An AI assistant shown in a demo: management convinced, business users unimpressed. It was never put in their hands.
Nothing was slapped together, though. Two months of clean work by the technical team, based on the documentation it had been given. The question it was asked: is this feasible. It answered correctly. The question that decides a rollout is a different one, what is this assistant for, for whom, at what point in the day and what does it change in the work of several hundred business users. No one had written it down.
What the POC proved, and what it could not prove
This POC is not a waste, and I say so at every debrief. It made the subject tangible for management and for the teams, which no presentation does. Showing the business what the technology can do surfaces usage ideas that a cold workshop never brings out. But a demo does not become a rollout on several hundred workstations through a version upgrade.
The technology has moved on by two generations since. The blocker hasn't moved, and it is still the one I meet in scoping this year.
Uses first, then the employee experience, the solution last
I take these subjects in the same order, and always have. The uses: what people do with their days, where it jams, what costs them time, what they look up three times a day in a procedure. The employee experience: where the assistant fits into the gesture, what it replaces, who checks what. The solution last, once we know what it has to hold up.
It starts with workshops with the people who will do the work, and with observing them at their desks. I come out with named uses, the questions they actually ask, a success criterion the business recognises as its own. That is also where adoption is settled: a team that has formulated the need itself doesn't then have to be convinced to use the tool.
The rest follows that scoping rather than preceding it. The quality of the documentation, for instance, is judged against what the selected uses demand of it, not in the absolute. Who updates what, at what rhythm, with which source of truth: those are process questions and they are settled with the business.
AI is a lever, it produces nothing on its own
AI is a technological lever. It produces nothing on its own: it needs a process that holds and people who agree to change the way they work. Without a stated need, a successful POC remains a fine object.
Did your last AI project start from a business use or from a technology to try out? If you have a POC that has plateaued, that is where I would pick it up. Write to me.
