I wasted $8K applying regression when a real business needed classification. This article breaks down the failure, the lesson, and how better problem framing leads to better machine learning decisions.