Articles
Credit score bias checks before you release a changed score
Learn how to test whether near-equal approval rates hide group differences inside score bands, then decide which field to investigate before release.
Reject inference: test each decline rule before replacing it
See what approved-only scorecards and reject inference methods can tell you, then use a random test arm to decide whether each decline rule should stay.
What ArkOwl can reveal about an email address
ArkOwl shows the history behind an email address, including how long it has been seen and where it has accounts. That evidence helps fraud teams distinguish an established address from one that may be difficult to trust.
How many bads does a credit model need?
Three hundred bads is not a pass mark for a scorecard. The answer depends on the model's expected Gini, the parameters the build will try, and how precisely you need to know the bad rate behind pricing and provisioning.
How to use Trestle IQ Decision Signals for better anti-fraud
A fraud or credit team evaluating an identity data service needs to know what its checks cover and where the call belongs in a decision flow. Trestle IQ Decision Signals combines name, phone, address, email and IP checks with a composite identity score, with component results to read beside the policy action.
Why product attributes make your scorecard look strong and should not be in it
A public SBA loan analysis found that product attributes can dominate a scorecard because the lender chose them, while borrower attributes carry less separation. Test the top five characteristics by asking who sets each one, then move lender-set choices into a visible cutoff.
How life insurance underwriting turns an application into terms
Life underwriting does not reduce every application to accept or reject. It starts with standard mortality, adds or removes the effects of relevant factors, and then maps the result to terms such as standard rates, a loading, an exclusion, postponement, or decline.
When to Set Up Credit Scoring in Lending
New lenders often reach for a scorecard too early. That usually creates weak approvals, false rejects, and higher acquisition cost before the model has enough data to help. In market entry, antifraud comes first. Credit scoring matters later, when you have enough performance data to separate risk, reduce hard rules, and raise approval rates without weakening the book.
Your round-trip check produces false matches when amounts differ
A small allowance for fees can turn a round-trip check into a source of false matches. On synthetic SME bank statements, a 0.5% allowance produced 87 false matches on an applicant and 84 on a clean company when split payments could be combined.
How to Automate Factoring Approvals and Keep Control
Factoring approvals need 2 separate decisions. The first is whether the small company and the invoice are real. The second is whether the big company, the debtor, and the exposure fit your risk policy.