The challenge
Meridian's underwriting team was processing roughly 4,000 loan applications per week. Each application included a payslip, two bank statements, and a government-issued ID. The team of 12 operators reviewed each document set manually, with average turnaround of 48 hours and a 6% error rate that drove rework and customer churn.
The approach
Meridian piloted fluex against their existing manual workflow over a 14-day evaluation period. Real production documents (de-identified for the pilot) were processed in parallel; outputs compared against operator-verified ground truth. fluex hit 99.4% field accuracy on the first pass with no custom training, against the operator team's 94% accuracy at 48-hour latency.
The outcome
Meridian moved 90% of underwriting volume to fluex within 60 days. Two operators were retained for exception review of low-confidence extractions; the rest were reallocated to portfolio risk monitoring. Decisions on clean applications now complete in under 3 seconds; complex applications with 50+ page bank statements run async with webhook callback. Underwriting capacity increased 8x without headcount growth.
What this proves
Case studies don't generalize perfectly — every customer's volume, document mix, and compliance environment is different. But the architectural pattern repeats: replace the data-entry layer with a structured-extraction API, route the genuinely uncertain cases to humans, preserve a defensible audit trail, and the constraint shifts from headcount to judgment. That's the offer fluex makes; the metrics here are one shape of what it looks like in production.
For a side-by-side evaluation against your current workflow with your real documents, talk to our team. For pricing, see pricing.