First Hawaiian Bank

First Hawaiian Bank, a regional bank serving Hawaii's diverse population, relied on traditional FICO-based credit scoring that systematically excluded creditworthy borrowers with non-traditional or limited credit histories — including recent graduates, immigrants, and thin-file customers common in island communities. These applicants were declined not due to actual credit risk but due to insufficient data depth in bureau files. Simultaneously, the manual underwriting process was too slow to compete with fintech lenders offering near-instant decisions. The combined effect was lost lending volume and reduced financial inclusion, both strategic priorities for the bank.

First Hawaiian Bank deployed Zest AI's machine learning credit underwriting platform to automate consumer loan decisioning. The system uses machine learning and predictive analytics to evaluate applicants by incorporating alternative data signals alongside traditional bureau data, identifying repayment patterns that FICO models miss for thin-file borrowers. Zest AI's platform integrated with First Hawaiian Bank's existing loan origination workflow, enabling automated decisions without replacing the underlying infrastructure. The model was trained and validated against the bank's own portfolio performance data, ensuring predictions reflected local borrower behavior rather than generic national patterns. Automated decisioning replaced manual review for a significant share of applications, compressing approval timelines from days to near-real-time.

First Hawaiian Bank achieved measurable improvements in both approval volume and operational throughput after deploying Zest AI's underwriting platform:

• Increased loan approval rates for qualified borrowers who would have been declined under FICO-only scoring
• Automated underwriting decisions at significantly faster speed than manual review
• Maintained portfolio loss performance, demonstrating that expanded approvals did not come at the cost of credit quality

The AI model proved particularly effective for thin-file borrowers — recent graduates and immigrants — improving financial inclusion without relaxing credit standards. Underwriting staff were redeployed toward complex exception cases rather than routine decisioning.

• Alternative data unlocks creditworthy borrowers that bureau-only models systematically decline; lenders with diverse or immigrant populations should evaluate this gap quantitatively before assuming FICO is sufficient.
• Model training on local portfolio data matters — generic national ML models may not reflect regional borrower behavior; insist on institution-specific calibration.
• Automation expands capacity without headcount — redeploying underwriters to complex cases is a more sustainable path than hiring to match fintech speed.
• Financial inclusion and risk management are not in tension when ML models are built and validated correctly; approval rate gains can coexist with stable loss rates.