you there has an obligation to say that most readers do not need step-change, and that reaching for it unnecessarily costs more than the improvement it replaces.

The test is the divisibility sort above. Write down the five things that most need to change. If all five sit in the divisible pile, you do not have a structural problem, you have an execution problem, and a reorientation would destroy a working organisation to fix something a disciplined improvement cadence would have handled.

Step-change is expensive in ways that do not appear in its business case. It consumes management attention for a year, it stops normal improvement while it runs, it loses people who joined the previous configuration, and it resets relationships with customers and suppliers who understood the old arrangement. Those costs are worth paying when the structure is genuinely wrong and are pure waste when it is not.

Where to go instead, and none of it is mine. For the evidence itself, Romanelli and Tushman’s paper is the primary source and is better than my summary of it. For the counter-tradition, James Quinn’s work on logical incrementalism argues that strategic change does proceed incrementally, and reading the opposing case properly is more useful than reading my version of the argument it loses. And if your five things are all divisible, any competent continuous improvement practitioner will serve you better than a transformation specialist, because the specialist will find something structural to do.

Where does this break down?

This conclusion happens to sell what I sell. An article concluding that structures need replacing is an article recommending my own line of work, and you should discount it accordingly, especially in the parts where the evidence here is thinnest.

The thinnest part is the political section. The three-tier finding about when firms reorient is real; my explanation of why, in terms of personal risk and diffused blame, is a plausible story I have assembled rather than anything demonstrated. It fits what I have watched happen and that is a weaker basis than it sounds.

Three further limits. The punctuated equilibrium model has a serious academic counter-argument holding that it overlooks the latent incremental processes underpinning revolutionary changes, which would mean the increments do accumulate and simply become visible all at once; a later line of work poses ambidexterity as an alternative account entirely. My own 3-A cadence is my practice rather than a studied intervention, and the eighteen-month claim I make for it is observation rather than measurement. And the divisibility sort is my framing, useful for forcing a conversation and not validated as anything more.

The one thing I would defend without qualification is the indivisibility point, because it is not really an empirical claim. You cannot close a plant gradually. That part is arithmetic.

A leadership team that agrees privately the structure is wrong, and keeps commissioning improvement work inside it. Everybody knows. Nobody will be the one to say it in the room where it would count.

People also ask

Can continuous improvement replace transformation? Not for indivisible change. Research found small changes in strategy, structure and power distribution did not accumulate into fundamental transformation. Divisible things like cycle time and defect rate do respond to steady improvement.

How do I know which one my company needs? Write down the five things that most need to change and sort them. If all five could be five per cent better next month, you have an execution problem. If any requires a discrete act, you need step-change for that one.

Does running more improvements faster change this? Possibly, and it is untested. Most programs run a handful a year, so the research may describe low-frequency improvement rather than improvement in principle. At fifty-two a year a business is genuinely different, though it still cannot exit a business line.

Why do companies choose continuous improvement more often? Because nobody has to say the structure is wrong and no previous decision is implicated. One option carries personal risk and the other does not, and that asymmetry decides many agendas without being discussed.

When should a company reorient? Before a performance crisis, according to the research. The most successful firms initiate reorientation ahead of competitive pressure; mediocre firms wait for a crisis to justify it; failing firms never start.

About the Stagnation Assassin

Todd Hagopian is a Fortune 500 transformation executive who has generated $3B+ shareholder value across five turnarounds at Berkshire Hathaway, Illinois Tool Works, and Whirlpool. He is VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books, The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn. His newest book is 10-Minute Transformation (February 23, 2027).

The talk version of this has a room sort their own five changes into divisible and indivisible, which usually ends an argument they have been having for a year.

Want the short version you can use this week? Pick your problem and get the matching free minibook.

Find my free minibook

The full diagnostic and 90-day playbook: 10-Minute Transformation, February 23, 2027.

Preorder 10-Minute Transformation

Not what you needed? Find Your Solution →