How Machine Learning Keeps Your Money Safe

What’s the first thing that comes to mind when you think of a bank? Cash, cards…bank robbers!? If you thought of the last one, think again.

Jokes aside, when it comes to banking, security is extremely important. While the move to digital money and mobile banking have made some forms of crime redundant, fraud prevention is often like a game of cat and mouse. By using the latest technologies, we simplify your banking experience, so you can focus on what matters. Be it spending time with your family, carefree traveling or growing your business, we’re always on the lookout for the next best thing that gives you more freedom.

When we launched in 2015 we set out to challenge the status quo of what a bank should be. This led to some early innovations like in-app CVC codes (goodbye card skimming) and no physical branches (hello, speed and convenience!). Today, we’re exploring a less talked about benefit of mobile banking: fraud detection and prevention.

The challenge is to check millions of transactions in real-time for illegal activities and prevent fraud. Sounds simple, right? Oh, and also to make sure none of this affects the great user experience you’re accustomed to.

Consider a scenario where an account starts receiving irregular payments from various unknown bank accounts, and then suddenly all the money is withdrawn. Do you think this is:

a) A case of money laundering?

b) Friends paying you back for pizza?

c) Going full-on Marie Kondo mode and selling your vinyls on eBay?

Good question! To answer this, and to make sure we’re tough on criminals and easy on you, we use supervised and unsupervised machine learning.

If you talk to our compliance team, you’ll hear terms like Nigerian Prince, smurfing or money-muling thrown around, showing that many types of bank frauds are actually well known. To detect these patterns we use supervised machine learning. This is an algorithm that looks at over 500 data points per transaction and creates hundreds of competing decision trees to decide if a transaction is suspicious or not. The algorithm learns based on years of labeled transaction data, and is designed to keep you safe at all times!

This works well for spotting existing patterns, but what about completely new types of fraud? Say hello to …unsupervised learning!

By running all non-suspicious transactions through an auto-encoder deep learning model, we find highly complex and unusual patterns that may be linked to previously unknown fraud schemes. If confirmed, the data is labeled and used by the supervised learning models as input — another one bites the dust!

If all of this sounds complicated, you may ask yourself: isn’t there a simpler way?

Yes, there is! And it’s called rules. Until recently, most traditional banks didn’t use machine learning to spot fraudulent patterns. And even in 2020 they mostly rely on simple, rule-based models which means defining suspicious patterns and then manually checking each time they happen.

These rules are usually not as effective as machine learning. However, in certain cases they can be very powerful so we use them in two ways: when a pattern is always fraudulent, and when our supervised learning models are still learning to detect the new pattern using data generated by the rule.

Here’s why our hybrid approach keeps you safer:

1. We don’t discriminate based on nationality or looks

Of course, no one does so on purpose—but by focusing on behavioral patterns instead of stereotypes and biases, our model is more fair than human compliance officers.

2. We can tailor the user experience to be effortless for you and terrible for fraudsters

Machine learning allows faster detection of fraudulent behavior than rule-based models and it’s a lot more accurate.

3. By always staying one step ahead of criminals, we have more time to focus on what really matters: building the best bank for you

So there you go! This is one of the ways we keep your money safe each day, but of course there’s a lot more happening under the hood.

If you’re interested in hearing more about how we prevent fraud and keep you safe, please let us know here!

Spread love, not fraud!