What Kotter and ADKAR Teach Leaders About the Real Barrier to AI Adoption

3 days ago

5 min read

A RAND Corporation study published in 2024, based on interviews with sixty-five data scientists and engineers with at least five years of experience building AI and machine learning systems, set out to understand why so many AI projects fail. The researchers noted that by some estimates, more than 80% of AI projects fail to deliver the business value they were built for, roughly twice the failure rate of ordinary IT projects.

The most important detail is what the interviewees identified as the primary cause: 84% pointed to leadership-driven problems rather than technical ones, citing a persistent gap between what executives believed AI could do and what the teams building it were actually equipped to deliver (Ryseff et al., 2024). All of which goes to say: the models mostly work, but the organizations around them are not changing fast enough to use them well.

An Old Problem in a New Package

This is not a new problem wearing a new name. It is the oldest problem in change management, arriving in an unfamiliar package. Leaders who would never announce a merger, a restructuring, or a new strategic direction without a change plan are routinely rolling out AI tools with none at all, treating adoption as a training question when it is really a change question. Two frameworks that predate the current AI cycle by decades explain much of why that approach fails, and together they give coaches a useful diagnostic for clients stuck somewhere in an AI rollout that has quietly gone nowhere.

Kotter's Eight Steps, Applied to AI

The first framework is John Kotter's eight-step model, built from his research on why large organizational changes so often collapse (Kotter, 1996). Kotter's sequence begins with establishing a sense of urgency, a step most AI rollouts skip because the urgency feels self-evident to a leadership team that has been reading about AI for two years. It rarely feels as self-evident to the employee whose daily workflow the tool is about to disrupt. From there, Kotter's steps move through building a guiding coalition, forming a strategic vision, enlisting a volunteer army to communicate it, removing barriers to action, generating short-term wins, sustaining acceleration, and finally institutionalizing the change so it survives once the initial push fades.

Skipping urgency leaves an initiative without the organizational will to survive its first budget review, echoing RAND's finding that leadership misalignment on what problem AI is meant to solve is the most common failure driver. Skipping the guiding coalition produces tools that live entirely inside one department, resented rather than adopted anywhere else. Skipping short-term wins leaves projects that technically launch and then fade because no one can point to a specific result that justified the disruption, a pattern RAND traces to underinvestment in the deployment infrastructure that would let early wins actually surface. Skipping institutionalization explains why so many AI pilots that succeeded quietly disappear a year later when the champion who drove them changes roles.

ADKAR and the Individual Level of Change

Kotter's model operates at the organizational level, and that is precisely its limitation for AI adoption, which depends on whether individual employees actually change how they work. This is where Jeff Hiatt's ADKAR model, developed by Prosci from research on hundreds of organizations, becomes a necessary complement (Hiatt, 2006). ADKAR breaks individual change into five sequential outcomes: Awareness that a change is happening and why, Desire to participate in it, Knowledge of how to do so, Ability to actually execute it, and Reinforcement that makes the new behavior stick instead of reverting under pressure.

Most AI training programs start at Knowledge, handing employees a tool and a tutorial, and skip past the two stages that actually determine whether anyone uses it. Awareness without a credible business reason produces polite compliance followed by quiet abandonment. Desire is the stage leaders underestimate most severely, because an employee's motivation to adopt an AI tool is entangled with a genuine, often unspoken concern about what fluency with it will mean for their role. A training session cannot manufacture desire that a leader has not first addressed directly. Reinforcement is the stage almost every organization drops once the launch event ends, which is exactly why usage metrics so reliably spike at rollout and decay within a quarter.

Where the Two Models Meet

Used together, the two models cover ground that either one leaves exposed alone. Kotter explains whether the organization has built enough structural support for an AI initiative to survive contact with reality. ADKAR explains whether the individuals inside that structure have actually been taken through the psychological sequence required to change how they work. A rollout can execute every Kotter step well and still fail if individual employees never move past Awareness. A rollout can win over every individual employee and still fail if there is no guiding coalition protecting it past the first reorganization.

A Diagnostic Conversation for Coaches

For a coach working with a leadership team on a stalled AI initiative, this pairing turns a vague complaint (e.g. people just are not using it) into a specific diagnostic conversation. Ask where the guiding coalition sits and whether anyone outside the initial sponsor's own department has a stake in its success. Ask what short-term win the team can point to that took less than ninety days to produce. Then ask the individual-level question underneath the organizational one: for the people closest to the work, has anyone addressed directly what this tool means for their role, moving them from awareness to desire, or has the organization simply assumed that training equals adoption?

How This Plays Out in Practice

Consider the shape this takes in practice. A mid-sized professional services firm rolls out an AI drafting tool to its client-facing teams. Leadership announces it at an all-hands meeting, points to the productivity numbers from the pilot group, and gives everyone a login. Three months later, adoption has flatlined at the same handful of early enthusiasts who used it during the pilot. Nothing about the tool has changed. What is missing is not a second training session. It is a guiding coalition that includes respected senior staff who were not part of the original pilot, a short-term win specific to the teams who have not yet adopted the tool, and, underneath all of that, an honest conversation about the desire gap, the quiet worry among mid-level staff that visible fluency with the tool is being read by the firm as a signal about which roles it plans to need less of. No amount of additional Knowledge closes a gap that actually sits at Desire.

The Bottom Line

The number from the RAND report that should reframe how leaders think about AI adoption is 84—the percent of practitioners who pointed to leadership-driven problems, not technical ones, as the primary reason AI projects fail. AI initiatives are failing for reasons that have little to do with whether the model is good enough and everything to do with whether the organization did the unglamorous work of taking people through change in order. Kotter and ADKAR were built to diagnose exactly that gap. The tools involved are new. The way organizations resist changing around them is not.

References

Hiatt, J. M. (2006). ADKAR: A model for change in business, government and our community. Prosci Learning Center Publications.

Kotter, J. P. (1996). Leading change. Harvard Business School Press.

Ryseff, J., De Bruhl, B. F., & Newberry, S. J. (2024). The root causes of failure for artificial intelligence projects and how they can succeed: Avoiding the anti-patterns of AI. RAND Corporation. https://www.rand.org/pubs/research_reports/RRA2680-1.html