FORUM Credit Union

Forum Credit Union, a $2.3 billion institution based in Indianapolis, faced a scaling problem common across the credit union sector: loan volume was growing while staffing was not. Loan officers spent 15–20 minutes per application on document classification alone — manually sorting income verification, tax returns, bank statements, and employment letters before any underwriting work could begin. Data extraction was entirely manual, with staff keying figures directly into the core system. Underwriting preparation required pulling information from multiple sources while applying complex, institution-specific policy logic. Off-the-shelf SaaS tools offered limited relief, delivering only 10–15% improvement because they could not accommodate Forum's proprietary document formats, custom underwriting criteria, or deeply configured core system — leaving no viable path to scale without hiring or accepting slower turnaround times.

Advisor Labs developed custom AI models trained specifically on Forum Credit Union's own document types, formats, and underwriting policies rather than deploying a generic lending platform. Using computer vision and document AI, the system performs automated classification at over 95% accuracy across Forum's complete document taxonomy — including internal loan forms, regional bank statement layouts, and member-submitted photo documents. An intelligent extraction layer maps recognized fields directly to Forum's core system fields, eliminating manual re-keying and reducing data entry errors. An underwriting acceleration module applies Forum's actual policy logic — calculating debt-to-income ratios, evaluating collateral, and flagging exceptions according to Forum's own criteria — then assembles complete packages for human review. The entire solution was built to integrate with Forum's existing core system and preserve current workflows, which drove immediate loan officer adoption without retraining or process disruption.

The compound effect of automating each manual stage produced a 70% increase in loan processing volume without adding headcount. Improvements spanned the full workflow:

• Document classification: reduced from 15–20 minutes per application to seconds
• Underwriting preparation: reduced from 45 minutes to approximately 5 minutes of human review
• Data accuracy: improved through automated extraction mapped directly to core system fields, removing manual re-keying

Loan officers adopted the system immediately because it fit into existing workflows rather than replacing them. Advisor Labs attributed the 70% volume gain — versus the 10–15% typical of generic tools — directly to the fact that models were trained on Forum's actual documents and policies, not industry-generic samples.

• Custom AI trained on institution-specific documents and underwriting policies substantially outperforms generic SaaS platforms, which typically yield only 10–15% improvement due to process misalignment.
• Core system integration (Symitar, Corelation, DNA, etc.) must be a first-class design requirement from day one — retrofitting integration after model development is where most credit union AI implementations stall.
• Training on real institutional documents, not synthetic data, is what produces the classification accuracy required for production-grade lending automation.
• Keeping humans in the loop for final approvals satisfies NCUA regulatory expectations while still capturing the majority of efficiency gains.
• Target the highest-friction bottleneck first; Forum started with document classification because it was the most time-intensive step and the easiest to measure, then expanded from there.