Loading…
7
Pro tip: I used a simple checklist to catch biased AI loan denials and it cut my error rate by 80%
I work as a loan officer at a small credit union in Boise, and 3 months ago we started using an AI tool to pre-screen applications. I noticed it was denying a lot of applicants from certain zip codes for no clear reason. So I made a one-page checklist with 5 checks: verify income source, look at zip code, check if the denial reason matches the data, compare to similar approved apps, and log any weird patterns. After 6 weeks, my error rate dropped from about 15% to 3%, and I caught 4 denials that were clearly biased. Has anyone else tried building their own bias check before trusting the AI? What did you include?
1 comments
Log in to join the discussion
Log In1 Comment
lee_thompson9d ago
15% to 3%" sounds like a lot, but if the AI was flagging people in certain zip codes, wouldnt the rate have been way worse than 15%? Not saying bias in lending isnt real, just feels like one person's 6 week test on a small sample isnt enough to say you fixed anything.
5