Loading…
12
Finally got a biased hiring model to show its hand after 6 weeks of digging
I spent 6 weeks trying to prove a client's resume screening tool was quietly penalizing older workers. At first every test came back clean because the data was so messy. Then I started splitting the results by graduation year buckets and bam, there it was. Candidates over 45 got a 32% lower pass rate even with identical skills and experience on fake resumes. It took 43 separate test runs and a bunch of late nights to pin it down. The model was basically treating any job gap longer than 18 months as a huge red flag, which hit older folks harder. Now I'm trying to figure out if we can retrain it or if we need to scrap it entirely. Has anyone else had luck fixing a bias once you actually find it in the training data?
1 comments
Log in to join the discussion
Log In1 Comment
wyatt1351mo ago
Damn, that's wild. 43 test runs is dedication, lol. Honestly if the gap rule is that baked in, retraining might just bury it somewhere else. I'd scrap it and start fresh with the feature removed, way less risk of it sneaking back. Good luck with the client though, that's a rough conversation.
4