Set the oversight level
Ten AI use cases. Read each one, then decide how much oversight it needs. Drag it into a tier, or use the dropdown on the card.
Tier definitions
Tier 1 · Audit-only
Lower riskAI acts autonomously. A human reviews outputs after the fact, through sampling and the audit trail.
Assign here: low-risk, low-impact, easily reversible use cases: routine drafting, internal search, low-stakes summarisation.
Tier 2 · Human-on-the-loop
Moderate riskAI acts autonomously in real time. A human monitors and can intervene or override before harm lands.
Assign here: moderate-risk use cases with real-time monitoring in place: fraud flags, operational recommendations, escalation paths.
Tier 3 · Human-in-the-loop
Higher riskA human reviews and approves each material decision before it takes effect.
Assign here: high-risk, high-impact, hard-to-reverse decisions with legal or substantial effect: credit, employment, insurance.
Tier 1 · Audit-only
Lower riskAI acts autonomously. A human reviews outputs after the fact, through sampling and the audit trail.
Assign here: low-risk, low-impact, easily reversible use cases: routine drafting, internal search, low-stakes summarisation.
Tier 2 · Human-on-the-loop
Moderate riskAI acts autonomously in real time. A human monitors and can intervene or override before harm lands.
Assign here: moderate-risk use cases with real-time monitoring in place: fraud flags, operational recommendations, escalation paths.
Tier 3 · Human-in-the-loop
Higher riskA human reviews and approves each material decision before it takes effect.
Assign here: high-risk, high-impact, hard-to-reverse decisions with legal or substantial effect: credit, employment, insurance.