Change Management Models Compared: What the Evidence Supports

Here is a hypothetical situation many operations and HR leaders will recognize. This guide follows it throughout.

A regional insurer is merging its three claims offices into one national claims team. Simple claims will go to a fast-track group, and complex claims to specialists. The chief executive has read John Kotter and wants an eight-step plan. HR has budgeted for ADKAR training. A consultant’s slide deck shows the change curve and opens with the claim that 70% of change programs fail.

You are the operations lead, and you have to decide how to run the change. Which change management model fits? What does each model actually claim? And what does the evidence say about any of them?

The short answer starts with six models: Lewin’s three-step model, Kotter’s 8 steps, Prosci’s ADKAR, Bridges’ transition model, the Kübler-Ross change curve and McKinsey’s 7-S framework. Most are process checklists or descriptive frameworks built from consulting and management experience. Each can help organize part of the work.

The research reviews discussed below found no controlled comparison showing that any of them makes organizational change succeed more often than a sensible alternative. And none tells you which behaviors have to change, for whom, or what is stopping them. That diagnosis is your job, and it should decide how you use the models.

Four kinds of change management model

Organizational change management is the work of moving an organization to a new way of operating: a merger, a restructure, a new process or system. A change management model is any framework that claims to help, and the label covers four kinds of tool.

• Process checklist: a sequence of actions for the people leading the change, such as Kotter’s eight steps. It tells you what to do and roughly when, not why people act as they do.
• Descriptive stage model: an account of how people are said to experience change over time, such as the change curve. It predicts a pattern of reactions, not what to do about a particular one.
• Diagnostic framework: a set of categories to check, such as McKinsey’s 7-S. It helps you look in the right places; it does not rank causes or predict effects.
• Causal theory: an explanation of what produces a behavior and what would change it, stated so evidence could prove it wrong. Kurt Lewin’s field theory comes closest here.

The difference shows up when a change stalls. Suppose adjusters keep working claims from their old regional lists. Kotter’s checklist says you are at the barrier-removal step (“enable action by removing barriers,” in Kotter Inc.’s current wording), but not which barrier. A causal account says why they stay: on the regional list an adjuster can pick claims that close quickly and count toward the office bonus, and the shared queue takes both away. That tells you what to change: the scorecard and the bonus, not the speeches.

A model’s popularity is not evidence that it produces that kind of answer. A model can be taught in every MBA program because it is memorable and easy to apply, not because using it makes change more likely to succeed.

The six models compared

Here is the comparison at a glance. Four models have glossary entries, linked in the table, with fuller histories. The sections below explain each verdict and apply it to the claims merger.

Lewin’s three-step model: unfreeze, change, refreeze

What it claims. As usually taught, Lewin’s change model says you unfreeze the current way of working, move to a new one, then refreeze it so it lasts. Its origin is disputed: Stephen Cummings and colleagues argue that Lewin wrote only a brief passage and later writers built the model, while Bernard Burnes replies that the labels summarize Lewin’s field theory.

What kind of thing it is. As taught, a checklist. As Lewin wrote it, part of a causal theory: behavior holds at its current level because forces for change balance forces against it. You can add forces or reduce the opposing ones, and Lewin wrote, as Burnes quotes, that “as a rule, the second method will be preferable.”

Evidence. Neither side of the dispute points to comparative tests of the three-step version as a method. Reema Harrison and colleagues’ 2021 healthcare review found it in 11 of 38 studies; across all models, the reviewers could not tell whether using one contributed to the results.

In the claims merger. A typical “unfreeze” phase adds pressure. Lewin’s preference points the other way: reduce what holds adjusters to their regional lists first.

Kotter’s 8 steps

What it claims. Kotter’s 8-step model comes from his 1995 Harvard Business Review article, “Leading Change: Why Transformation Efforts Fail,” and his 1996 book Leading Change. It says transformations fail when leaders skip or rush steps that run from creating urgency to instituting the change. Kotter drew it from watching more than 100 companies. Most of their efforts, he wrote, fell between success and failure, “with a distinct tilt toward the lower end of the scale.”

What kind of thing it is. A process checklist for senior leaders.

