Published August 14, 2026 · Hoopi team
Demos prove capability. Nobody needed that proved.
A good demo answers a question the team was not really asking. Everybody already believes the technology can do impressive things; what they doubt is whether it will do their thing, on their messy data, reliably enough to stop doing it themselves.
So the pilot has to answer that question directly, on the first attempt, with a job whose output someone already depends on.
How to pick the first job
Look for four properties. It recurs on a predictable schedule. It crosses at least two systems. Somebody senior enough to be expensive is doing it. And there is an existing artifact — last month's version — that defines what good looks like.
The last one matters more than people expect. A task with a known-good previous output can be evaluated in minutes. A task without one turns into a debate about standards, and the pilot stalls there.
What to expect in week one
The first output will be about eighty percent right and wrong in a way that is easy to name. That is the good case, and it is what you want: specific, correctable errors rather than vague dissatisfaction.
The corrections made in week one are the actual deliverable of the pilot. They are what turns a capable general system into something that knows how your company does this particular thing — and they are what the next person inherits.
Want to try this on your own worst task?