The hype cycle rewards announcements. Delivery rewards adoption. The gap between those two things is where most enterprise AI investment quietly disappears.
On a recent enterprise rollout, the entire story was the difference between "deployed" and "adopted". Deployment is a press release; adoption is a behaviour change. Measured properly, adoption reached well into the nineties, with thousands of hours returned to the people doing the work — but that only happened because it was run as a change programme, not a software install.
Three things made the difference. We measured the baseline first, so the value was provable rather than asserted. We made that value visible to the people on the floor, not just the steering committee. And we treated resistance as information, not obstruction.
For regulated industries there's a further discipline. The governance and data-sovereignty questions have to be answered before the tool runs, not retrofitted afterwards. In financial services especially, "where does the data live and who can see it?" is the first question, not the last.
AI doesn't remove the need for delivery discipline. If anything, it raises the bar because the technology moves faster than the organisation's ability to absorb it, and closing that gap is delivery work.