1
Hit a 70% accuracy ceiling on my hiring algorithm and it made me rethink everything
I built a screening algorithm for entry level roles at my company. After 6 months of tweaking it, I got it to score 70% on matching hires who stayed past 90 days. That number felt solid until I looked at the false negative rate. We were passing over people in the other 30% who might have been great. Now I wonder if 70% is actually good enough or if we need to accept lower accuracy to catch more potential. Has anyone else run into this tradeoff with their own models?
3 comments
Log in to join the discussion
Log In3 Comments
bennett.harper1mo agoMost Upvoted
Is it just me or does this same pattern show up EVERYWHERE in life? I've noticed it with online dating apps and even restaurant reviews. People give a place 4 out of 5 stars but their friend gives it 2 stars and they're both right. The 70% is probably as good as it gets for people. We're all too complicated for a computer to really nail down who's a good fit.
6
hollywhite1mo ago
Did you actually track what happened to the people your algorithm rejected? I'd want to know if any of them ended up succeeding somewhere else, or if they really were all bad hires. That 30% feels like the only way to tell if your threshold is actually right.
6