Discussion about this post

User's avatar
State of Play's avatar

Sixty-one percent of flagged candidates scored above the passing threshold and got the offer anyway: detection worked, and the hiring process absorbed the flag without changing the outcome for most of them. That tracks with a University of New South Wales experiment on algorithmic ratings, where managers followed high scores far more readily than low ones.

The same asymmetry shows up in your numbers: the technical score cleared the bar and got acted on, while the fraud signal that contradicted it didn't. Detection software fixed the measurement problem. On my read it didn't fix the organizational one: acting on a negative AI signal still requires someone to override a decision already forming before the offer goes out.

Ready for more?