Why machine learning detects payment fraud more accurately

Any business that sells goods or services online is increasingly vulnerable to attack by fraudsters. From using stolen credit card details for online purchases to creating false accounts and even voucher code abuse, traditional approaches to tackling this problem such as heuristic rules and business logic are quickly being outpaced by fraudsters.

A 2016 survey from Cybersource, one of the leading rules-based fraud detection solutions, revealed that over 25% of all transactions, across a variety of companies surveyed, were manually reviewed. These numbers are astonishing, especially when you think about how much time would take reviewing manually 25% of all transactions.

While manual review can be effective, it is clearly not the most efficient method to prevent fraud and can end up costing a business more than fraud itself. A solution that we have found to be efficient at tracking and stopping fraudsters is a machine learning-based approach.

Machine learning in ecommerce

Whilst machine learning has been around for a number of years, many ecommerce companies continue to rely on rules-based systems to detect fraudsters. Rules-based systems require quite a lot of manual review, and since ecommerce companies traditionally had a window of time to review transactions before the goods were shipped, they worked just fine.

Today however, this window is getting smaller and smaller, which gives retailers less time to manually catch fraudsters. This is where machine learning comes in. Machine learning-based systems use a company’s historical and live data to make fraud predictions based on patterns in behaviour from both genuine and fraudulent customers. Companies are already amassing huge data points about their customers and whilst this could be applied to a traditional static, rules-based system, the rules need to become increasingly complex to be effective, making it virtually impossible to maintain.

Whereas, machine learning by nature is perfect for dealing with large datasets and weighing multiple signals appropriately. The historical data acts as the training data because it contains chargeback information. From a data science perspective, having large historical datasets collected across many clients and industries and a very accurate set of training data (chargebacks versus non-chargebacks), means increased precision for spotting fraud.

Choosing the right models to apply to the historical dataset will optimise the levels of recall and precision that they provide. Recall is the proportion of fraud that was successfully flagged by the models and precision refers to the proportion of transactions flagged that were actually fraudulent.

In an ideal world, the models would identify 100% of fraud with 100% accuracy, but this is nearly impossible as fraudsters behave differently and there will always be some overlap between the behaviour of genuine and fraudulent users. To make matters worse, sophisticated fraudsters can often look like your best customers. So, in order to stop the most sophisticated fraudsters, you need to be willing to potentially flag genuine customers in the process. Machine learning allows you to make an informed decision on how you balance fraud exposure and conversion based on your companys risk appetite.

Moving beyond manual review

Are ecommerce companies ready to let machine learning do the manual labour for them? For businesses where speed, scale, and efficiency are paramount, they have no other option if they want to reduce fraud significantly. However, there is scope for this type of solution outside of the on-demand ecommerce industry too. Fraud detection platforms need to provide more accurate fraud decisions faster than those that rules-based systems can provide for businesses operating online. And machine learning is the only mature technology that is able to do that.

This does not mean that human insight is not needed. A machine learning approach needs manual review in order to decrease the risk of mistakenly flagging good customers as bad. Gleaning insights into customer behaviour and providing explanations for the decisions this is the way forward in fraud detection and it can only get better.

About Gerry Carr

Gerry is CMO of Ravelin, which provides fraud protection for online businesses. He joined Ravelin from its inception to help define and articulate a product vision for the changing face of fraud in ecommerce. Prior to Ravelin, Gerry has led the product marketing functions for products as diverse as Ubuntu and Sage CRM. Gerry loves to snowboard and compete in iron man contests when time allows.

About Ravelin

Ravelin prevents fraud and protects margins for online businesses. Companies all over the world are accepting more transactions with fewer chargebacks thanks to our machine learning-based approach to fraud prevention.