Take me to...
Debit cards remain the most popular payment method in Australia, representing about 70% of the country’s electronic transactions each month.
Most Australian debit cards are dual-network cards that allow payments through either the national eftpos network or one of two international networks. The ability to route these dual-network debit card transactions through the lowest-cost network option is driving significant growth in both Card-Present (CP) and Card-Not-Present (CNP) transactions for eftpos.
However, criminals are now focusing on exploiting debit card payments, particularly Card-Not-Present transactions. Despite existing fraud detection systems and strategies, financial institutions often struggle to detect and prevent fraudulent activities without causing unnecessary friction for cardholders.
To complement its members’ fraud prevention strategies and deliver secure and seamless transaction experiences, eftpos aimed to utilize a full card fraud machine learning solution covering both CP and CNP transactions.
Despite not having historical Card Not Present (CNP) data, eftpos leveraged ARIC Risk Hub’s ability to analyze transaction data across the entire network to make inferences about behavior and quickly identify anomalies in cardholder behavior.
ARIC Risk Hub utilizes Adaptive Behavioral Analytics, a Featurespace invention, to build profiles of genuine cardholder behavior and adapt to changing behaviors over time. For eftpos, the solution involves a bespoke machine learning model for risk scoring CP transactions, a new approach to analytics to deliver a rules-model hybrid (RMH) for CNP transactions, and a set of complex rules to cover fraud sub-classification codes.
As eftpos had a rich set of historical CP data, they launched ARIC™ Risk Hub with the CP model first within a set threshold. Any transaction that achieves a risk score above a set threshold will trigger an alert. The threshold can be increased or decreased depending on risk appetite.
Featurespace has developed a “cold-start” model for electronic funds transfer at point of sale (EFTPOS) to address the challenge of having no historical data. The model is designed to provide accurate scores for card-not-present (CNP) transactions, which are significantly riskier than card-present (CP) transactions.
During offline research, the transferred models caught 65%-85% as much fraud as otherwise identical models trained on data drawn from the same source. Featurespace’s SCC field, shared with members along with the fraud score, provides greater explainability. Some SCC risk types show accumulating fraud rates that are 10-100 times higher than the no-risk typology code, demonstrating the accuracy of the score.Download the full case study
Today, the network intelligence that feeds the eftpos model ensures all members benefit from the model’s learnings on attacks and new typologies, whether or not their organization is the initial target.
eftpos members are achieving strength in numbers against rising global and domestic levels of financial crime.Download the full case study