AI adoption
Give an AI workflow a fallback before it becomes routine.
Define how work continues when the AI tool, source information or confidence needed for the task is unavailable.
Consider a hypothetical operations team using AI to turn incoming notes into a structured daily briefing. The first few trials work well. Then one morning the source file is incomplete, the tool cannot be reached and the briefing still needs to go out. A useful AI workflow needs a clear route for the moment when the normal path cannot be trusted or completed.
Describe the normal path first
Write the workflow as a short sequence: what starts it, which information is used, what the AI produces, who checks it and what action follows. Keep the description close to the real job. If the process depends on a particular folder, system permission or person, make that dependency visible.
Define the minimum acceptable output as well as the ideal one. The team may need a concise briefing by 9am; a polished layout may be useful but not essential. This distinction helps people protect the business outcome when part of the workflow is unavailable.
Choose the failures worth planning for
Do not attempt to predict every technical problem. Start with a few conditions that would prevent a safe or useful result: the required source is missing, the tool is unavailable, the output contains unsupported statements, or the reviewer cannot complete the check in time.
For each condition, state how the team will recognise it. A vague instruction to use judgement leaves people deciding under pressure. A simple trigger such as “two required sources are missing” or “the reviewer cannot confirm the figures” makes the handover clearer.
- Stop: what prevents the normal AI-assisted route continuing?
- Switch: which approved manual or reduced process should be used?
- Own: who decides to switch and who completes the work?
- Recover: what must be checked before the normal route resumes?
Make the fallback genuinely usable
Keep the fallback proportionate. It might be a short manual template, an existing report or a smaller output created from confirmed information. Make sure the people expected to use it still have access and understand the steps. A process that exists only in a launch document will not provide much continuity.
Avoid copying sensitive information into a second system merely to create a backup. Use approved locations and retain only what the underlying task requires. If the fallback changes who can see or approve the work, check that access and responsibility remain appropriate.
Rehearse the switch and learn from it
Run one labelled exercise with synthetic information. Pause the normal route at a chosen point, use the fallback and check whether the required outcome can still be delivered. Record unclear steps, unavailable access and decisions that took too long.
When a real fallback is used, capture the reason and the practical impact without turning the record into a large incident process. Repeated switches may indicate an unstable input, an unsuitable use case or a need to redesign the workflow.
Rethink’s AI working sessions can help teams map a workflow, define its controls and make the fallback usable. Ongoing adoption support can then keep responsibilities, checks and working practices current as the process changes.