Credit Union of Colorado

Credit Union of Colorado faced a structural challenge common across the credit union industry: traditional underwriting models built around FICO scores, debt-to-income ratios, and credit bureau data systematically excluded a meaningful share of creditworthy applicants. Members with thin credit files — recent immigrants, young adults, or those who had relied primarily on cash — could not be evaluated through conventional metrics regardless of their actual financial behavior. For a member-owned institution whose mission centers on financial inclusion, this created a compounding problem: foregone loan volume, an underserved membership, and no reliable way to distinguish genuinely high-risk applicants from those simply lacking a conventional credit footprint.

In October 2022, the credit union deployed Scienaptic AI's credit decision platform across its full consumer lending suite — auto loans, credit cards, personal loans, and lines of credit. Rather than replacing existing underwriting criteria, the platform layers alternative data on top of traditional bureau and FICO inputs. Rent payment history sourced from LexisNexis and bank account cash flow data from Plaid are combined with conventional credit signals to produce a holistic risk score that captures financial behavior invisible to bureau-only models. The system operates in real time: approximately 60% of consumer loan applications receive an automated approval or decline decision instantly, while borderline cases are routed to human underwriters for review. The platform also runs continuous post-origination monitoring, flagging behavioral shifts — such as disappearing direct deposits or irregular payment timing — as early warning signals for potential delinquency before accounts reach default.

The AI-assisted underwriting delivered measurable gains on both sides of the lending equation simultaneously. By approving applicants previously screened out by traditional criteria, the institution unlocked incremental loan volume while improving portfolio quality — outcomes that conventional underwriting wisdom would treat as trade-offs. Key results include:

• $40 million in additional consumer loans approved that would have been declined under the prior model
• ~50% reduction in overall consumer loan losses, with charge-offs effectively eliminated from the newly approved segment
• 60% of consumer loan applications processed via automated real-time decision
• Portfolio growth achieved without adding underwriting headcount
• Real-time pre-qualified bundled offers now available to members at point of need

• Alternative data closes the thin-file gap: Rent payment history and bank account cash flows surface creditworthiness that bureau scores miss, making them particularly powerful for serving underbanked members.
• Augmentation outperforms replacement: Layering AI on top of existing underwriting — rather than substituting it — improved both approval rates and loss performance simultaneously, defusing the traditional access-versus-risk trade-off.
• Automation enables scale without proportional cost: Handling 60% of decisions automatically allows lending teams to grow portfolios without equivalent headcount growth.
• Post-origination monitoring is as valuable as origination: Real-time behavioral signals turn early delinquency risk into a member service opportunity rather than a write-off.