The quick read.
Lewin (three stages)
- Built for
- Understanding why change feels hard
- Strongest at
- Simple shared language: unfreeze, change, refreeze
- Strains with AI because
- AI never refreezes. There is no end state to settle into
Kotter (8 steps)
- Built for
- One big organizational change with a finish line
- Strongest at
- Executive sponsorship, urgency, momentum
- Strains with AI because
- It is episodic. AI change is continuous and has no finish line
ADKAR (Prosci)
- Built for
- Individual adoption, person by person
- Strongest at
- Diagnosing exactly where each person is stuck
- Strains with AI because
- Light on workflow redesign, governance, and value measurement
McKinsey 7-S
- Built for
- Diagnosing organizational alignment
- Strongest at
- Showing how strategy, structure, and skills interconnect
- Strains with AI because
- It describes one moment in time and gives no sequence or cadence
Lewin's three stages.
Developed by psychologist Kurt Lewin in the 1940s, this is the model many later change models build on: unfreeze the current state, make the change, refreeze the new state. It is still the clearest way to explain to a team why change feels uncomfortable. Its limit in an AI transition is the third word. Refreezing assumes the ground stops moving, and with AI model releases arriving quarterly, it does not.
Kotter's 8 steps.
John Kotter published the model in Harvard Business Review's 1995 article 'Leading Change.' Eight sequential steps that run from creating urgency through building a coalition to anchoring the change in culture. We find it the strongest framework for winning executive sponsorship. Its limit is that it was built for a single change event, and AI change keeps going, so running eight steps every quarter is not practical.
ADKAR.
Developed by Jeff Hiatt, who founded Prosci in 1994, ADKAR tracks five outcomes for each individual: Awareness, Desire, Knowledge, Ability, Reinforcement. It is the best diagnostic on this list for a stalled deployment, because it tells you precisely where each person is stuck. Is this an awareness problem or an ability problem? Its limits are the things it never claimed to cover: it will not redesign a workflow, set your governance, or tell you whether the change produced business value. We put ADKAR and Kotter side by side if you are choosing between them.
McKinsey 7-S.
Developed at McKinsey in the late 1970s, 7-S maps seven interdependent elements: strategy, structure, systems, shared values, skills, style, and staff. It is a diagnostic lens, and that is its value. It shows leaders that buying a platform (systems) without touching skills, style, or shared values is why the platform sits unused. But it gives you no sequence and no cadence. It tells you what is misaligned, not what to do on Monday.
McKinsey Rewired.
Rewired is McKinsey's approach to digital and AI transformation, set out in McKinsey's book Rewired and in the McKinsey Quarterly article Rewired to outcompete by Eric Lamarre, Kate Smaje and Rodney Zemmel. Drawing on McKinsey's work with more than 200 large companies, it names six capabilities a company has to build: a business-led road map, a talent bench, an operating model that can scale, technology for distributed innovation, data embedded everywhere, and adoption and scaling. McKinsey delivers the work alongside the technical and implementation capabilities of QuantumBlack, AI by McKinsey.
The strength is breadth: the article finds that no digital and AI transformation can succeed without a baseline of competence across all six capabilities. The limit is that the approach is built for enterprise-wide programs, with large transformation budgets and delivery teams. For a marketing, creative, or talent function trying to get value from AI already in the building, the weight of the approach can outrun the problem.
People, Process, Platform.
People, Process, Platform is the framework behind everything we build at Average Robot, and the spine of our book, The Elephant in the Algorithm. It was designed from the AI adoption failure patterns themselves.
People asks whether your teams are aligned and equipped to use AI with intent. Process asks whether your ways of working have adapted, or whether AI is bolted onto old workflows. Platform asks whether your choices of AI and AI models serve your business strategy and your customers. The People layer is where most AI programs break, and it is where this framework puts its weight.
Using them together.
These frameworks are not mutually exclusive. A common and effective pattern is ADKAR for the human transition, McKinsey's Rewired approach for enterprise-wide transformation at scale, and People, Process, Platform as the AI-specific read that tells you what needs to change for AI to land. The diagnostic comes first. It tells you which gaps are real, so you do not spend transition effort in the wrong place.
How to choose.
- - One defined change with a hard deadline and executive attention: Kotter.
- - A specific AI product has been deployed and people are not using it: ADKAR.
- - Leadership cannot agree on what is wrong: 7-S as the diagnostic conversation.
- - A team that needs to understand why this feels hard: Lewin, in one slide.
- - Continuous AI change across many workflows, with spend already committed: none of the above alone. You need a loop that repeats every quarter.
Where the classics fit.
We built our seven-layer framework with full awareness of these models, and out of respect for what they get right. We do not repackage them, and we are not affiliated with any of them. What we do is apply the same underlying logic where it belongs: our Enable layer works from the same insight ADKAR is built on, that adoption happens one person at a time and stalls at identifiable points. Our Decide and Measure layers reflect the same discipline about sponsorship and momentum that Kotter codified. Our Diagnose layer asks the same alignment questions 7-S taught a generation of leaders to ask. If your team is already trained in one of these models, nothing in our framework asks you to abandon it. The classics handle the human transition. Our framework adds the layers AI demands around them: continuous redesign, governance, and a measurement loop that survives an AI model upgrade.
Related reading.
- - An AI change management framework: the seven-layer loop we use, in detail.
- - AI change management software, compared: what to buy and when.
- - AI readiness assessment: the diagnosis layer in detail.
- - Our book, The Elephant in the Algorithm: the longer argument behind the framework.
Questions people ask.
Which change management framework is best for AI?
None of the classics alone. Pick the one that fits your immediate change, then add continuous redesign, governance, and measurement around it, because AI change does not end.
Is ADKAR still relevant for AI?
Yes, for individual adoption it is still the sharpest diagnostic available. It was never designed to cover workflow redesign or value measurement, so pair it with a framework that covers them.
What is the difference between ADKAR and Kotter?
ADKAR tracks change one person at a time. Kotter runs change one organization at a time. ADKAR tells you why one person is not using the AI. Kotter tells you why the program lost momentum.
Can these frameworks work together?
Yes. A common pattern is ADKAR for the human transition, McKinsey's Rewired approach for enterprise-wide transformation at scale, and People, Process, Platform as the AI-specific read that tells you what needs to change for AI to land.
Do I need a new framework for AI?
Mostly you need a wrapper around the framework you already have. Keep the classics for what they are good at, and add the layers they leave out: redesign, governance, and a measurement loop.
ADKAR is a registered trademark of Prosci Inc. Kotter's 8-Step Process, Lewin's change model, the McKinsey 7-S framework and McKinsey's Rewired approach remain the work and intellectual property of their respective creators. Average Robot is not affiliated with, certified by, or endorsed by any of these organizations. References here are for comparison and commentary.