ScrumLaunch designed an AI workflow to replace a manual, error-prone order process. It ingests customer orders downloaded from SAP Cloud, cleans and normalizes them from a patchwork of Excel formats into the client's internal structure, and reconciles forecast demand against actual realized demand — built with Claude paired with Python scripts and connected to the client's JobBoss ERP. The hardest part was capturing the business logic: the tribal knowledge living around the workflow, like the fact that orders actually leave the facility on a different date than the one listed or requested. Those small, insider-only rules drove much of the fine-tuning needed to reach acceptable accuracy. With that logic encoded, what had taken a person fifteen hours a week collapsed to under two, and monthly revenue forecasting — previously thrown off by disparate formats, forecast windows and business logic — improved by 30%. The build came together in under four weeks.