New users struggled to understand what they needed to do, in what order, and when they were "done." I redesigned the end-to-end audit workflow from blank state to completed review.
Complify is an AI compliance platform that automates financial audit workflows (User Access Reviews) that auditors currently do manually in spreadsheets. The setup experience exposed too many requirements at once, leading to confusion, hesitation, and early churn.
The interface relied on rigid, structured fields. When information didn't fit predefined schemas, or was only temporarily relevant, agents had no appropriate place to record it.
Started with a node-based configuration approach for audit setup. Through user testing, discovered auditors don't think in terms of configurable nodes. They think in sequential steps.
Pivoted to a wizard-based workflow that mirrors how auditors actually set up a review: select control type, fill details, define test plan, configure sampling.
The core product is a 3-panel layout: source document viewer, AI test results, and sample sidebar. Each sample runs 4 automated tests against the source evidence, with confidence scores and extraction reasoning visible on every field.
Shipped 18+ hosted prototypes over 3 months, testing each with real audit data from Zillow Home Loans (365 employees, 53 flagged samples) and SeatGeek (1,500+ review items). New prototypes went out fast enough to run real auditor sessions the same sprint they were built.
Complify signed a paid pilot with RSM, one of the largest audit firms in the US, based directly on the prototype work. No industry competitor had removed the friction from the laborious node configuration, and prospective clients bought in on the new design immediately.
The wizard-based setup reduced audit spec creation from a multi-day manual process to under 30 minutes. Population-level testing runs automatically on upload.
Tested across two real client datasets (Zillow, SeatGeek) to confirm the design handles different org structures, scale, and review formats. The engagement kept going and I continued leading the next set of product work.
Want to see the prototypes or discuss the work?