Evidence. Steven Appelbaum and colleagues’ 2012 review found support for most individual steps but no formal studies of the whole model. Its popularity, they judged, rests more on its usable format than on scientific consensus. Julien and Rachel Pollack’s action research found the linear, top-led picture too simple. Kotter appeared in 19 of Harrison’s 38 studies, more than any other model, and the reviewers could not tell whether it contributed.

In the claims merger. The merger will succeed or fail at removing barriers, and Kotter does not say which ones matter here.

Prosci’s ADKAR model

What it claims. Prosci’s ADKAR model is, in Prosci’s words, “a model for individual change.” Each person needs Awareness, Desire, Knowledge, Ability and Reinforcement. Jeff Hiatt, Prosci’s founder, developed it in the 1990s and set it out in a 2006 book.

What kind of thing it is. An individual change sequence that can serve as a diagnostic: find the first element a group lacks.

Evidence. Harrison’s review found one ADKAR study among 38, used with another model. In Prosci’s practitioner survey of more than 2,600 change practitioners, 88% of those who rated their change management excellent met or exceeded objectives, against 13% of those who rated it poor. That survey covers change management in general, not ADKAR. The same people rated both the program and the result, so it shows a correlation in self-reports, not an effect.

In the claims merger. A survey might show adjusters low on Desire. ADKAR frames that as a stage in each person’s progress, but here the cause sits in a system, the scorecard, which 7-S is better placed to find.

Bridges’ transition model

What it claims. William Bridges distinguished change, the external event, from transition, the internal process of coming to terms with it. William Bridges Associates describes three phases: an ending, a neutral zone “when the old is gone but the new isn’t fully operational,” and a new beginning. Bridges first wrote about personal life transitions in 1980 and applied the model to organizations in his 1991 book Managing Transitions.

What kind of thing it is. A descriptive stage model with practical advice, especially about endings.

Evidence. Thin. Logan Holdsworth and Todd Bridgman’s 2025 history of the change curve places Bridges in that lineage. They note that his claim of a universal emotional pathway was not firmly grounded in empirical research.

In the claims merger. Bridges’ question is still a good one: what is ending for these people? Adjusters lose their regional office identity, a manager many have reported to for years, and working relationships with local repair shops. Naming those losses can expose a design choice. If local relationships speed up certain claims, perhaps they should not end at all.

The Kübler-Ross change curve

What it claims. The change curve says people move through emotional stages similar to denial, anger, bargaining, depression and acceptance, while performance dips and recovers. The stages come from Elisabeth Kübler-Ross’s 1969 book On Death and Dying, about terminally ill patients. Kübler-Ross herself extended them to grief and loss in 1974. Later writers generalized them to all life transitions without testing that step, and brought them into organizations from the 1980s.

What kind of thing it is. A descriptive stage model borrowed from a very different situation.

Evidence. Holdsworth and Bridgman found little high-quality research validating the curve, from Kübler-Ross onward. The curve treats emotions as universal and sequential, and if every objection counts as denial, it can never be wrong. They still credit it as a relatable way to talk about emotions.

In the claims merger. If adjusters say the shared queue will hurt their numbers, calling that “denial” skips a factual claim you can check in a day.

McKinsey 7-S framework

What it claims. Robert Waterman, Thomas Peters and Julien Phillips first published the framework in “Structure Is Not Organization” (Business Horizons, 1980), after working it out with Anthony Athos and Richard Pascale. They argued that organizational effectiveness comes from the interaction of seven elements: strategy, structure, systems, style, staff, skills and superordinate goals, later renamed shared values. The diagram has no starting point, and the authors wrote that it is difficult, perhaps impossible, to make significant progress in one element without progress in the others.

What kind of thing it is. A diagnostic framework. Peters, in his own history of the model, calls it an “organization effectiveness diagnostic.”

Evidence. The authors’ support was practical. They said they had tested it in teaching, workshops and direct problem solving, and that “it seems to work.” That is consultants’ judgment, not a controlled test. Harrison’s review found one healthcare study that used 7-S alongside Lewin. None of the reviews discussed in this guide reports a controlled test of 7-S.

In the claims merger. Here, where the obstacle sits in the scorecard, 7-S is the model most likely to point at it. The structure has changed to one national queue. The systems have not: the scorecard still counts claims closed, and bonuses still follow office results. 7-S does not say how to fix that, but it points straight at it.

