The question comes up in every procurement process for change support, and it deserves a straight answer. The straight answer is that none of them has been tested.
That is not a rhetorical framing. There is no published study comparing organisations that applied Kotter's eight steps with matched organisations that did not, no equivalent for ADKAR, and none for the three-step model. The primary studies that a meta-analysis would pool do not exist for any of the three.
What the models do have is evidence of three different kinds, and the differences between those kinds are worth understanding, because they determine what each model is entitled to claim.
Three models, three sorts of warrant
| Model | What it rests on | What that supports |
|---|---|---|
| Kotter's eight steps | Ten-year observation of more than 100 companies, published 1995 as eight errors and 1996 as eight steps | A list of recurring failure modes. Not a success rate and not a tested remedy. |
| ADKAR | Practitioner benchmarking surveys run by its owner, with a self-selected sample of participants | A description of what practitioners report doing. Not a comparison with practitioners who did something else. |
| Unfreeze, change, refreeze | Derivation from Kurt Lewin's field theory, with the attribution itself disputed | A theoretical account of why change might work. No outcome data of any kind. |
Kotter. The origin is a ten-year observation of more than a hundred companies attempting transformation. That method is well suited to identifying what keeps going wrong and poorly suited to producing a rate or testing a remedy, and the step from eight observed errors to eight prescribed steps assumes that avoiding the errors produces the opposite result. The details are in what Kotter's model is based on.
ADKAR. The model breaks change adoption into awareness, desire, knowledge, ability and reinforcement, and it is supported by the longest running body of practitioner research in the field. Prosci has run benchmarking studies since 1998, twelve editions covering more than 10,800 professionals, and the model was distilled from that data in 2003.
That is a substantial dataset and the size is not the issue. The design is. Prosci describes its recruitment as soliciting participation through online media outlets and directly from the 90,000 members of its own Change Management Learning Center, which means the respondents are disproportionately people already invested in change management as a discipline and often already trained in this method. There is no comparison group of organisations that used no method at all, and the research is designed, run and published by the organisation that owns and licenses the model. What it establishes well is what practitioners in that community report doing and how they rate the result. What it cannot establish is what would have happened otherwise.
The three-step model. Whether Lewin proposed it at all is an unresolved dispute in the literature. Either way it has never been evaluated as an intervention.
Why certification is not evidence
All three models have certification programmes attached, and certification is regularly presented in procurement as a quality signal.
A certification tests whether a person can reproduce a method. It says nothing about whether the method produces better outcomes, because the certification body has no incentive to find out and no mechanism for doing so. The number of certified practitioners measures the commercial success of the certification, which is a different thing from the effectiveness of the model, and the two are frequently confused in vendor material.
This is worth separating from the question of whether certified change managers are useful. Someone who has been trained in a structured method, and has run twenty programmes, brings experience and a shared vocabulary that a team without either does not have. That is a real benefit. It is a benefit of the person and the practice, not a validated property of the framework on the certificate.
The state of the field
The scholarly literature has been aware of this gap for some time and has not closed it. Stouten, Rousseau and De Cremer opened their 2018 review in the Academy of Management Annals by observing that organisational research lacks an easily accessible consensus on basic change management, which is a measured way of saying that a field with several competing models and no outcome data has not produced an answer that practitioners can use.
Contemporary organizations often struggle to create meaningful, sustainable changes. At the same time, relevant organizational research lacks an easily accessible consensus on basic change management.
The contrast with neighbouring fields is instructive rather than damning. Leadership training has been tested across 335 independent samples. Workplace coaching has three meta-analyses. Both are interventions delivered to individuals, which makes them far easier to randomise than an organisational restructure, and that asymmetry explains most of the gap. You cannot run a controlled trial of a merger.
What you can do, and what has largely not been done, is measure comparable changes across many organisations with a consistent definition of the outcome. The reason that has not happened is partly methodological and partly that the people with the data sell the models.
Choosing between them anyway
Since the evidence does not discriminate, the choice has to be made on fit. Four questions do more work than a comparison of the models.
Is the change primarily technical or primarily behavioural? If the difficulty is that several thousand people must each do something differently, a model organised around individual adoption, which is ADKAR's design, matches the problem. If the difficulty is political and the obstacle is a coalition that has not been built, Kotter's framing is closer.
Is there a defined end state? Refreezing assumes there is. A continuous programme with no completion date does not fit a model with a terminal step, and forcing it produces a fictional milestone.
Who has to use the model? A framework that a change team applies to an organisation needs different properties from one that four hundred line managers are expected to internalise. Eight steps is already too many for the second case.
What is already in the building? An organisation that has trained fifty people in one method will get more from using it than from switching, and the cost of the switch is real while the benefit is unevidenced.
And one thing not to do: do not accept a failure rate as the reason to buy a method. The 70 percent figure has been traced through five published sources with no empirical basis found, and it is the standard opening of a change management sales conversation precisely because it converts a discretionary purchase into a necessary one.
Common questions
Which change management model is most effective? None has been shown to be more effective than another, because none has been tested against a comparison group. The choice is about fit with the change at hand rather than about evidence.
Is ADKAR evidence-based? It is supported by benchmarking research conducted by the organisation that owns it, with a self-selected sample of practitioners and no comparison group. That describes what practitioners report doing; it does not establish that the model improves outcomes.
Has Kotter's eight-step model been tested? No controlled evaluation has been published. It originates in a ten-year observation of more than a hundred companies, which is a design for identifying recurring errors rather than for testing a remedy.
Does change management certification mean a method works? No. A certification tests whether a person can reproduce a method. Experienced certified practitioners bring real value, but that is a property of the person and their practice rather than evidence for the framework.
Why is there no good evidence in change management? Largely because organisational change cannot be randomised the way an individual intervention can. You cannot run a controlled trial of a restructure. What could be done, and has not been, is consistent outcome measurement across many comparable changes.
How should we choose a model then? On fit: whether the change is technical or behavioural, whether it has a defined end state, who has to use the model day to day, and what the organisation already knows.
Does this mean change management does not work? No. It means the case for a specific framework cannot be made from outcome evidence, because none exists. Individual practices within these models have support from other literatures, and experienced practitioners are useful for reasons that have nothing to do with which certificate they hold.
Sources
- Stouten, Rousseau & De Cremer (2018), Successful Organizational Change: Integrating the Management Practice and Scholarly Literatures, Academy of Management Annals 12(2), 752-788
- Hughes, M. (2016), Leading changes: Why transformation explanations fail, Leadership 12(4), 449-469
- Cummings, Bridgman & Brown (2016), Unfreezing change as three steps: Rethinking Kurt Lewin's legacy for change management, Human Relations 69(1), 33-60
- Hughes, M. (2011), Do 70 Per Cent of All Organizational Change Initiatives Really Fail?, Journal of Change Management 11(4), 451-464
- Prosci, Change Management Research Process
- Prosci, Best Practices in Change Management, 11th Edition executive summary (PDF)
