In Prosci’s research with 1,107 change professionals, user proficiency accounted for 38% of reported AI implementation challenges and technical challenges for 16%. These are categories in that study, not a universal failure rate. They should still make leaders ask whether their people have the judgement, time and support to work differently.
And yet, when I look at how organisations allocate their AI transformation budgets, the ratio is inverted. The vast majority goes to technology, including licences, infrastructure and model development. A smaller share goes to change management, and less still to the harder work of helping people reconsider their roles, expertise and value.
This is the cognitive bottleneck, and it helps explain why some AI transformations stall even when the technology is ready.
The Nature of the Bottleneck
Earlier technologies changed how people communicated, recorded information and coordinated work. They also changed roles and processes. AI adds a further challenge when a system produces a recommendation that someone may act on without examining the underlying evidence.
In some workflows, AI recommends an action; in others, it may execute a step within agreed limits. People need to understand when to accept, question or stop that output. This does not remove human responsibility. It changes the knowledge, authority and support needed to exercise it.
Consider a loan officer who has spent fifteen years developing expertise in credit assessment. They have intuitions about risk that are grounded in thousands of individual decisions. Now suppose an AI model produces a credit score that performs better on an agreed test measure. That result does not settle whether it is reliable for every customer or suitable for the live decision. What is their role? To rubber-stamp the model's output? To override it when their intuition disagrees? To monitor it for errors they may not have the technical knowledge to identify? None of these answers is satisfying. The discomfort they create is not only a training problem; it is also a question of professional identity.
Why Training Is Necessary but Not Sufficient
The standard response to the human challenge is training. Teach people how to use the new tools. Run workshops. Create e-learning modules. Certify competency. This is necessary work, and I am not dismissing it. But it addresses the surface of the problem while leaving the depth untouched.
Training teaches people what to do. It does not address what they fear. And in the context of AI, the fears are substantial and legitimate. Fear of obsolescence. Fear of losing the expertise that defines their professional identity. Fear of being held accountable for decisions made by a system they do not fully understand. Fear of looking incompetent in front of colleagues who seem to be adapting faster.
These fears are not irrational. They are a normal response to uncertainty about work, status and accountability, and a two-day workshop on prompt engineering will not resolve them.
The Middle Management Paradox
In my experience, middle managers are among the most important and most neglected groups in an AI transformation. Senior leadership sets the vision, and front-line workers use the tools. Middle managers sit between those groups, expected to drive adoption while processing their own uncertainty about what AI means for their roles.
Middle managers in many organisations derive their authority from two things: domain expertise and information asymmetry. They know things their teams do not. They have access to data their teams cannot see. AI erodes both of these. When a junior analyst can query an AI system and get insights that previously required years of experience to develop, the manager's expertise advantage shrinks. When dashboards and AI-generated reports make information transparent across the organisation, the information asymmetry disappears.
This does not mean middle managers become irrelevant. It means their role must evolve: from gatekeepers of information to interpreters of context, from decision-makers to people who assure decision quality, and from technical experts to people who can bridge the gap between what AI recommends and what the organisation should do. But this evolution does not happen automatically. It requires deliberate support, new frameworks for evaluating performance, and honest conversations about how roles are changing.
What Effective Change Management Looks Like
The organisations that manage the cognitive bottleneck effectively do several things differently. They start the change management work before the technology is deployed, not after. They involve the people who will be affected in the design of the solution, so that the AI system reflects their expertise rather than replacing it. They create safe spaces for people to express concerns without being labelled as resistant. They also redefine success metrics for the new operating model. People are assessed not only on the volume of decisions they make, but on the quality of oversight they provide.
Most importantly, they are honest. They do not pretend that AI will not change roles. They do not promise that everyone's job is safe. They acknowledge the uncertainty, provide support for navigating it, and show through practical decisions, not communication alone, that the organisation is willing to invest in its people as their work evolves.
In the UAE, where the pace of AI adoption is among the fastest in the world, this human dimension is particularly critical. Deloitte’s State of AI in the Middle East reports that over 80% of organisations in its study feel intense pressure to adopt AI. That pressure creates urgency. Urgency can create shortcuts, and one of the most common is to skip the human work and focus on the technology.
That shortcut can leave the organisation with working technology but weak adoption. The more useful question is whether people have the clarity, authority and support to work with it. And that question is not answered by the IT department. It is answered by the quality of the change management, the honesty of the leadership, and the willingness of the organisation to invest in its people with the same conviction it invests in its platforms.