Model Cards: One-Page Governance Artifacts

ACRGA-EXPLAIN introduces "model cards"—one-page documentation for each AI credit model.

What's included:

• Intended use: "Predicts PD for CRE loans >$10M in primary markets"

• Training data: "50,000 loans, 2013-2023, 87% primary markets, 13% secondary"

• Known limitations: "Model trained during recovery period; performance in sustained downturn uncertain"

• Bias audit: "Shows 2% higher predicted loss rates for secondary markets vs. primary"

• Performance metrics: Validation results, backtesting performance, sensitivity analysis

This is required documentation. Not optional.

Model cards serve multiple audiences: credit officers (understand model scope), governance committees (identify limitations), regulators (understand design).

The discipline of writing a model card forces clarity about what the model actually does.

#RiskManagement #AIGovernance #DocumentationStandards