First Hawaiian Bank

First Hawaiian Bank (FHB) operated in a challenging lending environment where legacy credit scoring models constrained both accuracy and reach. Hawaii's population includes a significant share of thin-file and unscorable borrowers — segments traditional FICO-based models systematically exclude. Manual underwriting created decision bottlenecks, slowing application throughput and limiting scalability. With only 10% of applications receiving instant decisions, the bank's ability to compete on speed and serve its full community was materially hampered. Building better models in-house was impractical given the resource and time investment required, leaving the bank stuck with risk models that restricted growth.

FHB partnered with Zest AI to replace its legacy scoring infrastructure with an AI-driven automated underwriting platform built on machine learning and predictive analytics. Zest AI's models ingest a broader set of credit signals than traditional scorecards, enabling more accurate risk differentiation across both prime and non-traditional borrowers. The platform was integrated directly into FHB's application decisioning workflow, automating a significantly larger share of credit decisions without adding manual review overhead. Critically, Zest AI compressed what would have been a multi-year in-house model development effort into a roughly six-month deployment timeline — allowing FHB to move from implementation to production results within a single fiscal year.

Within one year of deployment, FHB's instant decisioning rate jumped from 10% to 55% — a fivefold increase reflecting both the model's confidence thresholds and the breadth of applications it could evaluate automatically. Overall approval rates rose to 65%, extending credit access to previously underserved and unscorable populations. The model deployment was completed in approximately six months, well below typical in-house timelines. Qualitatively, the initiative shifted underwriting from a manual, case-by-case process toward a scalable, data-driven operation. The results were independently validated when Celent awarded FHB its 2025 Model Bank Award for AI-Augmented Retail Lending.

• Expanding instant decisioning requires replacing scoring models wholesale, not layering AI on top of legacy rules — FHB's results came from a full platform transition.
• Serving thin-file and unscorable populations is both a business opportunity and a community obligation; ML models that ingest non-traditional signals can unlock this segment without relaxing risk standards.
• Vendor partnerships can compress model deployment from years to roughly six months, making speed-to-value a realistic expectation rather than an aspirational one.
• Clear, pre-defined success metrics — instant decisioning rate, approval rate, deployment timeline — are what separate transformation from experimentation.
• Third-party validation (such as industry awards) provides useful evidence that AI-driven lending outcomes meet regulatory and peer scrutiny.