Stripe, which started out in 2010 by providing a set of unified APIs and tools that enables businesses to accept and manage online payments for any type of credit or debit card, is using that experience to branch out in a new security area.
The San Francisco-based payments-and-security software company has launched a new version of its Radar platform, featuring tools designed for enterprises that use machine-learning models in the software.
The company contends that Radar prevented $4 billion in attempted fraud in 2017 alone by learning from the transactions processed on the Stripe network for hundreds of thousands of businesses and helping users tailor defenses for their individual company. With the announcement, Stripe makes a new bundle of advanced fraud prevention tools available to risk professionals within large businesses.
Radar for Fraud Teams, designed for sophisticated teams of fraud professionals, improves visibility and offers granular control for identifying and preventing fraud.
Fraud professionals within large organizations can use Radar 2.0 to optimize for:
- Faster and more accurate reviews: When reviewing payments, Radar shows relevant info and related payments that a user’s business has processed. By gaining a broader view into attributes like a typical purchase pathway or a mismatch between country of incoming IP address and country where a card was issued, risk professionals can more quickly evaluate fraudulent activity.
- Custom rules with real-time feedback: Radar’s fraud prevention logic can now be customized with unique rules (i.e. “block all transactions above $1,000 when the IP country does not match the card's country.”). It also previews the rule on historical data to help risk professionals evaluate its impact on live transactions.
- Custom risk thresholds: Radar helps risk professionals to maximize revenue by allowing them to set custom thresholds at which to block payments.
- Block and allow lists:Users now have an easy way to create and maintain lists of attributes—card numbers, emails, IP addresses, and more—that should be consistently blocked or allowed.
- Rich analytics on fraud performance: Radar highlights dispute trends for a user’s business, the effectiveness of reviewing flagged payments, and the impact of rules customized for a user’s business.
Faster Machine-Learning Improvements
This 2.0 release marks the biggest update to Radar’s machine learning models since its launch in 2016. Stripe has added hundreds of new signals that distinguish legitimate customers from fraudsters, including purchase patterns that are highly predictive of fraud. As a result, the upgraded machine-learning models help businesses reduce fraud by up to an additional 25 percent, while keeping payment acceptance rates high.
Proxy detection is one example of a new, highly predictive signal that has been incorporated into Radar’s machine learning models, the company said. It measures the round-trip time between Stripe and a potential fraudster’s browser, helping to pinpoint if a fraudster is using a proxy or VPN.
Radar also constantly evaluates patterns that are unique to a user’s business. Now on a daily frequency, Radar updates and re-trains its models, in the process evaluating each user’s unique transaction profile to determine which model will achieve the best performance. By training machine learning models for specific use cases, Radar can precisely serve businesses of all sizes and types with more accurate and performant results.
By using a cloud-based service, users will automatically benefit from future daily updates, with defenses that will adapt even more quickly to constant changes in fraudster tactics.
Radar for Fraud Teams is being used in production by Watsi, Fitbit, Restocks, Patreon, and others. For more information, go here.