// Audit · Institutional decisions

What should an AI decision audit examine?

The complete decision produced around the AI—not only the model’s design, validation and performance.

An AI decision audit should establish what the institution intended, how that intent became data, rules and system logic, where judgement and authority entered, what was executed, what outcome followed and whether feedback could correct the chain.

// 01

The wrong unit of analysis

Model validation, bias testing, data lineage, access control and monitoring are necessary. But each examines a component. A consequential decision often crosses several models, rules, teams, systems and human hand-offs. The audit fails if no one examines that decision as a complete object.

Knight Capital, 2012: seven servers received the new software and one did not. The material failure was not merely old code; it was the broken join between an authorised deployment decision and the state actually executing in the market. The firm lost more than $460 million in about 45 minutes. See the worked replayability example.

// 02

Eight layers to examine

L1–L3

Purpose, strategy and intent: what outcome was authorised?

L4–L6

Rules, judgement and decision: how did authority become action?

L7–L8

Outcome and feedback: what happened, and could the chain correct itself?

// 03

The audit questions

Can one material decision be replayed end to end?
Which joins have no owner or evidentiary record?
Where did delegated authority exceed the encoded mandate?
Which divergence could become irreversible before detection?
What single remediation would restore the most control?

// Questions people ask

Common questions

Is this the same as model risk management?

No. Model risk management examines whether models are sound and operate within approved limits. A decision audit examines whether the complete institutional outcome remained connected to authorised intent.

Does an AI inventory provide enough evidence?

No. It identifies systems and use cases, not necessarily the specific decisions and joins through which authority was exercised.

What should the board receive?

A prioritised finding showing where chain integrity broke, the exposure created and the action required to close it.

// The practical test

Test one institutional decision

The ten-day Decision Drift Audit™ maps one material decision across all eight layers, assesses replayability and authority boundaries, and delivers one prioritised board finding.

Test one decision →Canonical definitions →Decision Integrity Chain™ →