Materiality calibration

Materiality Calibration

Materiality calibration is the explicit, documented threshold that separates routine AI use from consequential AI use that must receive governance attention. It defines which AI-supported decisions, uses, and incidents trigger:

• governance review,
• evidence production, and
• escalation or oversight,

and which can proceed under lighter-touch controls.

Without this calibration, an AI governance programme either:

• Over-governs: treating every prompt or interaction as a control event, exhausting the function and slowing delivery; or
• Under-governs: treating all activity as routine until a failure occurs, leaving no evidence that higher-risk decisions were treated differently.

The calibration itself is the artefact: a named, versioned statement of thresholds that can be inspected and audited.

Calibration Dimensions

A defensible programme calibrates materiality across at least four dimensions:

1. Decision consequence
   • Does the AI output affect a client matter, regulatory filing, HR decision, or external commitment?
   • If yes, materiality is higher because the downstream impact of error is greater.
2. Reversibility
   • Can the action be undone or corrected if the AI output is wrong?
   • Low-reversibility actions (e.g. filings, regulator communications, external legal or professional advice) sit higher on the materiality scale.
3. Audience exposure
   • Does the AI-generated or AI-shaped output leave the organisation?
   • Client-facing, court-facing, and regulator-facing outputs are higher materiality than internal drafting or exploratory analysis.
4. Tier of autonomy
   • Tier 1 – Augmentation: AI assists but humans drive and decide.
   • Tier 2 – Co-pilot: AI drafts or proposes, with systematic human review before action.
   • Tier 3 – Workflow operator: AI executes steps in a process with limited human intervention.
   • Tier 4 – Autonomous agent: AI plans and acts with high autonomy.
   • Tiers 3 and 4 are calibrated as higher materiality than Tiers 1 and 2.

Where Materiality Calibration Appears

Materiality calibration is embedded into core governance artefacts and processes:

• Risk Register entries
• AI Governance Charter (GOV-01)
  • named materiality tiers or levels,
  • the thresholds that trigger AI Council review,
  • when post-incident analysis is mandatory, and
  • which use cases must be covered in quarterly retrospectives.
• Use Case Intake (USE-01)
• Risk Register (GOV-02) lifecycle updates
  • changes in scope,
  • changes in autonomy tier,
  • new audiences or jurisdictions, and
  • observed incidents or near-misses.

Why Materiality Calibration Matters

1. Defensibility
2. Operability
3. Comparability over time
   • compare use cases across time,
   • re-evaluate older deployments against updated standards, and
   • demonstrate how its risk posture has evolved.

A programme that merely asserts that “everything important is reviewed” but lacks a documented materiality calibration is asserting a posture, not evidencing one. A robust AI governance function treats the calibration itself as a first-class artefact and keeps it under change control.