The standard playbook for rolling out a new AI tool is to find the enthusiastic early adopters, get a visible win, and use that momentum to bring everyone else along. It works, up to a point, and then it often stalls exactly where it matters most.
The people who were going to be skeptical stay skeptical, because nobody addressed their actual objections; they just watched a demo from people who were already convinced. When the rollout reaches their team, the skepticism has not gone anywhere, and now there is added resentment about being the last ones brought in.
A more durable approach involves the skeptics early, not to convert them in a single session, but to genuinely hear what they think will break. Operations staff who have seen automated systems fail before usually have specific, valid concerns: edge cases the tool will not handle, workflows it was not designed for, situations where a wrong output looks confident enough to go unquestioned.
Those concerns, taken seriously, usually improve the rollout. They surface failure modes before they cost anything, and they give skeptical staff a genuine stake in getting the implementation right, rather than a reason to quietly work around the tool once it launches.
AI adoption that only listens to enthusiasm tends to look successful in the first quarter and fragile by the second. The rollouts that hold up are usually the ones that made room for the people who said "wait, what about this" before it became a problem.