Nanny Agency Jobs How-to / checklist

Plan a controlled manual fallback for candidate updates and records when an approved AI service is unavailable.

Name the essential work

An AI outage should not erase who owns an active candidate, which family communication is pending or when a promised update is due. List the small set of activities that must continue: candidate contact, confirmed stage tracking, time-sensitive appointments and correction of inaccurate records.

Define the minimum authorized record for each activity. Use the agency’s existing approved system and access controls. Avoid an unapproved spreadsheet export of candidate and household details.

Write the manual route before an outage

Specify who can view the source record, who verifies the current stage, who drafts a message without AI, who approves it and where the sending result is recorded. Keep a reusable plain-language template that says what is known and what remains unknown.

Fictional example: an agency’s drafting assistant is unavailable on Thursday. Recruiter A checks the approved case system, sees no family decision and writes the promised Friday update from a manual template. Recruiter B reviews the job reference and next-contact commitment; the normal messaging system records delivery status.

Test access and recovery

Practice a fictional outage with the AI tool disabled. Confirm that staff can reach authorized records, cover an absent owner and find the latest approved message template. If the normal case system itself is unavailable, use the agency’s approved continuity procedure and minimum necessary notes; do not create a permanent shadow database.

NIST’s AI RMF Core includes roles, incident processes and monitoring. Its framework is voluntary guidance, not proof that any particular fallback meets an agency’s obligations.

Reconcile when service returns

Compare manual actions with the primary record before re-enabling automation. Close duplicate tasks, enter any missed sending result and check whether queued drafts are now stale. Do not bulk-send everything the AI system prepared before the outage.

Assign an owner to review the incident: duration, affected promises, records unavailable and the one change that would make the fallback more reliable. Keep the outage record within the agency’s normal privacy and retention rules.

Sources