Financial Services – Regional Credit Union (8 Branches): Loan Origination & Decision-Support AI

Industry: credit union lending. Size: $1.9B assets, 8 branches. Complexity: HELOC + consumer + indirect auto. Systems: LOS, core, credit bureau hooks.

Fragmented apps, manual stip tracking, underwriter queue starvation. Member drop-off at document upload.

Application Agent, Doc Classify Agent, Stipulation Chaser Agent, Underwriting Summary Agent, Member Comms Agent with fair-lending guardrails.

1. Trigger: application start.
2. AI guides required fields.
3. Docs OCR’d and validated.
4. Stips identified; chasers scheduled.
5. UW packet auto-built.
6. UW decides; exceptions logged.
7. Continuous improvement on abandonment points.

Intake Agent, Document Agent, Stipulation Agent, Summary Agent, Compliance Comms Agent.

Bank-grade VPC, LOS APIs, encrypted storage, LLM with policy filters, human-in-loop on decisions.

MeridianLink, Fiserv core, Twilio, PostgreSQL, OpenAI, eOriginal

Pull-through +24%; cycle time −33%; processor manual minutes per loan −61%; member satisfaction +18 points.

Build: 11 weeks. Deploy: branch pilot then all. Optimization: 10 weeks credit policy alignment.