Markdown

Human-In-The-Loop Approval Matrix

Use this when deciding what humans approve, sample, audit, or own in an AI workflow.

Workflow Facts

Consequence Matrix

QuestionYes / NoNotes
Can a wrong output affect a customer, employee, supplier, or regulator?
Can the workflow send, publish, delete, charge, refund, or change records?
Can it expose personal, confidential, financial, legal, HR, or security data?
Would the mistake be hard to detect later?
Would the mistake harm trust even if reversible?

Review Pattern

ConsequenceDefault patternSelected?
Internal, reversible, low visibilityAudit after the fact
Repeated, quality-sensitive workSample review
Routine flow with ambiguous casesException review
Customer-visible or external actionHuman approval before action
Destructive, financial, legal, HR, or regulated decisionHuman owns final decision

Review Screen Requirements

Stop Rules