“People need more training” is one of the most expensive untested assumptions in change management.
Training can solve a Knowledge problem. It cannot make an unclear decision understandable, turn an unfair trade-off into willing participation, create system permissions or stop an old incentive from rewarding the previous behaviour.
ADKAR is useful because it separates these problems. Its five outcomes—Awareness, Desire, Knowledge, Ability and Reinforcement—give a change team a diagnostic vocabulary. The value is not the acronym itself. The value is choosing a different response when the mechanism is different.
Start with an observable change
Before assessing people, define what adoption would look like.
Weak:
Employees adopt the new AI assistant.
Stronger:
Field technicians review the assistant’s recommendation, record any safety-critical exception and obtain human approval before closing a visit.
The stronger statement establishes a role, context, behaviour and control. It also prevents login rate from being mistaken for safe adoption.
Separate three evidence layers:
| Layer | Question | Example |
|---|---|---|
| Delivery | Does the capability exist? | Mobile workflow released |
| Adoption | Is the behaviour occurring? | Exceptions recorded before closure |
| Outcome | Is the intended result changing? | Fewer missed safety checks without slower urgent response |
ADKAR primarily diagnoses the adoption layer. It does not prove that the change itself is strategically or ethically sound.
Diagnose by affected segment
An organisation-wide average conceals the mechanism. Consider four groups using the same mobile workflow:
- office dispatchers have reliable connectivity and controlled test data;
- experienced technicians believe the workflow removes justified discretion;
- contractors understand the purpose but lack access to the coaching channel;
- new technicians complete the task but receive conflicting instructions from supervisors.
Their earliest barriers may be Knowledge, Desire, Ability and Reinforcement respectively. One campaign or course cannot address all four.
Segment only on meaningful differences in role, impact or conditions. Avoid demographic profiling unless it is necessary, lawful and protected—for example, to meet accessibility or language obligations.
Use evidence, not labels
For each segment, ask what would support or weaken each outcome.
Awareness
Can people explain, in their own words, why the current state is changing, what evidence informed the decision and what remains uncertain?
Email delivery and town-hall attendance show exposure, not understanding.
Desire
Are people willing to participate after considering personal and collective effects? What workload, status, safety, trust or fairness issue shapes the choice?
Desire is ethically sensitive. Silence is not consent. Compliance under threat is not enthusiastic adoption. A person may understand the business case and still oppose a poorly designed change for defensible reasons.
Knowledge
Do people know the new procedure, decision rules, exceptions, roles and help path?
Knowledge evidence can include scenario questions, demonstrations and teach-back. A recall quiz alone is weak when the work requires judgement.
Ability
Can people perform under realistic time, data, tool, access and exception conditions?
Ability is where many “training problems” reveal themselves as system problems. If the account lacks permission, the interface fails offline or managers provide no practice time, another course will not create capability.
Reinforcement
Do feedback, systems, leadership behaviour and consequences support the new way after the initial launch?
Recognition can help, but reinforcement is not a celebration plan. It includes removing obsolete forms, aligning metrics, correcting defects, maintaining help and showing that responsible feedback changes the system.
Find the earliest gap
The practical sequence is:
If Awareness evidence is weak, later scores do not justify beginning with advanced training. If Awareness and Desire are supported but performance fails in the field, work on Ability.
Treat the earliest barrier as a hypothesis, not a diagnosis of personality. The Prosci source presents the elements as sequential while also stating that application is not linear: needs can shift as workflows and technology evolve. That makes reassessment essential.
Match the response
| Earliest barrier | Useful response | Common wrong response |
|---|---|---|
| Awareness | Show decision evidence, consequences and uncertainty; enable questions | Repeat the slogan more loudly |
| Desire | Address impacts, voice, trust, workload, incentives and genuine choice | Label dissent as resistance |
| Knowledge | Role-specific instruction, examples, rules and accessible help | Generic launch communication |
| Ability | Practice, coaching, time, permissions, usable tools and realistic exceptions | Retake the same course |
| Reinforcement | Align managers, measures, defaults, feedback and recognition | One launch celebration |
Every action needs a prediction. “Run coaching” is activity. “Increase correct offline exception handling from 58% to 75% without more safety overrides” is a testable adoption result.
