The most durable AI value is often invisible: embedded in operational workflows, running behind ERP systems, quietly removing manual work from processes unchanged for fifteen years. This session is about that harder, less glamorous category of AI work. The speaker covers the data quality problems to solve first, the integration constraints that shape what is possible, and how to make the ROI case to a finance team skeptical of AI promises. Attendee takeaways - The data quality baseline required before AI can add reliable value in a legacy operational context - How to scope AI features in back-office workflows so they reduce friction without creating new failure modes - How to build the ROI case for back-office AI with a finance team that has heard too many AI promises