ADKAR presents change as a sequence: Awareness → Desire → Knowledge → Ability → Reinforcement. You move through the phases, ideally in order. Once you’ve achieved Reinforcement, the change is sustained.

AI adoption rarely follows that path. It’s iterative. Teams experiment, hit walls, backtrack, try again. Awareness deepens over time—you don’t fully grasp the implications until you’ve lived with the tools. Desire waxes and wanes. Knowledge accumulates in fits and starts. Ability develops through trial and error. Reinforcement competes with reverting to old habits as tools evolve and org priorities shift.

The model isn’t wrong. It’s built for a different kind of change—one that’s more discrete, more planned, more linear. For AI, we need an adapted view.

Non-Linear Awareness

Awareness isn’t a one-time event. It’s recurring. Teams discover new implications as they use the tools: “oh, we didn’t realize the agent would do X” or “we didn’t anticipate that our process would break here.” Awareness deepens through experience. Design for it: regular retrospectives, sharing sessions, “what we learned” updates. Treat Awareness as ongoing, not a checkbox from the kickoff.

Cyclical Desire

Desire isn’t stable. Early excitement fades. Frustration with tool limits or failures can dampen it. A new capability or a visible win can reignite it. Desire oscillates. Reinforcement—the things that sustain the change—also sustain desire. When desire drops, it’s often a signal: the experience isn’t delivering, or the narrative has gone stale. Respond by improving the experience or refreshing the story.

Iterative Knowledge and Ability

Knowledge and Ability develop through use. You don’t “complete” them in a training program. You build them by doing, failing, and learning. Design for iteration: low-stakes practice environments, safe spaces to experiment, mechanisms to share learnings. Treat training as a starting point, not the end state. Ability grows in the field.

Continuous Reinforcement

Reinforcement isn’t “we did it, we’re done.” Tools evolve. People turnover. Orgs restructure. What reinforced the change last year may not work next year. Reinforcement is ongoing: defaults, norms, metrics, and narrative need regular attention. Build reinforcement into the operating rhythm—quarterly check-ins, annual refreshers—not as a one-time campaign.

The Adapted Model

Think of ADKAR for AI adoption as a spiral, not a ladder. You cycle through Awareness, Desire, Knowledge, Ability, and Reinforcement—but each cycle deepens. You’re not “done” with Awareness when you move to Desire; you revisit it as new implications emerge. The same for each phase. The model still applies; the timeline and the assumption of linearity don’t. Plan for iteration.