The short answer
ADKAR is a model of how one person takes on a change, developed by Jeff Hiatt, the founder of Prosci. The letters stand for Awareness, Desire, Knowledge, Ability and Reinforcement. In practice, a person needs to know why, want to take part, know how, be able to do it in the real job and have something that keeps the new way going. You use it by having each person rate the five elements from 1 to 5 and putting your effort into the first one that scores 3 or lower. Repeat the rating every four to six weeks, because the place where people get stuck moves as the change goes on.
- Created by Jeff Hiatt, founder of Prosci, in 1996.
- Level The individual. Kotter’s 8 steps cover the organization, and the change curve the emotions along the way.
- Start At the first element that scores 3 or lower on a scale from 1 to 5.
What is the ADKAR model?
ADKAR describes how a person goes from first hearing about a change to working the new way. Jeff Hiatt, who founded Prosci in 1994, developed the model in 1996 and first published it in a 1999 white paper called The Perfect Change, according to Prosci. Prosci says the model rests on research into the change patterns of more than 700 organizations (Prosci, in German, The Open University). The elements come in a fixed order. It’s hard to learn something you don’t want, and hard to want something when you don’t see why.
| Element | What the person needs | What you hear when it’s missing |
|---|---|---|
| Awareness | Understanding why the change is needed | “Why are we doing this?” |
| Desire | Choosing to take part in and support the change | People get it, but keep working the old way |
| Knowledge | Knowing how to change and what it takes | “I don’t know where to start.” |
| Ability | Doing it in practice, in the real job | It works in the training, but not on a busy Tuesday |
| Reinforcement | Keeping the new way going | Three months later, everything is back to how it was |
Prosci itself calls ADKAR a model for individual change, not a complete change management methodology (Prosci). Which workflows to change and how to measure the value is work for the whole organization, and I cover it in my guide to AI transformation.
What is the ADKAR barrier point?
Have each person rate the five elements from 1 to 5. Prosci calls the first element that scores 3 or lower the barrier point (Prosci). Put your effort there, since the elements after it won’t hold until it’s solved. In AI rollouts, I think a common mistake is starting with training when the team hasn’t understood why, or doesn’t think it’s worth the time.
An ADKAR assessment template for AI
Send these five statements in an anonymous form, with a scale from 1 (strongly disagree) to 5 (strongly agree), and add one question about how often people use the tool. For a change other than AI, swap the word.
- Awareness. I understand why we’re bringing AI into our own work.
- Desire. I want to use AI in my work and think it’s worth the time.
- Knowledge. I know how to use the tool in my own tasks and which rules apply.
- Ability. I use AI in my normal work, even when I’m busy.
- Reinforcement. We have follow-up and routines that keep us using AI.
- Use. How often do you use the tool in your work? Weekly, less often or never.
Find each person’s barrier point, meaning the first statement scored 3 or lower, and count how many people have it at each element. Put your effort where most are stuck. If several are stuck at Desire, offer everyone a short one-on-one. The answers are anonymous, so you don’t know who it is, and Prosci’s advice is that building Desire starts with listening (Prosci).
Repeat the assessment after four to six weeks and whenever the tool changes. Prosci recommends assessing at several milestones and, when a change rolls out in stages, measuring Knowledge and Ability before each stage.
A one-page ADKAR checklist for AI
Go through it with the team or the leadership team. The first element where you can’t tick most boxes is roughly your barrier point, and the unticked boxes there are what to do first. The template above gives a sharper picture per person.
Awareness
- ☐ We’ve written down why we’re bringing in AI, in one sentence per team.
- ☐ Everyone knows which tool is approved and which data must never go into it.
- ☐ The manager has said what we know and don’t know about how roles will change.
- ☐ Employee representatives (a union or works council, where you have one) were told early.
Desire
- ☐ Each person has picked a task of their own to start with.
- ☐ The manager shows how they use AI themselves.
- ☐ The manager has listened to what worries people, in one-on-ones or in the team.
Knowledge
- ☐ Training is built on the team’s own tasks, in the tool they’ll use.
- ☐ Prompts that worked are in a shared document.
- ☐ Everyone knows how to review AI output before using it.
Ability
- ☐ There’s time in the calendar to practice every week.
- ☐ Someone on the team can answer questions the same day.
