The ADKAR Model describes five outcomes individuals need to adopt and sustain a change: Awareness of the need for change, Desire to participate and support it, Knowledge of how to change, Ability to demonstrate the required skills and behaviors, and Reinforcement to sustain the change. Developed by Prosci, it focuses on the people side of transformation. An enterprise can deploy a cloud AI platform successfully while employees continue using old workflows because one or more of these outcomes is missing.
Awareness requires a credible explanation of why the organization is changing. For an AI-assisted support workflow, leaders should explain the business problem, expected improvements, limitations, and consequences of doing nothing. A general announcement that AI is strategically important is insufficient. Employees need to understand what will change in their own work, which decisions remain with people, and how customer data will be protected.
Desire concerns willingness to participate, not simply understanding the announcement. Resistance may reflect concerns about job security, workload, professional identity, unreliable outputs, or unfair performance evaluation. Leaders should invite affected employees into workflow design, explain likely role changes honestly, provide reskilling opportunities, and resolve conflicting incentives. More technical training rarely fixes a lack of trust or a belief that adopting AI will threaten someone’s career.
Knowledge covers what employees must learn, including approved use cases, data-handling rules, prompt techniques, output verification, and escalation procedures. Ability is different: people must demonstrate those skills under realistic working conditions. Engineers might understand retrieval-augmented generation in a classroom yet still need supervised practice to diagnose unsupported answers or configure access correctly. Protected learning time, coaching, accessible tools, and representative exercises help turn knowledge into performance.
Cultural barriers require targeted leadership interventions. Risk aversion can be addressed through bounded pilots, explicit approval criteria, least-privilege access, and clear accountability. Fear of failure calls for psychological safety: leaders should distinguish good-faith experimentation within approved boundaries from negligent behavior and use blameless reviews to extract lessons. Resistance to change often requires listening, employee participation, and visible action on legitimate concerns rather than labeling employees as difficult.
Reinforcement makes the new behavior sustainable. Managers should recognize responsible use, incorporate validated practices into operating procedures, provide ongoing coaching, and review whether the change improves business outcomes. For example, an AI support pilot should measure resolution quality, rework, and appropriate escalation alongside usage. Rewarding prompt volume alone can encourage unnecessary use or discourage employees from reporting errors.
ADKAR is useful as a diagnostic framework, not a rigid project schedule or a substitute for technical governance. Outcomes generally build on one another, but teams may revisit earlier needs as tools, roles, or risks change. Interviews, practical assessments, and workflow observations add context that telemetry cannot provide. The trade-off is that individualized support takes time and leadership attention; a uniform rollout is simpler but can overlook very different barriers across teams. Cloud controls can enable safe practice, but no service creates employee willingness or trust automatically.
Amazon CloudWatch