How we use machine learning to protect you from fraud

We speak with banks all of the time, and we're always really shocked to learn that many of them are still relying on 100 percent manual processes for detecting fraud.

When thinking about our own strategy for protecting our customers against fraud, we found this kind of dated process to be slow, expensive and often inaccurate at the expense of the customer.

And if we are to become a world leader in customer protection, that would not be possible by simply adopting an outdated and timely process that has failed to leverage the power of technology.

And so, over the last 12 months, we've been investing massively in our engineering and data science teams in line with our vision to automate, accelerate and improve the quality of decision-making when it comes to card fraud and money laundering.

Behind the scenes, we have been building machine learning systems to develop deep insights and predictions around customer behaviour so that we can identify new fraud patterns in real time, without explicit human intervention.

Fighting card fraud 👊

These systems work by applying complex mathematical models to large sets of data in order to identify anomalies, offering a greater degree of accuracy when it comes to decision-making, and saving both the customer and the business valuable time.

What this means is that we are able to build extensive data profiles on our customers to better understand their spending behaviour and patterns. Our anti-fraud systems will then be able to detect any suspicious activity in real-time, based on abnormal activity which deviates from their usual behaviour.

For example, when our systems flag something suspicious on a customer, it will automatically block the payment or transfer until the customer verifies the transaction from within the app. This process cuts out of the fat and avoids the customer having to go through a lengthy security process in order to get their account unlocked, especially for something as minor as making a payment abroad.

Since we put these new systems in place about two months ago, we have seen a massive reduction in card fraud levels, specifically tackling common fraud cases such as e-commerce payments, card cloning and card theft.

Fighting money laundering 👊

Separately, we've also just launched new clever piece of technology for detecting money laundering - a dynamic mathematical model that calculates a risk score for each user based on their activity history. It processes all transactions in Revolut live and statistically determines the probability of money laundering based on the profile of the user and various features extracted from the transaction.

All these estimations are then fed into the algorithm to output a unique score for each user. When this value exceeds a certain threshold, the customer is required to submit documentation to justify the activity, which the model incorporates to intelligently recalibrate the user’s risk exposure.

Just the beginning ⏰

We strongly believe a combination of machine learning and human intervention can best tackle areas such as card fraud and money laundering, which is why we are massively investing in both of these areas. And while it's still early days for this kind of technology, we are packed full of ideas as to where we can implement machine learning technology next. Stay tuned. ✌