Why every AI adoption stall point is a people problem, not a technology one
Published on LinkedIn August 3, 2026
This is the sixth issue in a series building on Digital Applied’s “Change Management for AI Adoption: A 2026 Playbook,” published June 14, 2026. Their research framework is the foundation. This newsletter adds the practitioner layer, covering what these patterns actually look like inside organizations, where the standard response falls short, and what to do first.
Every stall point in the ADKAR model traces back to a human problem. Not a model problem. Not an integration problem. A human one. That observation sounds obvious until you watch an organization spend six months troubleshooting its AI deployment infrastructure while the actual blocker sits two levels up, in an unanswered question about what the tool is supposed to do for the person using it.
ADKAR is not new. It has been the dominant individual change framework for two decades. What is new is applying it to a technology that keeps moving underneath the people being asked to adopt it, and understanding which stages break down differently for AI than for any other enterprise rollout.
What ADKAR is and why it fits AI
ADKAR maps five milestones that every individual moves through during a change: Awareness, Desire, Knowledge, Ability, and Reinforcement. The model was developed by Prosci and is built on a straightforward premise. Organizations do not change. People change, one at a time, and the organization changes when enough of them do.
For most enterprise technology rollouts, the model works well enough. A system goes live, people are trained, the project closes, and ADKAR gets applied once. AI breaks that assumption. The tools change monthly. Use cases keep expanding. The capability available in December is not what anyone was trained on in January. Applying ADKAR once to an AI rollout is like installing a navigation system and then never updating the maps.
The practical implication is that AI adoption requires ADKAR to run as a continuous cycle rather than a one-time project milestone. Most organizations are not designed for that, and most AI change programs are not funded for it.
Where each stage actually breaks down
The Digital Applied playbook, building on Prosci’s research across 1,107 professionals, maps the most common stall point at each stage. The pattern across all five is consistent: every stall is human. The training gap, the unanswered WIIFM, the tool that works in a demo but breaks down in real work, and the adoption metric that stops once the rollout team disbands. None of those are model failures.
Awareness is where most AI programs lose people before the change even starts. Understanding of AI’s actual impact is low across most workforces, and when people do not understand what the tool is for or what problem it is solving, they disengage before Desire ever gets a chance. The warning sign is familiar to any practitioner: the hallway conversation about whether jobs are at risk. When that conversation is happening, Awareness has not landed.
Desire is where the WIIFM gap from last week’s newsletter shows up in the model. Broadcast communications answer the organization’s desire, not the individual’s. When a role cannot articulate what changes specifically apply to them, Desire remains at zero, and the rest of the model works from an empty foundation. Rising shadow AI usage is the diagnostic signal here. When people route around the sanctioned tool to get work done, the desire to use it officially never formed.
Knowledge is the stage most organizations underinvest in and then underestimate its consequences. SurveyMonkey’s 2026 research found that only 13% of U.S. workers received any employer AI training. DataCamp reported that the share of organizations offering formal AI upskilling fell to 26% in 2026, down from 35% the prior year. Those numbers are moving in the wrong direction, while AI investment is moving in the opposite direction. Without Knowledge, usage spikes after a training session and collapses within weeks because people run out of road.
Ability is where the demo-to-reality gap lives. The tool works under controlled conditions with clean data and a prepared scenario. It breaks against the actual workflow, with its edge cases, its legacy systems, and its time pressure. When Ability fails, the workforce reverts to the manual path, not out of resistance but for practical reasons. The warning sign is the phrase every change practitioner has heard: “I’ll just do it the old way.”
Reinforcement is the stage that most AI programs treat as optional, only to wonder why usage drifts. Adoption is recognized and measured at launch; the rollout team disbands; the tools change beneath the workforce; and there is nobody left to close the loop. A decline in active usage rates six months after go-live is the most reliable indicator that Reinforcement was treated as a project milestone rather than an operating discipline.
The stage organizations most commonly skip
Practitioners who have run multiple AI rollouts will recognize the pattern. Organizations do not start at Awareness. They start at Knowledge, schedule the training, and assume the earlier stages took care of themselves somewhere in the communications plan.
They did not.
When Knowledge lands on a workforce that has not moved through Awareness and Desire, the training attendance is fine, and the comprehension scores look acceptable. Then usage data comes back six weeks later, and the question becomes why people are not applying what they learned. The answer is that they never wanted to in the first place, and nobody established why they should.
The fix is not more training. The fix is sequencing. Awareness and Desire require their own investment, earlier in the program, before the go-live calendar is built. That investment takes longer and does not produce the visible milestones that a project plan tracks. It is also the only thing that makes the rest of the model work.
The Reinforcement problem specific to AI
Every change program struggles with Reinforcement. AI programs struggle with it more acutely because the thing being reinforced keeps changing.
When a new CRM goes live, reinforcement means maintaining the habits established during training. When an AI tool goes live, reinforcement means maintaining habits around a capability set that will be materially different in six months. The prompt library from January is outdated by June. The use cases from the kickoff workshop are a fraction of what the tool can do by year-end.
This creates a specific failure mode that most AI change programs are not designed for. The reinforcement plan gets built around the tool as it exists at launch. The workforce falls behind the tool’s evolution, the gap between what people know how to do and what the tool can do keeps widening, and the organization mistakes declining usage for resistance when it is actually disorientation.
The organizations staying ahead of this are treating capability refresh as a standing operating cadence rather than an episodic training event. Regular office hours, prompt library updates tied to tool version changes, and champion networks that surface new use cases from early adopters are what Reinforcement looks like for a moving target. Most of that work is invisible on a project plan and critical to sustained adoption.
Sources
ADKAR model and stall points applied to AI: Prosci, “8 Ways AI-Driven Change is Different” (1,107 professionals) Building on: Digital Applied, “Change Management for AI Adoption: A 2026 Playbook,” June 14, 2026 https://www.digitalapplied.com/blog/change-management-ai-adoption-2026-overcoming-resistance-playbook
13% employer AI training figure: SurveyMonkey, 2026 AI Workforce Survey Cited in Digital Applied, June 14, 2026 (above)
26% organizations offering formal AI upskilling (down from 35%): DataCamp, 2026 Cited in Digital Applied, June 14, 2026 (above)
76% vs 25% adoption with and without employer support: Bright Horizons, 2025-2026 research Cited in Digital Applied, June 14, 2026 (above)
