How we replaced manual document review with a compliance-grade extraction and form-filling system built to survive a regulator's audit, not just a demo.
The client is a US-based fintech handling high-volume client onboarding, where every new account requires document review: IDs, proof of address, bank statements, and supporting paperwork submitted in dozens of formats and qualities. Every document had to be manually read, matched against onboarding rules, and keyed into downstream systems before an account could go live.
The bottleneck wasn't headcount. It was throughput per compliance reviewer. Off-the-shelf OCR handled clean, single-format documents fine. It fell over on scanned images, inconsistent layouts, and the edge cases that make up a meaningful share of real submissions — exactly the cases a regulator expects a firm to handle correctly, not wave through.
The engagement ran through our standard four-stage framework. Each phase had a defined deliverable and a milestone gate.
Discover (weeks 1–3). We embedded with the compliance team to map the manual review workflow end to end. We identified which document types were highest-volume and highest-risk, and which onboarding rules were genuinely ambiguous versus just poorly documented.
Co-create (weeks 4–8). Built an extraction pipeline combining layout-aware OCR with LLM-based field validation against the client's onboarding rules. Reviewed with the compliance team every week, tuning for precision before we touched recall.
Build (weeks 8–13). Production integration into the onboarding platform and case management system. Added a human-in-the-loop escalation path for low-confidence extractions, and full audit logging so every decision was traceable back to source.
Scale (weeks 13–16). Rolled out across the full document set. Documentation, handover, internal training for the compliance team. Monitoring and observability in place before we stepped back.
[Note: placeholder attribution. To be updated with a real, consented quote from the client before publication.]
This was one of the three engagements that shaped the studio's early thesis. Each one involved a regulated or compliance-heavy operator. Each one involved integrating AI capability into workflows that existing vendors had failed to handle. And each one produced a pattern that's now informing our next ventures.
Document AI that's built for audit, not just accuracy, is a product category of its own. Watch this space.
Most of our engagements start with a 30-minute discovery call. No deck. We'll tell you plainly whether we're the right team.