Practical AI · 3-minute read
The AI service is unavailable. Your customer is available. Enthusiastically. On line two.
An outage does not have to become a business emergency. It becomes much harder to manage when nobody remembers how the task worked before a chat window became the front door.
Choose one job that cannot simply wait
Granite Finch Construction is fictional. Its office uses an approved AI workflow to turn incoming delivery requests into draft scheduling entries. A person confirms the details before a delivery is arranged. This example concerns office coordination, not site safety or equipment operation.
The manager asks what happens if the service is unavailable for half a day. The useful question is whether staff can still find the request, see what has already been confirmed and record the next action.
The Canadian Centre for Cyber Security describes business continuity planning around critical functions, assets, roles and processes, and recommends testing and updating the plan. The same planning discipline is useful for an AI-dependent task. [1]
Write a fallback someone can actually use
Trigger: who decides the normal workflow is unavailable and activates the fallback?
Inputs: where are approved source records and current work status accessible?
Method: what template or manual steps keep the essential work moving?
Owner: who handles the task and who covers their absence?
Recovery: who reconciles temporary records with the restored system?
Keep essential instructions accessible through an approved route that does not depend on the failed service. Check permissions and whether a saved copy is current. A backup plan inside the unavailable chatbot is mostly a literary achievement.
Using this approved workflow description, draft a fallback for loss of [service or account]. List the trigger, accountable owner, essential inputs, manual steps, temporary record, customer communication and recovery checks. Identify shared dependencies that could disable both normal and backup methods. Mark missing details ‘Confirm.’ Do not invent credentials, access rights or technical recovery commands.
Rehearse the return to normal too
Run a tabletop exercise with fictional requests. Ask a trained colleague to explain how they would complete one request without the AI. Verify access and steps without deliberately disrupting production. Record where the instructions fail.
Before returning to the normal workflow, compare request IDs and status records. A failed screen does not prove that nothing was processed. Check what completed, what is queued and what needs attention before retrying, so customers do not receive duplicate confirmations.
Have the operational and technical owners verify the plan. Avoid moving confidential work to a personal account just because it still loads. Consider staff who need accessible formats and customers who cannot use the usual online channel.
Review the fallback when the workflow, supplier or account setup changes. The goal is a workable reduced service until normal operations return.
Your turn: Which AI-assisted task would your team struggle to complete if the tool disappeared tomorrow morning?
Source [1]: Canadian Centre for Cyber Security: Developing your business continuity plan. Supports continuity planning and exercises. The construction office and AI fallback checklist are original adaptations, not a tested recovery procedure for a specific platform.
