Key takeaways
- Commonwealth Bank of Australia (CBA) cut scam losses 76% from its 2022 peak and reduced fraud 30%, with H2O (CBA case study).
- CBA scores 16M+ customers in real time and has trained 900+ analysts (CBA case study).
- H2O sits underneath your existing fraud engine as a scoring and challenger layer, so your rules, decision flows and case system stay in place.
- Every model built in H2O Driverless AI ships with reason codes and model documentation for your model risk process.
Catch a scam before the money leaves, not after
In a scam, your own customer makes the payment. The login is real, the device is known and the authorization is valid, so the checks built for stolen credentials see nothing wrong.
The loss still lands on your books and on your customer. You are measured on net fraud and scam loss, and each false alert costs an analyst hour and a frustrated customer. Tightening the rules moves one number in the wrong direction to fix the other.
Any change also has to pass your model risk review. A score nobody can explain does not get past model risk, internal audit or your regulator.