- ☐ Access and logins work.
Reinforcement
- ☐ We’ve measured time or quality before and after in the chosen workflows.
- ☐ The new way is in routines, templates and onboarding, so it doesn’t depend on a few people.
- ☐ The manager follows up at team meetings.
- ☐ We reassess when the tool gets new features.
How do you use ADKAR in an AI rollout?
The example is hypothetical. An HR team of twelve has an AI tool the organization has approved, and is supposed to use it to write job ads, summarize anonymized free-text answers from pulse surveys and answer managers’ common policy questions. After three months, four of the twelve use it every week.
The team fills in the template. Two people have their barrier point at Desire, six at Knowledge, one at Ability and three at Reinforcement. Most are stuck at Knowledge, so more meetings about why would be wasted time.
1. Awareness. Why AI, and why now?
Everyone scores 4 or 5 here. Early on, the manager explained the reason (less time on job ads, more time for candidates), what nobody knew yet about how roles would change and which data must never go into the tool. Without rules like that, people turn to their own tools, which is how shadow AI starts.
Ask the team. Can you say in your own words why we’re doing this, and what you must never put into the tool?
2. Desire. Is it worth my time?
Two people score Desire 3 or lower. The form doesn’t say who, so the manager offers everyone a short one-on-one. In those talks, one person says their calendar is full and AI feels like one more thing to learn. Another worries about their job. The manager lets each person pick a task they’d rather not do as a first test, and at the next team meeting shows one of their own attempts that went badly. For the conversation about job security, my guide to AI anxiety at work has talking points.
Ask the team. Which task would you most like to hand over, and what worries you?
3. Knowledge. How do I do this in my own tasks?
This is where half the team sits. They had a general introduction, but nobody showed them the three workflows in their own tool. The manager books a three-hour workshop where the team runs the workflows on their own material, anonymized where needed. They save the prompts that work in a shared document and go through how to review an AI-written job ad. For prompts to start from, see ChatGPT prompts for HR, and for who to train first across the organization, the AI literacy playbook.
Ask the team. Do you know how to take the first step in each of the three workflows?
4. Ability. Does it work on a busy Tuesday?
Knowing how doesn’t mean doing it when the job ad has to go out today. One person is stuck here from the start, and after the workshop this is where the problem moves. What helps is time in the calendar and someone who answers questions the same day, which is where the team puts its effort after the second assessment.
Ask the team. When did you last write a job ad with AI, and if you stopped, what made you stop?
5. Reinforcement. Will we still do this in three months?
The usage question shows that three of the four weekly users have their barrier point here. They use the tool on their own initiative, but nobody follows up and nothing is written into the routines. The team starts measuring minutes per job ad and the number of rewrites, and the manager brings one example to every team meeting.
Ask the team. Will we still work like this in three months, and how will we know?
Six weeks later, the barrier point moves
The team repeats the assessment. Now one person has the barrier point at Desire, two at Knowledge, six at Ability and three at Reinforcement, so the effort moves to Ability. The team gets 30 minutes a week to do real tasks with AI together. A colleague who acts as AI champion answers questions the same day, and IT fixes the last access problems. Without the second assessment, the team would have kept training long after training stopped being the problem.
Where do AI rollouts get stuck?
BCG’s AI at Work 2026 survey asked 11,749 people in 14 markets, about 1,000 each in the US and the UK (BCG). Respondents are frontline employees (BCG’s term for white-collar staff without managerial roles), managers and leaders (BCG). The survey doesn’t ask about ADKAR, but four of its findings line up with four of the elements. The mapping is mine.
Swedish HR data points to gaps in two of the same elements, Knowledge and Reinforcement. In the AI & HR Barometern 2026, humAIn Academy’s survey of 329 HR professionals in Sweden, 61 percent say they’ve had formal AI training from their employer, but only 25 percent had training that covers law, risk or the AI Act. And 70 percent say they don’t measure the effect of AI in HR. Two questions for your own organization. Did the training cover rules and risks, and do you measure the effect of AI at all?
BCG, AI at Work 2026, published June 2026. AI & HR Barometern 2026, humAIn Academy, 329 HR professionals in Sweden, collected 8 June to 24 September 2026. Respondents opted in and are likely more interested in AI than HR professionals overall. Results and method are in State of AI in HR.
