Run an AI outage
tabletop exercise.
Pick one real workflow that depends on an AI tool. Invite the person who owns that workflow, someone who does the work, and someone who can approve restarting it.
Start with a concrete disruption
Read this aloud. “It’s 9 a.m. Your drafting assistant won’t load. Eighteen tickets need a decision before the noon cutoff. The inbox still works.” Ask the group what they would do in the first ten minutes. Which screen or record can everyone still trust?
Make the fallback a real handoff
Introduce a second event. “The afternoon shift can see the normal queue but has no record of manually assigned cases.” Ask someone to deliver the handoff aloud. Have the next lead read back what they think the priority is and name any cases that still need an owner.
Make someone own the restart
Introduce a recovery signal. “The vendor reports recovery, and three replies already sent by hand still have drafts in the AI queue.” Ask what needs checking in your own records, which cases to sample, and who can say the workflow is ready to resume.
End with changes you can test
Record which decisions were easy, which depended on information you lacked, and one change you can test this week. For example, ask a shift lead to show a new teammate the manual queue without help.