• Lifted quote conversion from 29.7% to 38.9% (+9.3 percentage points) • Designed natural-language risk input, so agents describe a risk in plain language instead of filling a rigid form • Built confidence-tagged recommendations, framed as "likely in this carrier's appetite" rather than a guarantee • Tested and dropped a Kanban-style tracking board after research showed agents didn't need it, keeping the product simpler and faster to ship