The problem
Skeps runs loan origination for lenders and processes $2.5B in loans a year. At that volume, some loans that get approved are risky from the start, and once the money is disbursed, recovering it is slow and expensive. The safest place to catch them is before disbursement.
Every lender sees risk differently. A rule that suits one lender's portfolio can block good customers for another. So a single fixed set of checks run by Skeps wasn't going to work.
What I did
I started with the data. I analysed payment behaviour across loans to find the patterns that came before a default, and turned those patterns into rules.
Then I built those rules into an engine lenders configure themselves. Each lender decides which rules apply to them and how strict they are, without asking our engineers to change anything.
What changed
- An estimated $16M in potential defaults prevented before disbursement.
- Lenders control their own fraud rules, instead of waiting on Skeps for every change.