Stripe, the Silicon Valley online payments infrastructure provider, is hoping to take on the £10 billion problem of online payments fraud. It has launched a new feature called Radar, the idea being to give customers greater visibility of threats thanks to Stripe's extensive machine learning models for spotting fraudulent behaviour patterns.

Naturally Stripe has been protecting customers against fraudulent activity as much as possible for years, making today's launch essentially a coming-out party for its anti-fraud software.

John and Patrick Collison
John and Patrick Collison

Instead of working in the background the models and analytics are now "default included with all Stripe accounts," cofounder John Collison told Techworld yesterday. "You will see it in your dashboards and as of tomorrow people will get 90 percent of the benefits immediately and can choose to build on it and take action on it, so that it gets better".

Read next: Stripe cofounder says mobile banking apps are a cornerstone for bad design

How does it work?

According to Stripe, old methods of fighting fraud have never been optimised for the internet, with manual review and rules-based detection systems taking up valuable time.

Read next: Which is the right online payments solution for your startup? Stripe vs Braintree vs Worldpay vs Adyen

Stripe wants to automate fraud detection as much as possible. With Radar, customers get a dashboard with alerts along with a level of certainty to the transaction being fraudulent and why Stripe has flagged it. Admins can then set their own rules for how to deal with certain patterns of behaviour or individual transactions, such as blocking or refunding a transaction.

Stripe Radar in action


Collison is naturally sceptical of specialist third party fraud detection vendors, saying they are hard to integrate and most importantly, only run on proprietary data. Where Stripe stands out, according to Collison, is the breadth of its transactional data across mobile and desktop.

This allows their machine learning models to spot what is normal and what is potentially fraudulent across the internet as a whole, rather than in a specific business domain.

Read next: Featurespace: can machine learning and maths help banks detect digital fraud?

By constantly reviewing their models Stripe can avoid rules becoming entrenched across the organisation, leading to user frustration. For example, an organisation gets hit with a raft of fraudulent transactions from Vietnam so it sets a rule to block these as standard.

This would be frustrating for legitimate Vietnamese customers and Collison says Stripe's "self-healing models" and granular controls for customers allow for more nimble fraud protection.