Where “70% of change programs fail” comes from

Now return to the consultant’s opening slide. The figure became famous through “Cracking the Code of Change,” a 2000 Harvard Business Review article by Michael Beer and Nitin Nohria. It states that “about 70% of all change initiatives fail” and cites no study.

Mark Hughes of Brighton Business School went looking for one. His 2011 paper reviewed five published instances of the figure and concluded that there is “no valid and reliable empirical evidence” to support it.

The earliest of the five is Michael Hammer and James Champy’s 1993 book Reengineering the Corporation. They offered what they called an “unscientific estimate” that as many as 50 to 70 percent of organizations that undertake reengineering do not achieve the dramatic results they intended. Hughes shows how that caveated guess turned into a headline failure rate. “There is no inherent success or failure rate for reengineering,” Hammer later wrote with Steven Stanton. When Hughes looked for support behind the general 70% figure, the trail led back to estimates like this one, and to Beer’s own earlier guess about quality programs, not to a study.

Failure statistics have long served as sales arguments; the 1980 article that introduced 7-S cited a Fortune suggestion that perhaps as many as 90 percent of carefully planned strategies don’t work. In his 2019 textbook, Hughes notes that academics can use myths such as the 70% figure to sell their theories, models and concepts.

Do not replace the myth with a better-sounding number. Nobody knows the failure rate of organizational change, partly because “failure” is rarely defined in advance. For the claims merger, write down what success means, and how you will measure it, before launch.

What the research on change management as a whole supports

The evidence gaps above are not unique to these six models. Eric Barends and colleagues systematically reviewed 30 years of organizational change management research. Of 563 studies, most were one-shot designs with low internal validity, giving weak grounds for separating an intervention’s effect from other causes. Replications were rare, and the authors urged skepticism.

Harrison’s healthcare review found 12 methods used across 38 studies from 2009 to 2020, mostly as guiding principles. The authors wrote that “it was not possible to detect whether the use of a model, method or process contributed to the success.”

Jeroen Stouten, Denise Rousseau and David De Cremer compared widely used practitioner models with the scholarly research. They note that popular models more often cite expert opinion than scientific evidence.

Rousseau and Steven ten Have put the starting point bluntly in their guide to evidence-based change management. “Acting without identifying the real problem is the root of many failed changes.” They recommend first-hand observation and weighing several definitions of the problem before choosing a solution.

In the behavior change guide, I described change management, including Kotter and ADKAR, as largely a practitioner discipline with limited experimental evidence. I also argued that most change initiatives name outcomes such as “improve collaboration” without specifying the behaviors that make them up. The research above supports the first point. The rest of this guide addresses the second.

Start with the behaviors, then choose the model

I treat behavior change as a matching problem. Most programs ask how to make people do X. The better question is what other behavior, Y, naturally appeals to the people in question and achieves the same outcome. So define the outcome first, explore several behaviors that could reach it, judge each on its appeal and practicality for the people involved, and only then design the intervention.

For organizations, the behavior change guide adds two points: redesign the context before trying to change the people, and use person-environment fit in role design, so people do work that suits them. Different groups usually need different behaviors, so run the sequence once per group.

Here is the claims merger in six steps, which the change diagnosis worksheet follows.

Step 1: Define the outcome and what success means

“One national claims operation” is a structure, not an outcome. The outcome the merger should produce is faster settlement of simple claims without more errors, and more consistent handling of complex claims.

Write down how you will measure it before launch: median days to settle fast-track claims, the share of settled claims later reopened, and complaints per thousand claims. Add one harm check. If settlement speeds up while reopened claims rise, you have built what I call a Pyrrhic Intervention: one metric improves while a more important one gets worse.

Step 2: List the behaviors that must change, and whose

Name the groups and a first candidate behavior for each. For the merger, three groups matter most:

• Intake staff classify each new claim as fast-track or specialist, using the triage rules, on the day it arrives.
• Adjusters take their next claim from the shared national queue instead of from their old regional list.
• Team leads run weekly reviews on cycle time and reopen rates by queue, instead of comparing offices.

Each has an actor, an action and an occasion; “embrace the new operating model” has none. These are candidates, not commitments. Step 4 tests alternatives.

Step 3: Find what prevents each behavior

Watch people do the work, ask about specific recent claims, and check the queue data. Record whether each finding was observed, reported or assumed. The COM-B model offers useful categories: capability, opportunity and motivation.

