There was a "study" about drug testing welfare applicants in Florida a few years ago where the authors claimed that it only caught a handful of people and cost more than the money saved.

I looked up the numbers and found that applications took a historic plunge when the law went into effect, and an equally large increase after the law was suspended. It was a huge multi-stage funnel where the study only looked at one stage after thousands of applicants had already self-selected for drug-freeness. Applications went down by thousands, and payouts by millions.

Basically if you hired a data scientist who told you this about some funnel you were analyzing, like a purchase or sign-up flow, you would fire this person within five minutes. But people repeated this gotcha story everywhere.