Niural AI is a global HR, payroll, and workforce platform building first-party AI-Emma across its product. The existing expense module had inherited a report-first pattern from the Expensify era, had no integrated AI surface, and routed several flows through CSV export a UX gap that read internally as a trust problem. I led a multi-phase redesign covering submission, approval, settings, and Emma AI integration, anchored on three decisions: expense-first submission, not report-first; Approve and Pay always separated into two distinct steps; and a three-state AI verdict vocabulary Clear, Issues found, Processing chosen so an admin scanning fifty rows could interpret a verdict in a fraction of a second. The approval policy builder was designed once as a platform pattern reusable across Expense, AP Bills, and AR Invoices. Wireframes were accepted by engineering as the v2 spec; hi-fi and live build are post-internship work.