For me, the proverbial final straw came when I delivered a dinosaur utility company it's first churn propensity score. Bit of background, my manager and their manager were non-technical and I had this thing running in evaluation mode for a few months with excellent lift in the top couple of percentiles. I had the statistics well grounded, and this thing was due to go in against basically nothing, renewal teams randomly calling prospects.

Anyway, before this thing goes live, I see a few managers printing out pages of Excel sheets and arranging them on a conference table to validate the model with literally pencil and paper. They didn't know what a percentile was until I explained it.

I handed in my notice the Monday following that weekend. Sometimes, if you're not trusted to do a good job, it's not worth trying.