From the blind thread > Our back tested model error as well as true error rates (models only) were considerably smaller than what was realized. Largely because

From the blind thread > Our back tested model error as well as true error rates (models only) were considerably smaller than what was realized. Largely because leadership wanted to be aggressive on acquisitions and incentivized all teams to err on the side of growth, which manifested in system overrides. Without proper oversight and controls these lead to compounding and unchecked error Sounds less like blaming the data science team and more like the entire organization didn't want to face the music.