The pilot that never became a process

Somewhere in your company there is probably a thing that worked. Somebody tried an AI tool on a real task last spring, it produced good results, everyone agreed it was useful, and it is not part of how anything is done today.

This was the main finding reported out of Davos last month, at a scale of billions. At your scale it costs an afternoon and some enthusiasm, but the mechanism is the same.

Why pilots stall, in order of frequency

It stayed with one person. The trial worked because a capable colleague did it carefully. Nobody else knows the instruction, so when they are away the old way returns and then simply stays.

It was never put where the work is. The pilot happened in a browser tab alongside the real process rather than inside it, and the one-click-too-far problem from 14 August closed it.

Nobody said it was now the way. A pilot ends with « that worked well », which is praise, not an instruction. Without somebody saying from now on we do it this way, a successful trial is just a pleasant memory.

The conversion, which takes an hour

Three things, and they are the same three every time.

Write the instruction down so a second person can follow it — the 23 July prompt, the 10 February paragraph. Name who does it and when, in the existing process description rather than beside it. And say plainly, once, that this is now how the task is done.

The test of whether it took

Ask, six weeks later, someone who was not involved in the pilot to do the task. If they do it the new way without checking, it is a process. If they ask how, it is still a pilot.

What to do about the ones that already stalled

Pick one, this month. Not the most impressive — the one whose result people still occasionally mention.

Converting an existing successful pilot is the cheapest value available to most Swiss SMEs right now, because the hard part, finding something that works, is already done.