Commonwealth Credit Union

Commonwealth Credit Union, a $1.36 billion asset institution serving more than 107,000 members in Kentucky, faced a structural limitation in its auto loan underwriting: legacy credit scoring models were poorly calibrated for mid-tier borrowers. Applicants who fell outside clean approval or denial thresholds were routinely declined or underwritten conservatively, even when their actual risk profile justified approval. This left profitable members underserved and constrained portfolio growth. Without a more granular way to assess risk across the creditworthiness spectrum, the credit union had no reliable mechanism to expand lending responsibly — growth meant accepting more risk, not finding more qualified borrowers already in the member base.

Commonwealth Credit Union partnered with Zest AI to deploy machine learning-based underwriting models specifically for its auto loan portfolio. Zest AI's platform replaces or augments traditional scorecards with models that evaluate a broader set of borrower signals, enabling more precise risk differentiation — particularly among applicants in the middle credit tier where legacy models produce the most decisioning noise. The Zest AI system integrated into Commonwealth's existing loan origination workflow, allowing automated instant-approval decisions to flow through without manual review queues. Jaynel Christensen, VP of Lending, led the implementation, bringing her two-decade lending background to bear on calibrating the models to Commonwealth's member profile and risk appetite. The result was a consistent, explainable decisioning engine that expanded the approval band without relaxing underwriting standards.

The AI underwriting deployment produced measurable gains across approval rate, yield, and speed — with risk held constant throughout:

• +15% loan approval rate: More mid-tier applicants approved without increasing portfolio risk
• 5x increase in instant approvals: Automated decisioning reduced manual review volume and accelerated member experience
• +30 basis points portfolio yield: Better risk differentiation identified more creditworthy borrowers previously declined or underpriced

Beyond the metrics, the partnership validated that AI-driven underwriting could simultaneously serve growth, efficiency, and member experience objectives — goals that had historically required trade-offs under legacy models.

• Mid-tier borrowers are systematically underserved by traditional credit scorecards — ML models that evaluate more signals can unlock approvals without adding risk.
• Instant approval rate is a practical proxy for model confidence; a 5x improvement signals meaningfully better risk stratification, not just speed.
• Portfolio yield gains (30 bps here) reflect better pricing alignment, not looser standards — a strong counterargument to institutions worried AI means accepting marginal credit.
• Credit unions with large auto loan books should treat the middle credit tier as the highest-leverage segment for AI-assisted underwriting deployment.
• Executive ownership from lending leadership (not just IT) is critical to model calibration and staff adoption.