AI Fair Lending Comparative File Analysis Playbook

Your compliance team is preparing for a CFPB fair lending examination focused on your mortgage pricing discretion policy. Your HMDA data shows a 47-basis-point unexplained pricing disparity for Black borrowers in three MSAs after controlling for LTV, DTI, and credit score. You have 200 matched-pair loan files selected by the examiner, your pricing exception log, and the underwriting discretion policy.

When to use this playbook

• Use this playbook when the decision looks like the situation above: Your compliance team is preparing for a CFPB fair lending examination focused on your mortgage pricing discretion policy.
• It is a fit when you have source files in hand and need a structured, reviewable analysis — not a generic chat answer about "Fair Lending Comparative File Analysis".
• Do not use it as a substitute for licensed, legal, clinical, or authorized official judgment in the domain.

What you'll need

• 200 matched-pair loan files (100 control, 100 comparison) - Pricing exception log (12-month period) - Underwriting discretion policy and guidelines - HMDA LAR data with regression residuals

Attachments: Multiple attachments (Spreadsheets, Documents)

You are a fair lending analyst and regulatory counsel preparing a defense analysis for a CFPB comparative file review.  I am attaching: - 200 matched-pair loan files (100 control, 100 comparison) - Pricing exception log (12-month period) - Underwriting discretion policy and guidelines - HMDA LAR data with regression residuals

Work only from the attached source files. If a conclusion is not supported, say so.

Produce:
1. Conduct a matched-pair analysis across all 200 files — for each comparison pair, identify differences in pricing, exception usage, and underwriting notes that are not explained by the matched credit risk variables.
2. Analyze the pricing exception log for disparate usage patterns — calculate exception frequency, approval rates, and average pricing benefit by race/ethnicity and loan officer.
3. Review the underwriting discretion policy for provisions that grant pricing flexibility without objective criteria — identify specific policy language that creates fair lending exposure.
4. Assess whether the 47-basis-point disparity in the HMDA regression is attributable to pricing exceptions, underwriting overlays, or product steering — and identify the largest contributing factor.
5. Produce a management response framework for the CFPB examination — identifying each finding category, the remediation already in place, and the monitoring controls implemented.

Call out where independent models are likely to disagree, and list follow-up documents a reviewer should request.

What to expect

• Multi-model consensus on disparate treatment risk classification
• Matched-pair pricing variance analysis with unexplained disparity flags
• Exception log disparate usage scoring by loan officer
• Policy language red flag register with ECOA/FHA citation map
• Draft management response framework with model-agreement score

Review before you act

• Validate this output against source files before relying on it: Conduct a matched-pair analysis across all 200 files — for each comparison pair, identify differences in pricing, exception usage, and underwriting notes that are not explained by the matched credit risk variables.
• Validate this output against source files before relying on it: Analyze the pricing exception log for disparate usage patterns — calculate exception frequency, approval rates, and average pricing benefit by race/ethnicity and loan officer.
• Validate this output against source files before relying on it: Review the underwriting discretion policy for provisions that grant pricing flexibility without objective criteria — identify specific policy language that creates fair lending exposure.
• Validate this output against source files before relying on it: Assess whether the 47-basis-point disparity in the HMDA regression is attributable to pricing exceptions, underwriting overlays, or product steering — and identify the largest contributing factor.
• Confirm every cited figure, date, counterparty, or requirement against the attached originals — models compress and can drop a qualifier.
• Treat disagreement between models as a review item, especially on classification, materiality, and recommended next action.
• Do not authorize an operational, clinical, legal, credit, or enforcement action solely because the models agree.