What is the change curve, and does it hold up?
The change curve goes back to Elisabeth Kübler-Ross’s 1969 book On Death and Dying, about how dying patients responded. The five stages from the book are denial, anger, bargaining, depression and acceptance (EKR Foundation). Her foundation’s workplace version, the Kübler-Ross Change Curve, has seven areas, which are shock, denial, frustration, depression, experiment, decision and integration. The foundation says people feel several at once and skip some, and that the curve is not a timetable for when anyone should be done (EKR Foundation). If you want the foundation’s diagram for a training deck, it says you need a license.
The research behind the stages is thin. A review by Margaret Stroebe, Henk Schut and Kathrin Boerner found that the stage model of grief lacks sound empirical evidence (Omega, 2017). Logan Holdsworth and Todd Bridgman, who traced how the model moved from dying patients to organizational change, add that the workplace version treats emotions as universal and linear and resistance too narrowly (Holdsworth and Bridgman, 2025). Bridgman also points out that someone who objects may think the change is unnecessary or badly designed, and that calling it denial is patronizing (Bridgman, 2025).
So I’d use the curve to talk with people about how they’re doing and keep it out of the project plan. And I wouldn’t treat an objection as a phase to wait out.
The change curve when AI arrives (my proposal)
The change curve when AI arrives is my proposal for how reactions may run when an organization introduces AI. Its four phases are first contact, the trial phase, the everyday dip and a new way of working. Unlike the classic curve, there is no single shock. The dip comes when the tool meets real work, and the curve starts over when the tools change.
| Phase | What happens | ADKAR element |
|---|---|---|
| 1. First contact | Curiosity, worry or indifference, but no shock. The tool shows up while nothing else changes. | Awareness and Desire |
| 2. Trial phase | First attempts on simple tasks | Knowledge |
| 3. Everyday dip | Real tasks under time pressure. Output is uneven and the old way feels faster. | Ability |
| 4. New way of working | The workflow is redesigned, the rules are clear and the effect is tracked | Reinforcement |
BCG’s 2026 survey finds both strain and enjoyment. Among regular AI users, 41 percent report more mental strain since they started using AI, and 67 percent say they enjoy their work more (BCG). BCG calls it a joy paradox, where AI makes work both better and harder. The survey doesn’t follow people over time, so it can’t confirm a dip.
A team can be in phase four with one workflow and in phase one with the next. The curve is a proposal for conversations with teams, and I haven’t tested it in a study.
How is ADKAR different from Kotter’s 8 steps?
ADKAR follows each person through a change, while Kotter’s 8 steps describe what leaders do for the whole organization. The change curve is about how people feel along the way. You don’t have to pick one.
| Model | Level | Question it answers | Use it for | Weak spot |
|---|---|---|---|---|
| ADKAR | Each person | Where is this person stuck? | Following up each team | Says little about redesigning work |
| Kotter’s 8 steps | The whole organization | What should leaders do next? | The leadership team’s plan | Tested only in parts |
| Change curve | Each person’s emotions | How are people doing? | Conversations with people | Weak evidence, and no timetable |
John Kotter presented his model in Harvard Business Review in 1995, after watching more than 100 companies try to remake themselves (HBR), and expanded it in his 1996 book Leading Change. Here are the steps as Kotter’s firm names them today (Kotter), applied to AI.
| Step | In an AI rollout |
|---|---|
| 1. Create a sense of urgency | Show why now, with your own examples. |
| 2. Build a guiding coalition | Leaders plus HR, IT, employee representatives and a few heavy AI users. |
| 3. Form a strategic vision | One sentence on what AI is for, and the first two or three workflows. |
| 4. Enlist a volunteer army | AI champions in each team. |
| 5. Enable action by removing barriers | Approved tools, clear rules, working access, time in the calendar. |
| 6. Generate short-term wins | One workflow that saves time within weeks, measured before and after. |
| 7. Sustain acceleration | The next workflow and the next team. |
| 8. Institute change | Routines, goals, onboarding and how managers follow up. |
Kotter and his firm have since turned the linear 8 steps into 8 Accelerators, described in his 2014 book Accelerate (Kotter). For AI, I’d run the steps in parallel, since one team can be at step 7 while the next starts at step 1.
What are ADKAR’s strengths and limits?
