Machine learning to mitigate risk
Using Machine Learning to Mitigate Payment Fraud
In today's rapidly evolving digital economy, merchants face the dual challenge of preventing fraud while minimizing false declines that can alienate genuine customers. Our Machine Learning (ML) algorithm provides a robust solution by screening payments during the pre-authorization assessment. ML in the pre-authorization phase of payment processing represents a critical step in fraud prevention. By screening payments and assessing risk before transactions reach the authorization stage, ML algorithms offer several key benefits:
Early detection: Proactively identify potential fraudulent activities based on transaction data and historical patterns before they proceed. This proactive approach stops fraudsters at the initial stage, reducing the likelihood of fraudulent transactions being processed and causing financial damage.
Cost savings: Preventing fraud early in the transaction process reduces the costs associated with chargebacks, lost merchandise, and administrative and payment fees. Additionally, by lowering false declines, merchants can retain more revenue from legitimate transactions.
This assessment utilizes information from payment requests alongside historical data from our extensive network, generating a riskiness classification (Very Low, Low, Medium, High, Very High).
Why Merchants Should Embrace Machine Learning
Merchants often struggle to strike a balance between detecting fraud and avoiding the blocking of genuine customers. False declines represent a significant financial burden, costing e-commerce businesses an estimated $443 billion annually—far exceeding the cost of actual fraud.
Historically, Protect has relied on manual risk rules; a complex and time-consuming approach requiring constant monitoring and reactive updates to stay ahead of evolving fraud patterns. This can leave companies vulnerable to fraud if not diligently managed. To support merchants in balancing fraud prevention with customer experience, our latest version of the risk engine introduced machine learning at the core of the risk profile with the goal to reduce manual workload and improve performance.
Advantages of ML in Fraud Prevention
Leverage Adyen global transaction data: Benefit from insights and patterns detected across the Adyen platform. Helps you recognise new and genuine shoppers faster, reducing false positives and improving customer experience;
Operational Efficiency: Reduce the manual workload for risk management teams
Enhanced Fraud Detection: Capture complex fraud patterns more accurately and proactively.
How Does It Work?
The premium tier utilizes transaction attributes, ShopperDNA, and Adyen global transaction data to identify fraudulent activity. It supports two distinct models:
Machine Learning: Bot Attack Risk: Automatically blocks high speed attacks such as bot and card testing. Relies on crucial data points like shopper email and IP address to detect high-velocity or repeating patterns.
Machine Learning: Fraud Risk: Predicts the likelihood of fraudulent disputes based on issuer labels, focusing on disputable payment methods (Cards (debit/credit) and PayPal)
Merchants have the flexibility to configure the strictness of the “Machine Learning: Fraud Risk” check based on their risk appetite. The default setting is set on the "High" threshold to balance fraud prevention and customer satisfaction, ensuring genuine shoppers are rarely blocked while maintaining acceptable fraud levels. However, merchants can adjust this threshold to suit their specific needs.
It's important to recognize that certain types of fraud are beyond the scope of our current ML models. We recommend creating custom rules using domain expertise to enhance overall performance.
The basic tier utilizes transaction and shopper’s attributes (i.e. shopper email, IP address, card chunk) to identify high-velocity attacks. This version supports a rule-based model aimed at saving you on excessive retry fees and cannot be customized.