In this case study, learn how we solved a real ingestion challenge at a major insurance carrier with 1,100+ file sources mapped to a single schema. We built a metadata-driven ADF framework with control tables, mapping dataflows, and a Copilot Studio agent that predicts mapping for unmapped fields. With a human in the loop, approved mappings flow back into ADF, creating a proven pattern for scalable, self-improving ingestion. We will dive into technical implementation and lessons learned.