Loading…
14
My tiny audit tool flagged 7 loans in a row for the same zip code and nobody noticed till I did
I built a small script to check our lending model's approval rates by area, mostly just for myself, and it pointed out that 7 denials in a row all came from the same three blocks in East Austin. I brought the list to my manager expecting a shrug, and instead we found the training data was missing that whole area from 2019 onward. Has anyone else caught a data gap like this just by poking around on their own, or did it take a formal review to surface?
1 comments
Log in to join the discussion
Log In1 Comment
rowanhernandez1mo ago
A data gap like that is worth looking at but calling it a red flag right away feels like a stretch. Missing training data for a few blocks in one zip doesn't mean the model is broken, it could just be a records issue or those addresses got merged into another dataset somewhere. Seven denials in a row sounds spooky until you remember small sample sizes make patterns look bigger than they are. Where the real test comes in is whether the approval rate for that area actually dipped after 2019, if it stayed flat then you found a paperwork bug, not a fair lending problem.
2