Suppose the investigation finds three obstacles. Intake staff are unsure how to classify water-damage claims, a capability problem. The scorecard counts claims closed and bonuses follow office results, so taking the next national claim can lower an adjuster’s count and earn nothing for their office: a system problem that shapes motivation. And many adjusters rely on local repair shops whose numbers live in their own phones, an opportunity problem the shared queue ignores.

It is tempting to call this status quo bias. But persistence alone does not establish bias. Switching carries real costs here, a worse score and lost local knowledge, and those costs explain the behavior.

Step 4: Test alternative behaviors for appeal and practicality

For each group, list alternatives that reach the same outcome, then ask of each: will these people find it appealing, and is it practical for them? The behavior change guide’s criteria help: is it compelling, simple enough to do on a worst day, rewarding soon enough, and useful?

For intake, the claim form could route most claims automatically from four fields, with staff reviewing only the exceptions. Interviews would show whether staff welcome that or see it as a loss of skilled work.

For adjusters, the strongest alternative is on the person side. Some may prefer the pace of fast-track claims; others, the depth of complex ones. Letting them choose a queue, within limits, puts people where the work suits them. Claims that need a local inspection could go to small regional pods, keeping the repair-shop relationships that already work.

This is the core of Behavior Matching; Behavior Market Fit explains why the chosen behavior can matter more than any later tactic.

Step 5: Give each model a specific job

Now the models become useful, because each has a defined task:

• 7-S checks whether systems match the new structure. Here it points to the scorecard and the office bonus pool.
• Lewin’s force-field logic sets the order: reduce restraining forces, such as the scorecard and unclear triage rules, before adding pressure.
• Bridges makes leaders name what is ending and decide what should continue, such as repair-shop relationships.
• Kotter sequences leadership communication. Count behavior measures, such as same-day classification, as the short-term wins.
• ADKAR checks, group by group, whether the reason, the method and the support are in place.
• The change curve is, at most, a reminder to listen. Treat every objection as a claim to check.

Step 6: Decide how you will know it worked

Measure behaviors and outcomes separately: the first show whether people do the new work, the second whether it helps. If one office launches triage a month before the others, track its same-day classification and queue uptake against targets set before launch, and fix obstacles before the other offices start. The later offices give a rough comparison for early outcomes they also record, such as complaints per thousand claims. Settlement time moves more slowly; compare it at three months with the months before launch, and treat that comparison with more caution.

The research methods guide and the field experiment entry explain sound comparisons. Research on the diffusion of innovations adds a useful question for the rollout: can other teams try the change and see its results?

Set the decision rule in advance. If shared-queue uptake rises but settlement time does not fall, the behavior is happening and not helping: revisit the design, not the communication plan.

Where this approach does less

A behavior-first diagnosis earns its keep when a change depends on many people choosing to work differently. It does less in two situations.

The first is a simple behavior the system can enforce, such as a second approval that the payment system requires before releasing large payments. Rousseau and ten Have give a similar example when explaining how to make a change last: if the only way to get expenses reimbursed is an online system, people will use it. Build the control well, explain it, and check for workarounds.

The second is a change that is mostly an ending, such as closing an office. For the people leaving, the questions are fairness, notice and support, and Bridges’ attention to endings matters most. So does Holdsworth and Bridgman’s report from a downsizing at their own university: staff found it patronizing to be told the curve proved they would come to accept the change. The diagnosis still applies to the people who stay.

If the change is mainly adopting a new tool, use a narrower guide. AI adoption at work treats low use as a possible sign of poor task fit, not automatically resistance, and the Behavioral State Model example diagnoses low use of a reporting tool.

Try it: bedside shift handoffs

A short hypothetical case: a hospital’s nursing director wants nurses to move shift handoffs from the nurses’ station to patients’ bedsides, and proposes an ADKAR readiness survey and a Kotter-style launch. What would you do first, and what can each model tell you?

A good answer. Specify the behavior: which nurses, which patients, what time, and what happens at the bedside. Then watch a few handoffs. The shift overlap may be too short, roommates may make private details hard to discuss, and some patients will be asleep. Those are opportunity questions a readiness survey cannot settle.

