Financial & Corporate Compliance November 20, 2026

HMDA & CRA data validation: A defensible approach to full-population review

Key Takeaways

  • Regulatory data integrity is an enterprise governance issue. Recurring HMDA and CRA exceptions often reveal weaknesses in ownership, system mapping, and controls.
  • Full-population review improves management visibility. It shows where risk is concentrated across products, fields, systems, and processes rather than relying on the results of a limited sample.
  • Automation must reinforce accountability. AI-enabled validation becomes defensible when it is supported by traceable lineage, documented rules, human judgment, and evidence of resolution.
  • The greater return comes from preventing recurrence. By reducing manual comparison work, compliance teams can focus on root causes and build controls that also support future Section 1071 obligations.

Few compliance problems are as well documented, as persistent, and as ready for automation as regulatory data integrity. Many banks still validate Regulation C (HMDA) data through manual file review and sampling, and they are looking for AI-enabled tools that improve accuracy, speed, and insight without a heavy technology lift. This article looks at why the problem persists, how AI-enabled tools are making 100% transaction review practical, the governance structure that makes automation defensible, and how USRC’s Data Integrity Services, powered by Digital Twin, brings the two together for our customers.

A known problem area, year after year

The Federal Reserve’s Consumer Compliance Outlook publishes the most-cited consumer violations at state member banks. Regulation C (HMDA) data collection and reporting has ranked first in every year from 2022 through 2025, with Regulation BB (CRA) data collection second each year.

Year Reg C (HMDA) share of all violations Reg BB (CRA) share of all violations
2022 59.4% 7.2%
2023 48.3% 4.9%
2024 38.2% 8.0%
2025 45.2% (285 violations) 6.0%

Section 1071 raises the stakes further. The CFPB’s final rule, issued May 1, 2026, narrows the original scope and sets a January 1, 2028 compliance date, but it remains a data-heavy rule that many banks are already planning for.

The rule adds another data set that must be collected, validated, and reported, with its own field definitions, data sources, and ownership questions. Banks do not need to wait for the compliance date to prepare. A well-thought-out approach that embeds AI for optimization early in the design of the process will bring efficiencies to this new data collection from the start, rather than layering automation onto a manual process after problems appear. Building in automated validation, exception flagging, and traceability as the process is designed and tested also lets the same governance structure described later in this article ownership, lineage, and reporting cover 1071 alongside HMDA and CRA. Institutions that build disciplined data validation now will be better prepared for all three.

Why manual HMDA sampling can miss systemic errors

HMDA includes 48 reportable fields. Their subcategories bring the total to 110 data fields that must be reported accurately on every reportable loan, every cycle. Manual review and sampling struggle at that scale: reviewers interpret rules differently, sampling can leave systemic errors undiscovered for cycles, and the same problems recur because root causes are never traced.

Sampling was a practical answer to a resource problem: reviewing every loan file by hand was simply too slow and costly. But a sample only describes the loans that were pulled. A bank that validates a subset cannot know what the rest of the population holds, and systemic errors, such as a field mapped incorrectly for one product or a rule applied inconsistently by one process, can sit undetected across many transactions. By the time they surface, they have often repeated through multiple cycles. Recurring errors are exactly what sampling is poor at exposing.

Why 100% review is within reach today

What has changed is the cost of looking. AI tools can now read loan documents, extract data values, and compare them with what is reported on the LAR and complete that exercise across every transaction and every product in a lending suite. That is the shift USRC’s Data Integrity Services, powered by Digital Twin, was built to support. Efficiency gains of this kind do not have to mean a large technology investment that strains budgets or balance sheets. Banks can adopt tools like this in measured steps: start by focusing on higher-risk fields or known problem areas, then build toward full-scale field review as confidence and capacity grow. Banks no longer have to choose between coverage and speed.

How full-population HMDA & CRA data reviews work

Reviewing every transaction changes what a bank can see. Instead of a handful of isolated errors, it sees patterns: which fields, products, data sources, or process steps produce the most exceptions, and how large the problem really is. That visibility arrives quickly, so issues can be addressed while they are small, before they spread across more loans and more reporting cycles. The insight is often more valuable than the error count itself.

But automation alone is not the answer. Reviewing every transaction faster only finds the same errors faster if no one owns the data, the data lineage is unclear, and no one is watching the results. The tool needs a structure around it.

A three-part framework for defensible regulatory data validation

Three elements make up that structure. All three must be present, and they must work together.

  • Ownership and accountability. Clear ownership of the data: who is accountable for each field and how reporting decisions are documented so they are applied consistently.
  • Taxonomy and data lineage. A taxonomy-based approach that ensures lineage follows the data on its journey from loan origination system (LOS) input to the Loan Application Register (LAR), so every reported value can be traced to its source.
  • Dashboards and reporting. Dashboards and reporting that support ongoing monitoring and testing and build an evidence file that supports scrutiny from internal or external examination.

Remove any one and the others weaken. Without ownership, no one is accountable; without lineage, reported values cannot be traced to their source; and without dashboards and reporting, none of it can be demonstrated to an examiner. In practice, most HMDA exceptions are not really data problems. They are role-clarity and mapping-consistency problems, and a taxonomy-based approach is where those root causes can finally be fixed.