Use PDCA/PDSA to plan the support action, state the prediction, study results and adapt. Use Stakeholder Mapping when power, harm and voice are not yet represented.
Mini-case: the successful launch that did not change work
A service company reports:
- 94% training attendance;
- 82% account activation;
- 36% able to explain the operational reason;
- 68% passing a happy-path simulation;
- 41% using the workflow correctly during connectivity loss;
- supervisors still accepting the old spreadsheet.
The launch activity is strong. Adoption is not.
There are at least three different actions:
- Rebuild Awareness with operational and safety evidence for groups that cannot explain the decision.
- Test Ability with offline exceptions, correct permissions and coached field practice.
- Repair Reinforcement by changing supervisor acceptance rules and retiring the old path only after a safe fallback exists.
The numbers do not justify inferring weak motivation. Desire needs its own evidence, including candid discussion of discretion, workload and trust.
ADKAR through fiction: distributed change
Fiction Lab’s analysis of The Rise of Endymion examines how a transformative idea can spread without making its originator a permanent centre of control.
ADKAR provides a useful, explicitly limited lens:
- a message reaching the network is not Awareness unless local participants can explain why it matters;
- charisma may create Desire but can also hide coercion or dependency;
- repeatable teaching distributes Knowledge;
- local practice, interpretation and decision rights create Ability;
- peer evidence and founder-independent institutions provide Reinforcement.
The analogy also exposes a limitation. A civilisation-scale movement is not an employee rollout, and collective legitimacy cannot be reduced to five individual outcomes. The fiction lens is valuable precisely because it asks what ADKAR leaves outside the frame: power, institutions, dissent and the right to refuse.
Do not turn ADKAR into surveillance
An adoption profile can contain sensitive inferences about employees. Use the minimum necessary data:
- prefer role or context aggregates;
- separate coaching notes from performance records;
- restrict access and retention;
- provide a confidential route for concerns;
- never let AI infer Desire, loyalty or emotion from behavioural traces;
- do not punish the person who reveals a design failure.
The purpose of diagnosis is to improve support and the system—not to make reluctance more legible for discipline.
Limits of the model
ADKAR is a practitioner model with strong usability and limited independent causal validation. Published applications show how organisations have used it, but case reports do not establish that ADKAR caused success or outperforms alternatives.
It also focuses on the individual. A complete change effort must address strategy, leadership, organisational design, resources, power, labour relations, technical quality and external conditions. Use the sequence as a practical diagnostic heuristic, not a universal law of human change.
Practical next step
Choose one affected segment and complete this sentence for each outcome:
We have evidence of [outcome] because [observable signal]; confidence is [low/medium/high] because [limitation].
Find the first weak statement. Design one bounded support action, one outcome measure and one guardrail. Reassess after the evidence date.
References
- Prosci. “The Prosci ADKAR Model.” Official overview (opens in a new tab). Accessed 15 September 2026.
- Hiatt, Jeffrey M. ADKAR: A Model for Change in Business, Government and Our Community. Prosci Learning Center Publications, 2006.
- Creasey, Tim. “The Prosci ADKAR Model: Why It Works.” Prosci (opens in a new tab), updated 18 April 2025.
- Adelman-Mullally, T. et al. “The use of change theory to facilitate the consolidation of two diverse Bachelors of Science in Nursing programs.” Nursing Outlook, 65(2), 2017. DOI (opens in a new tab).
- Mölders, S. et al. “Expanding the success factors of change management by incorporating crisis preparedness in the emerging AI world.” Review of Managerial Science, 2026. DOI (opens in a new tab).
Continue with the tool
Use the ADKAR guide and structured workspace to create a segmented diagnosis, matched support plan and reassessment record.