The discipline that lets ML projects survive contact with operations: measure what you are replacing before you build the thing that replaces it.

Why I pushed the business toward being data informed — pairing data with judgment, context, and experience — instead of blindly data driven.