Data Integrity Services, powered by Digital Twin

USRC’s Data Integrity Services, powered by Digital Twin, was co-created by USRC and Wolters Kluwer’s Data & Analytics team to put this approach into practice for HMDA and CRA data. It reviews 100% of transactions rather than a sample, flags exceptions in real time, and traces each reported value back to its source. Two layers of validation, the platform and an expert reviewer, work together to give customers and their examiners lineage back to where an error began, along with real-time visibility into problem areas. Used within the framework, it strengthens each part rather than replacing it.

Framework element How data integrity services supports it
Ownership and accountability Applies the documented source and rule for each field the same way on every record, every cycle
Taxonomy and data lineage Acts as a detective control, reviewing every transaction rather than a sample and flagging exceptions as they occur, with an expert reviewer resolving them, while tracing each value from LOS input to the LAR
Dashboards and reporting Shows where and why errors occur, informing new and stronger preventive controls, and builds an evidence file, giving compliance leaders, executives, and examiners evidence rather than assertions

The AI is used for the work machines do best. The platform reads loan documents with OCR, classifies them and extracts reportable values using trained machine learning, and compares those values with the reported data using configured rules. Judgment stays with people. The current workflow does not use large language models, generative AI, or autonomous agents, and the technology does not make regulatory determinations or change reported data. Customers provide their loan documents through an SFTP or API connection. The Data Integrity Services platform, combined with human-in-the-loop review by USRC, delivers a two-level review. The institution retains responsibility for reporting decisions and regulatory interpretation.

From manual effort to root cause

Deployed within this structure, Data Integrity Services removes the heavy manual effort of comparing loan documents to reported values, field by field, across every transaction. It standardizes the structure so the same logic applies to every record and every product. Most importantly, it frees compliance resources to work on root causes, such as unclear data ownership, inconsistent mapping, and training gaps, instead of hunting for individual errors. Data validation is itself a detective control that catches what preventive controls miss, and the patterns surfaced by the dashboard and the 100% review show where those preventive controls need to be built or strengthened. Fixing causes rather than symptoms is what reduces repeat findings, and it is what gives a bank greater confidence in its data governance approach.

Banks that pair automation with clear ownership, traceable data lineage, and active monitoring will find errors faster and will also be able to show that their regulatory data, and the process behind it, can withstand scrutiny. That is what it means to be defensible by design. To learn how Data Integrity Services can support your customers, connect with the USRC team.

{Sources: Federal Reserve, Consumer Compliance Outlook, Table 1, “Top Consumer Violations for State Member Banks,” 2022 (Second-Third Issue 2023), 2023 (First Issue 2024), 2024 (First Issue 2025), 2025 (Second Issue 2026); CFPB, Small Business Lending Under the Equal Credit Opportunity Act" (Regulation B), final rule, Federal Register, May 1, 2026.}

Frequently asked questions

  • What is HMDA data validation?
    HMDA data validation is the process of comparing reportable Loan Application Register data with underlying loan records, source systems, and applicable reporting rules. The objective is to identify inaccurate or inconsistent values, understand why exceptions occurred, and document how they were reviewed and resolved.
  • Why can sample-based HMDA reviews miss systemic errors?
    A sample describes the records selected for review, not necessarily the full population. If an error is concentrated in a particular product, field, source system or workflow outside the sample, it may remain undetected. Population-level comparison can expose recurring patterns, while expert review determines their significance and appropriate response.
  • What is full-population HMDA transaction review?
    Full-population review evaluates every transaction within a defined review scope instead of selecting only a subset. In an automated workflow, source documents and reported values can be compared using consistent rules, with potential exceptions routed to reviewers. The scope, timing, and fields reviewed should be documented clearly.
  • How can AI support document-to-LAR comparison?
    AI can classify loan documents, use OCR and trained machine-learning models to extract relevant values, and compare those values with corresponding LAR data through configured rules. This reduces repetitive manual comparison and helps reviewers focus on exceptions, root causes, and corrective controls.
  • Does automated HMDA validation replace human review?
    No. Automation can expand coverage and apply comparison logic consistently, but people remain responsible for evaluating exceptions, interpreting regulatory requirements, and making reporting decisions. Human oversight is especially important when source documents are incomplete, values are ambiguous, or an issue requires regulatory judgment.
  • What controls make AI-enabled data validation defensible?
    Key controls include named owners for reportable fields, approved definitions and comparison rules, traceable data lineage, secure access and transfer, documented exception resolution, monitoring dashboards, and periodic testing of the workflow. Together, these controls show how the process operates and how identified issues are addressed.
  • How does data lineage support HMDA and CRA reporting?
    Data lineage shows how a reported value moves from its original source through systems, mappings, and transformations to the final regulatory record. When an exception occurs, lineage helps teams identify where it originated, assign accountability, and determine whether the issue affects other records or reporting cycles.
  • Can the same governance approach support Section 1071 readiness?
    Yes. Although Section 1071 has its own scope, definitions, and reporting requirements, institutions can apply the same core disciplines: assign field ownership, document source-to-report lineage, establish validation rules, monitor exceptions, and retain evidence of review. Specific implementation decisions should reflect the final rule and current CFPB guidance.
Mark Miller
Director, US Regulatory Consulting, Data Governance
Mark is responsible for leading Data Governance and AI Enablement initiatives that help financial institutions prepare, manage, validate, and operationalize regulatory data through Data Management and Data Integrity solutions.
Back To Top