ADKAR helps an AI rollout when you need to understand what an individual still needs to make a change: a reason to participate, useful knowledge, the ability to apply it or support that keeps it going. It becomes insufficient when the answer stays at that level and nobody changes the work around the person.
A colleague may understand the purpose, want to try and know how to write a prompt, yet lack permission to export the source records. Calling that an ability barrier can help locate the problem. It does not grant the permission or decide whether exporting those records is appropriate. Those decisions need an owner.
Use ADKAR as part of the broader people and change approach, with a clear task, permitted inputs, a result someone can check and responsibility for removing obstacles.
Turn the five elements into questions about work
Prosci's ADKAR model describes five outcomes of individual change: awareness, desire, knowledge, ability and reinforcement. For an AI rollout, make those words concrete:
- Awareness. Can the person explain which problem the new practice is meant to address? For a warehouse coordinator, it might be making stock discrepancies easier for the stock controller to investigate.
- Desire. Does the person see a reason to participate, and what concern makes that difficult? They may welcome help writing a discrepancy note while worrying that a fluent draft could conceal a missing receipt.
- Knowledge. Do they know the permitted inputs, how to request the note and what to check? Knowing a prompt formula is only part of knowing this task.
- Ability. Can they complete and check the note with the actual records, permissions and time available? A successful training exercise cannot answer that by itself.
- Reinforcement. What keeps the useful practice going? The stock controller might review the first notes and explain which details made investigation easier, while the manager protects time for checking.
The questions help avoid giving another explanation to someone who needs practice, or another course to someone waiting for an access decision. They should start a conversation, rather than become five labels attached to employees.
In a Reddit discussion about practical ADKAR experience, a Reddit user described using plain phrases when introducing change management to organizations unfamiliar with the framework. Their ability question concerned whether someone could do the task in their real workflow. The original poster had found the terminology abstract; the response made it easier to connect the model to action. This was an account of general change practice, not a test of AI rollout effectiveness.
Role-specific explanation can also matter. In the comments on Tim Creasey's February 2025 LinkedIn article about AI and ADKAR, Aubree Côté described helping her organization adopt AI through topics and examples tied to people's roles. That account supports a practical question: what does this change mean in the employee's work? It does not tell us whether role-specific examples alone resolved the rollout's other constraints.
Keep the individual model inside the whole rollout
The distinction is part of Prosci's own explanation. In Andrew Horlick's account of the model and methodology, ADKAR is a model of individual change within a broader method that also addresses organizational change. Treating the acronym as a complete implementation plan gives it a job it was not designed to do.
An AI rollout still needs decisions about the work:
- Suitability. Is the proposed task worth supporting, and what would make AI assistance inappropriate?
- Access and permitted use. Which records may the person use, in which approved tool, and who decides an unresolved permission question?
- Quality and accountability. What must the finished output preserve, who checks it and who accepts responsibility for acting on it?
- Capacity and priorities. What time or competing work changes so that practice and checking are possible?
ADKAR can help surface these issues through a conversation about ability or reinforcement. It cannot settle them merely by naming them. The useful boundary is between identifying a need and taking responsibility for the action that meets it.
That also means avoiding an unfair diagnosis. A person who refuses to use an unapproved tool with customer records may be following the rules. A team that can generate summaries but cannot verify them may have an unsuitable task or inadequate sources. Low use does not establish low desire, and more encouragement may be the wrong response.
If you need to investigate capability, opportunity and motivation before choosing support, use the separate guide to diagnosing an AI adoption barrier with COM-B. Keep the frameworks useful by asking which decision each helps you make.
Follow one stock-discrepancy note through the handoff
Suppose Sam Taylor, a warehouse coordinator, is trying an approved assistant to draft a short note about a mismatch between a stock count and receiving records. Jamie Lee, the stock controller, will use the note to decide what needs investigation. The assistant must not infer a missing receipt or recommend changing the inventory quantity without evidence.
Sam understands the purpose and can produce a note in training. During the real task, one receiving record is unavailable through the approved export. Sam pauses rather than filling the gap with a plausible explanation.
The change lead can ask what Sam needs to complete the task. The answer may include practice distinguishing a documented mismatch from an unsupported explanation. It also requires the records owner to decide whether a permitted export is available, and Jamie to agree how an incomplete note should be handed over.
A useful next attempt has an explicit check: the note identifies the mismatch, cites the available records and marks the unavailable receipt as unresolved. Jamie checks those points before using it. If the source gap cannot be handled safely, the team keeps the existing process for that case.
Neither another awareness session nor a higher self-reported ability score supplies the missing receipt. Conversely, fixing the export does not show that Sam can recognize an unsupported claim. Check both the work condition and the person's practice.
Give the next action an owner and a check
Record the obstacle in ordinary language, then decide who can act. An ADKAR label can sit beside that record if it helps the change team; it should not replace the description.
| What the attempt reveals | Support or decision needed | Who should act | What to check next |
|---|---|---|---|
Sam cannot explain which details Jamie needs in the note. | Show an acceptable discrepancy note and discuss why its evidence matters. | Jamie and the change lead. | Sam can explain the required details before drafting. |
Sam knows the requirements but misses an unsupported explanation in the draft. | Practise checking the note against the receiving and count records. | Sam with Jamie's review. | The next note distinguishes evidence from an unresolved gap. |
A receiving record is unavailable through the approved export. | Decide whether permitted access is possible or the case needs the existing process. | The records owner. | The agreed route works without inventing or exposing information. |
Checking is being skipped because other work takes priority. | Agree when review happens and what work makes room for it. | The manager and Jamie. | A real note receives the required review before use. |
Revisit the record after the next attempt. If the action did not remove the obstacle, change the action or the scope. Do not keep increasing training simply because training is the easiest thing the program can supply.
Use the observed task to decide whether the new practice is usable. ADKAR responses can inform that judgment, but they are not a substitute for a permitted workflow, a dependable output and an accountable handoff.
Questions and answers
Is ADKAR enough for an AI rollout?
ADKAR can help identify the support an individual needs to change, but it is not a complete AI rollout plan. The organization still needs to choose suitable tasks, establish permitted access, define acceptable output and assign responsibility for checking and use. Prosci itself places the individual model within a broader change methodology.
Start with one task and connect the person's needs to the organizational decisions required to complete it. The rollout boundary shows which decisions need more than an ADKAR label.
What is the difference between knowledge and ability in AI adoption?
Knowledge concerns knowing how to carry out the new practice. Ability concerns being able to perform it in the working conditions that actually apply. A warehouse coordinator may know how to request a discrepancy note, yet need practice checking an unsupported explanation or a decision about an unavailable source record.
Observe a real, permitted attempt before prescribing more training. Separate missing practice from access or workflow decisions, then use the stock-note handoff to define a check for the next attempt.
Does low AI use mean an employee lacks desire?
Low AI use does not establish that an employee lacks desire. The task may be unsuitable, the required records may be unavailable or the employee may be following a restriction on the tool. Someone who wants to participate may also lack time or support to check the output.
Ask about the last attempt, the reason for pausing and the conditions required to continue. Address legitimate concerns directly. Use the owner and check table to turn the answer into an action rather than a judgment about attitude.
Can an ADKAR score prove a team is ready to adopt AI?
An ADKAR score alone cannot prove that a team is ready to adopt AI. A person's reported awareness, knowledge or confidence does not demonstrate that a particular task can be completed with permitted information and acceptable quality. A team average can also hide different obstacles.
Keep any scores as prompts for investigation. Check a specific workflow, identify who accepts its output and observe whether people can complete and verify it. The five work questions help choose what to investigate; the task evidence supports the readiness decision.