Then compare alternatives and ask nurses which they would choose, such as a brief bedside safety check after a station report, or bedside handoffs only for patients who are awake and in single rooms. The survey can then check whether nurses understand the reason and the format, and Kotter’s steps can organize the launch. Measure whether handoffs happen at the bedside, whether key information is missed, and how long they take.

Neither model told you about the shift overlap. Watching the work did.

Use the change diagnosis worksheet

The change diagnosis worksheet turns the six steps into a page your team can fill in before writing the change plan, with a blank copy and the claims merger as a filled, hypothetical example. Fill it in for one group first; if you cannot name a specific behavior for that group, you are not ready to choose a model.

Which change management model should you use?

For the claims merger, use several of the models, each with a narrow job, after the behavior diagnosis rather than before it. Treat the change curve as a reminder to listen, never as a diagnosis. And drop the 70% slide.

Your next step is to fill in the change diagnosis worksheet for one group in your own change before you commit to any change management process. For individual behavior change, Behavior Change Models Compared applies the same test to 16 frameworks, and Behavioral Design: An Overview explains why behavior selection comes before design.

References

• Appelbaum, S. H., Habashy, S., Malo, J.-L., & Shafiq, H. (2012). Back to the future: Revisiting Kotter’s 1996 change model. Journal of Management Development, 31(8), 764–782. https://doi.org/10.1108/02621711211253231
• Barends, E., Janssen, B., ten Have, W., & ten Have, S. (2014). Effects of change interventions: What kind of evidence do we really have? Journal of Applied Behavioral Science, 50(1), 5–27. https://doi.org/10.1177/0021886312473152
• Beer, M., & Nohria, N. (2000). Cracking the code of change. Harvard Business Review, 78(3), 133–141. https://hbr.org/2000/05/cracking-the-code-of-change
• Burnes, B. (2020). The origins of Lewin’s three-step model of change. Journal of Applied Behavioral Science, 56(1), 32–59. https://doi.org/10.1177/0021886319892685
• Cummings, S., Bridgman, T., & Brown, K. G. (2016). Unfreezing change as three steps: Rethinking Kurt Lewin’s legacy for change management. Human Relations, 69(1), 33–60. https://doi.org/10.1177/0018726715577707
• Harrison, R., Fischer, S., Walpola, R. L., Chauhan, A., Babalola, T., Mears, S., & Le-Dao, H. (2021). Where do models for change management, improvement and implementation meet? A systematic review of the applications of change management models in healthcare. Journal of Healthcare Leadership, 13, 85–108. https://doi.org/10.2147/JHL.S289176
• Hiatt, J. (2006). ADKAR: A model for change in business, government and our community. Prosci Learning Center Publications.
• Holdsworth, L., & Bridgman, T. (2025). From coping with dying to coping with organizational change: The bricolage of the change curve’s evolution. Journal of Organizational Change Management, 38(8), 254–269. https://doi.org/10.1108/JOCM-05-2025-0479
• Hughes, M. (2011). Do 70 per cent of all organizational change initiatives really fail? Journal of Change Management, 11(4), 451–464. https://doi.org/10.1080/14697017.2011.630506
• Hughes, M. (2019). Managing and leading organizational change. Routledge.
• Kotter, J. P. (1995). Leading change: Why transformation efforts fail. Harvard Business Review, 73(2), 59–67. https://hbr.org/1995/03/leading-change-why-transformation-efforts-fail-2
• Pollack, J., & Pollack, R. (2015). Using Kotter’s eight stage process to manage an organisational change program: Presentation and practice. Systemic Practice and Action Research, 28(1), 51–66. https://doi.org/10.1007/s11213-014-9317-0
• Rousseau, D. M., & ten Have, S. (2022). Evidence-based change management. Organizational Dynamics, 51(3), 100899. https://doi.org/10.1016/j.orgdyn.2022.100899
• Stouten, J., Rousseau, D. M., & De Cremer, D. (2018). Successful organizational change: Integrating the management practice and scholarly literatures. Academy of Management Annals, 12(2), 752–788. https://doi.org/10.5465/annals.2016.0095
• Waterman, R. H., Jr., Peters, T. J., & Phillips, J. R. (1980). Structure is not organization. Business Horizons, 23(3), 14–26. https://doi.org/10.1016/0007-6813(80)90027-0