Prosci's ADKAR model has been a cornerstone of organizational change management since Jeff Hiatt formalized it in the late 1990s. The five elements — Awareness, Desire, Knowledge, Ability, Reinforcement — describe the individual journey through change, and they describe it accurately.
What's surprising is how rarely this framework gets applied to AI adoption programs, despite AI adoption being, definitionally, a large-scale organizational change initiative.
Let me walk through the model and show exactly where most AI rollouts break down.
Awareness: "Why is this change happening?"
ADKAR starts with awareness of the need for change. Employees need to understand not just that AI tools exist, but why the organization is investing in them, what the expected business impact is, and — critically — what happens if adoption doesn't happen.
What most organizations do: Announce the AI initiative. Send a company-wide email. Brief the leadership team.
What's missing: A coherent narrative that connects AI adoption to something employees actually care about — their own work experience, their career development, the organization's competitive position. Without this, the announcement is noise.
Practical implication: Awareness communication needs to be role-specific, not generic. A data analyst and a sales manager have different AI adoption journeys. Generic "AI is coming" messaging creates awareness of a threat, not of an opportunity.
Desire: "What's in it for me?"
Desire is the stage most commonly omitted in enterprise AI programs. Organizations proceed as though awareness of the technology's existence is sufficient motivation to adopt it.
It isn't.
Employees will adopt AI tools when they believe the tools will make their specific work meaningfully easier, better, or more impactful. They will resist when they believe the tools threaten their job security, expose them to judgment for making errors, or add complexity without visible benefit.
What most organizations do: Skip this stage entirely. Assume that professional compliance will carry adoption.
What's missing: Structured conversations about individual AI benefits. Champions who model adoption and share wins. Explicit de-risking of the "will AI replace me?" anxiety that underlies most passive resistance.
Practical implication: You cannot train desire. You can cultivate it by showing, in concrete terms, how AI makes specific roles easier. Manager-level conversations matter more here than company-wide communications.
Knowledge: "How do I use this?"
Knowledge in the ADKAR model is distinct from Ability — it refers to knowing what to do, not yet being able to do it consistently.
Most AI training programs conflate these two stages and deliver "AI awareness" sessions that cover what the tools can do without giving employees the specific operational knowledge they need for their role.
What most organizations do: Run a general AI overview workshop. Provide access to documentation. Offer optional training modules.
What's missing: Role-specific use cases. Concrete examples of how AI applies to the actual work employees do every day. Prompt templates and workflow integrations calibrated to specific job functions.
Practical implication: Knowledge building must be specific to be effective. A financial analyst learning to use AI needs to see it applied to financial analysis, not content generation or code writing. Role specificity is the difference between training that transfers and training that doesn't.
Ability: "Can I do this consistently?"
Ability is where the Dunning-Kruger dynamic applies most directly. Employees who have completed Knowledge-stage training often believe they have reached the Ability stage. They have not.
Ability requires repeated practice under real conditions, with feedback, until the behavior becomes reliable and automatic. This takes time and coaching — it cannot be compressed into a training session.
What most organizations do: Conflate Knowledge completion with Ability development. Move on after training.
What's missing: Structured practice opportunities. 1-on-1 coaching for employees who are stuck. Workflow integrations that force productive AI use rather than making it optional.
Practical implication: Ability is built through doing, not through watching. The organizations that achieve genuine AI adoption create explicit opportunities for employees to practice in low-stakes contexts before AI use is expected in high-stakes workflows.
Reinforcement: "How do we make this stick?"
Reinforcement is the stage that ensures change is maintained. Without it, employees revert to old habits — not because they don't know how to use AI, but because the organizational environment doesn't consistently reward the new behavior.
What most organizations do: Complete the rollout. Declare success. Move on to the next initiative.
What's missing: Ongoing measurement of adoption quality. Recognition systems for AI power users. Continuous learning opportunities as tools evolve. Management practices that normalize AI-assisted work.
Practical implication: Reinforcement is not a phase you complete — it is an ongoing organizational practice. AI tools evolve. Workflows change. Employee skill levels are not static. Building reinforcement into how work is managed is the only reliable path to sustained adoption.
The Pattern
The failure mode I see most consistently in enterprise AI adoption looks like this:
- Strong Awareness (the announcement was loud)
- Weak Desire (no one addressed the individual value question)
- Generic Knowledge (training covered features, not role-specific use)
- Assumed Ability (training completion was treated as capability confirmation)
- No Reinforcement (the rollout was declared complete)
The result: 90% of the organization has been through an AI training program. 10% is actually using AI in a way that produces measurable output improvement.
ADKAR is not a new framework. It has been validated across thousands of organizational change programs. The organizations that apply it rigorously to AI adoption will outperform those that treat AI as a software deployment problem.
The ADKAR model is a registered trademark of Prosci Inc. This article applies the framework to AI adoption contexts based on field research and practical implementation experience.