Danielle Wright

Expensify processed 8M+ expense reports monthly, but 35% required manual categorization review. The ops team of 45 reviewers cost $3.8M/year. Average review time was 72 hours, causing customer complaints about delayed reimbursements.

Built an AI categorization system using Claude to parse receipt images, merchant names, and transaction metadata. Created a feedback loop where reviewer corrections trained a custom classification model. Implemented confidence thresholds: high-confidence auto-approves, medium routes to a simplified review, low gets full manual review.

Manual review rate dropped from 35% to 8%. Reduced reviewer team from 45 to 12 FTEs, saving $2.8M/year. Average categorization time went from 72 hours to 4 minutes for auto-approved expenses. Customer satisfaction for expense processing increased 28 NPS points.

Bill.com had no visibility into invoice processing times. Enterprise customers complained about missed SLAs but the team couldn't identify bottlenecks.

Built a real-time ops dashboard in Looker tracking every invoice from receipt to payment. Added automated alerts for SLA breaches and bottleneck detection.

SLA compliance improved from 71% to 94%. Identified that 60% of delays came from a single approval step, which was then automated.