Executive sponsorship is not usually where AI adoption loses momentum. Neither is staff curiosity — people are already using AI tools, whether or not anyone has approved them. Ask where an AI rollout actually stalls and the answer is often the same layer: the managers sitting between the strategy and the work. It is almost never resistance. It is that they have been handed a change to deliver on top of the numbers they were already accountable for, with none of the room that change requires.

Why the squeeze lands there

The executive team gets the strategy days and the vendor briefings. Frontline staff get the tool and access. Middle managers get neither the airtime nor the slack. They are asked to hold output steady while their team learns a different way of working, to judge whether an AI-assisted piece of work is good enough — often in tasks they no longer do themselves — and to answer questions about job security they have never been briefed on. Faced with that, the rational move is to let the team experiment quietly, keep the process exactly as it is, and report that adoption is progressing. Which is precisely what the dashboard then shows.

What actually helps

• Move a target, visibly — if the volume and turnaround expectations do not shift during the transition, the manager has to protect the numbers, and the numbers always win.
• Train them to review, not to prompt — their job is judging AI-assisted work, so teach them what good looks like and, more usefully, where it tends to fail.
• Make the decision rights explicit — which tasks may be AI-assisted, what a person must check, and who signs off. Ambiguity quietly defaults to “don’t”.
• Give back the redesign time — a few protected hours a week to rework the steps is worth more than another round of tool training.
• Ask them what to automate — this layer can see the rework, the workarounds and the re-keying more clearly than anyone above or below it.

The pragmatic takeaway

The distinction that matters is between change done to a team and change done by it — and the manager is the person who decides which one it feels like. So pick one process, one manager and one quarter. Change that manager’s target, give them the authority to redesign the steps rather than just add a tool to them, and let the result be visible to their peers. Adoption does not stall because people dislike AI. It stalls where nobody was given permission to change how the work is done.