QUESTION: Should AI Be Used for Credit Decisions?

So, what is the answer?  I believe it depends on the individual CFI, but in my opinion, AI should be considered by all CFIs as it is constantly improving, and “bleeding-edge” CFIs are already using it.  However, the need for strong internal governance, documented transparent processes and continuous oversight can make deploying AI a challenging engagement.

My opinion stems from early in my banking career, (almost 40 years ago), when I struggled with my inexperienced lending decision-making.  Though credit scores would have helped, an AI tool would have been a huge enabler to advance my abilities faster.  AI’s ability to expand access to credit information can improve decision accuracy and streamline underwriting, but its efficiencies also introduce new challenges.  The new requirements developed around compliance, fairness in lending verification, and operational risk identification create an even heightened responsibility for CFIs.

Regulators’ Messaging Is Consistent: AI Is OK, But With Full Compliance Requirements

The CFPB, FDIC, OCC, and Federal Reserve have all reiterated that there is no “fancy technology exemption” for fair lending, consumer protection, or model governance requirements. Whether a credit decision is made by a loan officer, a traditional scorecard, or a sophisticated machine‑learning model, the same rules apply.

This means CFIs must ensure their AI‑driven lending decisions comply fully with:

• ECOA (Equal Credit Opportunity Act)
• Regulation B
• The Fair Housing Act
• UDAAP standards
• Model risk management expectations

Regulators are not discouraging the use of AI as they acknowledge its potential benefits. But they expect CFIs to demonstrate their AI models are explainable, fair, well‑documented, and continuously monitored.

Requirements To Consider Before Deploying An AI Tool:

1.      Adverse Action Notices Must Be Specific and Accurate

One of the clearest regulatory expectations concerns adverse action notices. Institutions must provide specific, understandable reasons for a credit denial. A vague explanation such as “the algorithm determined you were ineligible” is not acceptable.

No matter what complex variables, models or interactions their AI tool employs, a CFI must be able to translate their lending decision into a consumer‑friendly explanation. This requirement alone has eliminated many “black box” models from consideration.

CFIs should partner with vendors who can provide:

• Transparent model logic
• Clear factor‑level explanations
• Documentation that supports compliant adverse action notices

If the you can’t explain the model’s decision, you shouldn’t use the model.

2.      Disparate Impact Testing Is Mandatory

Any AI model could amplify a bias if not carefully designed and monitored. Regulators expect CFIs to continuously conduct thorough impact testing to ensure protected classes are not disproportionately harmed by automated outcomes.

This would include:

• Pre‑deployment fairness testing
• Ongoing monitoring for drift
• Comparative analysis against traditional models
• Documentation of testing methodologies and results

CFIs must take corrective action on any identified disparities. Ignoring or failing to detect disparate impact would be a significant compliance risk.

3.      Institutions Must Document Less Discriminatory Alternatives (LDAs)

An often-overlooked requirement is the need to evaluate and document less discriminatory alternatives. When an AI model produces disparities, the CFI must determine whether another model or methodology could achieve similar predictive performance with less discriminatory effect.

This expectation applies to:

• Vendor‑provided models
• Internally developed models
• Hybrid human‑AI underwriting processes

Simply put, AI adoption is not a “plug‑and‑play” exercise, but requires thoughtful evaluation, documentation, and governance.

4.      AI Is Allowed, But It Requires Higher Scrutiny Than Traditional Models

The implication of all this guidance is straightforward: AI is allowed, but it must be treated like any other credit model, but with even higher scrutiny.

CFIs must ensure:

• Model transparency
• Explainability
• Bias testing and mitigation
• Strong data governance
• Ongoing monitoring and recalibration
• Clear documentation for examiners

CFIs should not deploy AI in credit decisioning until they build the necessary governance capabilities to meet these expectations.

Why AI Still Matters for CFIs

Even though AI introduces more regulatory complexity, (like we want more of that), it does offer some meaningful advantages:

• Faster credit decisions or improved customer service
• More consistent underwriting enabling younger lenders to be more productive
• Expanded access to credit for underserved borrowers
• Improved risk modeling for better consistency
• Reduced manual workload for lending teams as employment challenges continue to increase

When implemented responsibly, AI can strengthen the mission of community finance: fair, accessible, relationship‑driven lending.

The Bottom Line

As I said above, I believe AI should be considered by CFIs as a tool to aid credit decisions but does not replace the “human factor” completely.  When implemented responsibly, AI can strengthen the mission of all CFIs: fair, accessible, relationship‑driven lending.

I also believe AI must be thoughtfully deployed with rigorous controls, transparent processes, and continuous monitoring.  It is not a “plug’n play, out of the box” solution. Though regulators have opened the door for AI‑driven lending, they expect CFIs to walk through it carefully, and with full accountability.

I think CFIs who are willing to invest in strong governance, AI can be a powerful tool to enhance fairness, expand access, and improve the member experience. For those not quite ready, the priority should be building the frameworks that make future AI adoption possible.

Atlas Advisory Partners does not consult on AI policies or lending decision-making regulations but does want to help CFIs continue to thrive.  We do specialize in assisting with vendor contract negotiations and in non-interest revenue enhancement engagements.  Our approach is NOT a “one size fits all”, but a curated, proven approach where we deliver a specialized strategy for each FI.  Contact us to discuss how a tailored engagement could support your institution.