ADKAR’s strength is that it gives managers and HR a shared language for where someone is stuck. The assessment takes a few minutes and can be repeated, and it turns a vague “people are resisting” into one element you can work on.
- A low score can point to the change itself. If a whole team scores low on Desire because the tool doesn’t fit their work, fix the tool or the workflow before you work on motivation.
- ADKAR comes from a consulting firm. Prosci sells certification built on the model (Prosci). That doesn’t make the model wrong, but the source has a commercial interest.
- Kotter’s model has only been tested in parts. A 2012 review found support for most of the steps but no study of the model as a whole (Appelbaum et al., 2012).
- ADKAR and Kotter weren’t built for AI. Both assume a change that leaders start, but with AI many people started on their own. In BCG’s 2025 survey, 54 percent said they would use AI tools even if they weren’t authorized (BCG). One person can be at Ability with their own tool and at Awareness with the one you bought, so find out what people already use before you plan. My own framework starts with leaders using AI in the open, and the guide to AI adoption explains how it differs from ADKAR.
- None of the three says much about the work itself. In BCG’s 2026 survey, companies that redesign workflows end to end do better than those that only deploy tools, on both value and how much employees enjoy their work (BCG). How to pick and redesign the first workflows is in my guide to AI transformation.
Frequently asked questions
What is the ADKAR model?
ADKAR is a model of how one person takes on a change, developed by Jeff Hiatt, who founded Prosci. Its five elements come in a fixed order, and the idea is that an organization only changes when the people in it do.
What does ADKAR stand for?
Awareness, Desire, Knowledge, Ability and Reinforcement. Each letter is an outcome a person needs to reach before a change sticks.
How do you do an ADKAR assessment?
Have each person rate one statement per element from 1 to 5. The first element that scores 3 or lower is their barrier point, and that’s where you focus. Repeat the assessment after a few weeks, since the barrier point moves as the change progresses.
What is the Kübler-Ross change curve?
The Kübler-Ross change curve describes how people react emotionally to a major change, based on Elisabeth Kübler-Ross’s 1969 account of how dying patients responded. It can help you recognize reactions. Research doesn’t support fixed stages, though, so use it to talk with people and keep it out of the project plan.
What is the difference between ADKAR and Kotter’s 8 steps?
Kotter’s 8 steps describe what leaders do to drive a change across the organization. ADKAR describes what has to happen for each person. In an AI rollout, I’d use both.
Does ADKAR work for AI adoption?
It can, as a way to see where each person and team is stuck. It says little about redesigning workflows, though, and AI tools change often. Repeat the assessment after a few weeks and whenever the tool gets new features.
Sources
- Prosci: The Prosci ADKAR Model
- Prosci: ADKAR, core to the people side of change
- Prosci: About
- Prosci: The ADKAR model (German edition)
- The Open University: Prosci’s ADKAR change model
- Prosci: Individual barriers to change and what to do about them
- Prosci: Change management certification program
- EKR Foundation: The five stages of grief
- EKR Foundation: The Kübler-Ross Change Curve
- Holdsworth and Bridgman (2025), Journal of Organizational Change Management
- Bridgman (2025), Victoria University of Wellington: Four problems with applying ‘five stages of grief’ model to workplace restructures
- Stroebe, Schut and Boerner (2017), Omega
- Kotter (1995), Harvard Business Review
- Kotter Inc: The 8 steps for leading change
- Appelbaum et al. (2012), Journal of Management Development
- BCG (2026): AI at Work, strategy matters more than tools
- BCG (2026): AI at Work slideshow (PDF)
- BCG (2025): AI at Work press release
- State of AI in HR (AI & HR Barometern 2026)
ADKAR® is a registered trademark of Prosci, Inc. Kübler-Ross Change Curve® is a registered trademark of the Elisabeth Kübler-Ross Family Limited Partnership.
Want to see where your AI rollout is stuck?
In an AI change program, I work with your leaders, managers and HR for 3 to 12 months, with pulse surveys and adoption tracking that show where each team gets stuck. Many organizations start with a leadership workshop. The leadership teams I’ve worked with give an average rating of 4.97 out of 5, measured across 13 groups. If you’d rather keep reading on your own, I write about AI, HR and leadership in the newsletter FullStack HR.