Reference / preflight checklist
Automation Preflight Checklist
Run this before choosing an AI tool, building a workflow or giving software permission to act. Automation should earn authority from the operating case.
Value case
- What recurring time, delay, error, missed opportunity or visibility problem is worth changing?
- Does the workflow happen often enough for automation to justify build and maintenance?
- What real evidence would show the automation created useful value?
Process definition
- Is the trigger explicit?
- Are the required inputs, normal states and expected output clear?
- Do competent people currently perform the normal workflow in roughly the same way?
Rules vs interpretation
- Which steps are deterministic rules?
- Which steps require language interpretation or variable context?
- Which decisions still require accountable human judgement?
Consequence + reversibility
- What happens when the system is wrong?
- Can the action be reversed or corrected quickly?
- Does the consequence justify a human approval gate before action?
Permissions + human gate
- Which tools, records and actions may the system access?
- Which actions are explicitly forbidden or require approval?
- Who owns the material decision when the system is uncertain?
Observability + escalation
- Can a person see what the workflow did and why?
- Are failures, duplicates, skipped actions and low-confidence cases visible?
- What condition makes the system stop, ask, defer or escalate?
Ownership after launch
- Who owns broken integrations, source-data quality and permission changes?
- Who reviews recurring exceptions and changes business rules?
- What evidence would justify expanding, narrowing or retiring the automation?
Passing the checklist does not create a guaranteed ROI case. Real volume, error rates, maintenance cost and failure behaviour still have to be observed. High-consequence judgement stays human-owned unless the evidence supports a different operating model.