Fair Lending Compliance and HMDA: Evidence Over Rule Summaries

Why Fair Lending Is an Evidence Problem

In our recent Ask Kaia report, fair lending, HMDA, CRA, and appraisal-bias questions appeared in 292 tagged questions during the 90-day period, about 6.1% of substantive questions, and that share rose in June.

Those questions are valuable because they show where bankers need more than a rule summary. They needed help:

• Assessing evidence
• Documenting reasoned decisions
• Connecting the answer to data, loan files, and policy language

What Counts as "the Record"

When examiners and litigators look at fair lending, they look at a record. That record is built from four things:

• Quantitative data such as the HMDA loan application register
• Individual loan files, including appraisals and adverse action notices
• Policy language that defines how decisions should be made
• Contemporaneous documentation showing the stated process was followed

The Fair Lending Laws Are the Floor

Fair lending compliance in the United States rests on a small set of anchoring laws:

• The Equal Credit Opportunity Act and its implementing Regulation B
• The Fair Housing Act
• The Home Mortgage Disclosure Act with its implementing Regulation C
• State-law overlays add further requirements

The work happens in matching those standards to evidence. Fair lending risk is generally analyzed through three theories, and each theory calls for different proof.

The table below maps those theories of lending discrimination to what typically triggers scrutiny and what evidence a bank relies on to substantiate or defend a decision.

HMDA Data: The Evidentiary Backbone

For mortgage lending, HMDA data is the primary quantitative evidence set behind fair lending analysis. The loan application register captures the structured record of what an institution did with each application, and it is the dataset regulators and the institution itself use to look for patterns.

Regulation C requires reporting institutions to record specific fields, and several of them carry direct fair lending weight, including:

• Action taken on the application
• The reasons for denial where applicable
• Applicant demographic information
• Pricing information such as rate spread

Read together, these fields let an analyst compare outcomes across groups and flag disparities that warrant a closer look.

How HMDA data feeds disparate impact analysis

If HMDA data shows a statistically meaningful gap in denial rates or pricing for a protected class, that is a signal to investigate. It is a prompt to examine files and policies for a legitimate explanation.

This is where data quality becomes an evidence problem. When action-taken codes are wrong, demographic fields are incomplete, or denial reasons do not match the file, the register cannot support a defense.

Accurate HMDA reporting is the foundation the entire fair lending record stands on. In the Ask Kaia analysis, HMDA reporting and LAR decisions accounted for 108 tagged questions, reflecting steady practitioner attention to getting the register right.

Appraisal and Valuation Bias

Appraisal bias has become one of the more active fair lending concerns, and it lives inside the individual loan file rather than in aggregate data. The Ask Kaia report analysis found appraisal-bias language appearing repeatedly, tagged in 93 questions over the period.

Bankers were asking a practical question, “how do you review an appraisal or property evaluation for potential bias or concerning language?”

The review looks for two kinds of red flags:

• Language: references to the racial or ethnic composition of a neighborhood, to "desirable" or "undesirable" areas, or to demographics that have no legitimate bearing on value.
• Value: an estimate that is hard to reconcile with comparable sales or with a second valuation.

When either surfaces, the reconsideration-of-value process gives the applicant a route to challenge the appraisal. What makes this a compliance discipline is documentation. A reviewer who spots a concern, escalates it, and records both the concern and the resolution has built evidence.

Adverse Action Notices as Fair Lending Evidence

When an institution denies an application, the adverse action notice it sends is both a disclosure and a piece of evidence. Under the Equal Credit Opportunity Act and Regulation B, a creditor must give applicants a statement of specific reasons for adverse action. That notice becomes part of what an examiner tests.

The evidentiary risk is a mismatch between the stated reason and the file. If the notice says "insufficient income" but the underwriting record points to collateral, the inconsistency undermines the decision regardless of whether the underlying call was correct.

In the Ask Kaia data, adverse action and denial notices drew 95 tagged questions, and how to handle denial reasons, incomplete applications, and multiple applicants was a recurring theme. Common gaps include:

• Vague or generic reasons
• Reasons that do not appear in the file
• Notices that are technically sent but not retained in a way that proves what the applicant received

How AI Supports Fair Lending Compliance: Evidence Over Rule Lookup

Because fair lending is evidence work, it is a fit for source-backed AI assistance. The analysis concluded that fair lending AI use cases should be evidence-first and should:

• Define review criteria
• Retain source evidence
• Protect sensitive data
• Route outputs through qualified fair lending or compliance reviewers

A structured appraisal review can prompt the reviewer for the property, the comparable sales, and the specific language flagged. A lending scenario triage can walk a user through the facts needed for a Regulation B, HMDA, or adverse action question and note where a matter should be escalated.

Platforms like Ask Kaia implement this pattern by tying answers to cited source material and keeping an audit trail that records the source, the prompt, the answer, the reviewer, and the final decision.

Frequently Asked Questions