One of the core challenges was the absence of a feedback loop between pre-repair estimates and actual invoice outcomes, making it difficult to identify cost anomalies, track patterns, or adjust payouts based on real data. Structuring invoice data at scale added complexity, particularly when dealing with a wide range of vendors, formats, and regional differences. Manual validation introduced delays, increased friction, and reduced trust in the process. Any viable solution needed to deliver the speed required for scale while maintaining enough control to support high-value, judgment-driven decisions.