How We Saved a $2.5B Regional Mortgage Lender from FDIC Enforcement Action in 90 Days

When a $2.5 billion regional mortgage lender rolled out their proprietary AI credit scoring model, the early wins were clear: faster decisions, better risk filtering, happier originators. But then came the FDIC exam with a serious red flag.

Examiners identified major gaps: no formal SR 11-7 third-party validation ever completed, evidence of potential disparate impact on protected classes, limited explainability, and no ongoing monitoring program. They issued a 90-day deadline to produce comprehensive, defensible validation documentation or face enforcement action in the form of fines, reputational risk, and a potential freeze on AI-driven lending growth.

That’s the kind of high-pressure situation our team has navigated successfully before and exactly the kind of challenge we built RegVizion to solve for community and regional institutions today.

The Challenge: A High-Stakes Regulatory Deadline

The lender was already relying on the AI model for thousands of mortgage applications. The FDIC’s main concerns boiled down to:

• Complete absence of independent SR 11-7 validation
• An 8% lower approval rate for minority applicants with comparable credit profiles
• Inability to clearly explain model decisions to regulators or borrowers
• No structured process to detect model drift or performance degradation

With just 90 days to respond, the risk and compliance team needed a partner who could deliver fast, regulator-ready work without disrupting originations.

Our Solution: Fast, Focused, and Fully Compliant

Drawing on deep experience from similar high-stakes validations, we assembled the right expertise and moved quickly. Over 12 intensive weeks, we collaborated closely with the lender’s risk, credit, and technology teams in three focused phases.

Phase 1: Rapid Triage (Week 1)

We kicked off with an emergency assessment:

• Thorough model and documentation review
• Immediate regulatory risk prioritization
• Clear 90-day roadmap with buy-in from leadership

This gave everyone confidence that the response would be complete and timely.

Phase 2: Full SR 11-7 Validation (Weeks 2–6)

We executed a complete, independent validation aligned with SR 11-7 requirements:

• Conceptual soundness review of the model’s theory, assumptions, and mathematics
• Detailed data quality assessment, uncovering legacy biases in the training set
• Comprehensive fair lending and bias testing across protected classes
• Performance and stability analysis using holdout and out-of-time samples
• Sensitivity and stress testing under varied economic scenarios

Documentation was built in real time to create a connected audit trail.

Phase 3: Remediation & Lasting Governance (Weeks 7–12)

Findings turned into action:

• Model recalibration to close the identified bias gap while preserving accuracy
• Data enhancements to add new features and remove problematic variables
• Deployment of a clean, automated monitoring dashboard tracking 15 core metrics
• Establishment of quarterly governance reviews and escalation protocols
• Hands-on training so the internal team could sustain the process long-term

The Results: Compliance Secured + Tangible Business Gains

The full SR 11-7 validation report, over 140 pages of clear, defensible evidence, was delivered ahead of the FDIC deadline. The examiners accepted the submission without requiring further revisions. No penalties. No enforcement action. No model-related findings on the follow-up exam.

The upside went well beyond compliance:

• Loan approval rates increased 13% after bias remediation. This led to more qualified borrowers funded
• Manual underwriting reviews fell 15%, freeing capacity
• Approximately $200,000 in annual operational savings from greater efficiency
• Out-of-sample model accuracy improved 18% post-refinements

As the lender’s Chief Risk Officer reflected:
"The team didn’t just meet the FDIC deadline, they delivered a fairer, more accurate model and the framework to keep it that way. That level of partnership is exactly what we needed."

Key Takeaways from This Engagement

These kinds of intense, time-bound projects reinforce the lessons we bring to every RegVizion client:

1. Proactive validation prevents enforcement surprises — waiting for the exam letter is risky.
2. Fair lending testing is essential for any AI model in credit decisions.
3. Documentation is defense — SR 11-7 compliance means comprehensive, organized evidence.
4. Continuous monitoring is non-negotiable — models drift; early detection saves money and stress.
5. Specialized expertise accelerates results — combining ML knowledge with regulatory fluency gets you across the finish line cleanly.

We founded RegVizion to bring exactly this kind of focused, high-impact support to community and regional banks without the big-firm overhead.

Facing model validation pressure, bias concerns, or an upcoming exam? Schedule a consultation today and get ahead of the curve by turning compliance into a real advantage for your institution.

Related Services:

• Model Risk Management
• AI Governance