What should happen when an AI experiment has run its course?
Plan how an AI experiment ends before it starts. Give the trial an end review, with a named person who can extend it and a clear idea of what evidence would justify continuing. If it stops, give leftover tasks to a named person, have the administrator remove access and record what the experiment taught.
Teams usually plan the start of an experiment more carefully than the end. A tool nobody uses can still have connected access and outdated reference material, plus unfinished work.
I'd want a named person taking over every outstanding task. Check whether any external commitments or scheduled actions remain. Access comes off through the responsible administrator, and disuse won't switch it off. A paid subscription or an installed integration shouldn't be the only sign the work is still going.
Keep the records your organization requires and remove material according to its policy. Tell employees which process replaces the retired one and where the current instructions live.
A documented decision to stop can keep another team from repeating the same mistake. Ending a workflow responsibly is part of managing it, and it says nothing bad about having tested it.
What to try this week
Set an end review date for one AI trial that's still running. Record who can extend it and the evidence that would justify continuing, and note the access, scheduled actions and reference material someone must remove if it stops.