Sign up for The Paypers newsletter Follow The Paypers on LinkedIn Follow The Paypers on Twitter Follow The Paypers on Facebook Follow The Paypers on Google +
The Paypers, paypers, Insight in payments, News, Reports, Events
 advertisement
Expert opinion

Play by your own rules: how machine learning helps tackle fraudsters

Friday 9 March 2018 | 08:40 AM CET

Tackling fraud while unlocking the highest potential of machine learning – best practices revealed by CyberSource

Fraudsters have grown more sophisticated. They’re more adept than ever at using technology – even big data and analytics – to conduct fraud. And they use it not only to increase the number of fraudulent orders but also to be cleverer about disguising illicit activity. Fraudsters have also worked out the best moments to attack, and how to capitalise on their opportunities. As a result, it’s getting tougher for organisations to determine which transactions to accept and which to reject.

For many companies, the answer lies in machine learning. It’s an important tool in helping businesses take on fraudsters, especially because it can analyse vast amounts of data. This provides a sophisticated defence against sophisticated attacks.

The more, the merrier

As transactions flow through global payment networks, the data behind each transaction can help provide vital insights. Machine learning flourishes when it’s dealing with high volumes of transaction data. By looking at the information to identify high risk or low risk transactions, machine learning gets better at making accurate predictions and guiding businesses to make the right call – and the more data there is, the more accurate it becomes.

Which is your best option?

In general terms, machine learning offers two very different approaches. Static models use huge amounts of historic transaction data to identify historical fraud patterns and work well when they are brand new. However, fraudsters keep on changing tactics, creating pressure for fraud managers to keep up with fresh patterns.

Self-learning models, on the other hand, thrive on new data. They use it to recognise and adapt to the latest fraud patterns. Even so, they’re complex; that can make it difficult for people to track, control or adjust what the machine learns.

Machine learning can help with:

Simpler real-time decision-making. Many fraud mitigation platforms use rules to determine which orders to accept and which to reject. But they’re manual and time consuming to set up or change. Machine learning makes rules work. It can also speed up data analysis.

Greater accuracy. Criminals continually create subtler, more non-intuitive patterns. People can find it hard to recognise these – but machine learning doesn’t.

Faster response to change. Fraud recognition is a constant game of cat-and-mouse. The right machine learning models can use the latest data and update their approach to reflect new trends.

Lower costs. Major technological advances have cut the costs of machine learning and the computing systems that can run it. Machine learning can also reduce costly false positives as well as the time and cost of manual reviews.

But machine learning:

  • Depends on receiving good input data, and plenty of it. Without this, the machine can learn the wrong thing and could make the wrong assessments

  • Is a great tool for automating the pre-determined patterns associated with fraudulent behaviours. However, it needs significant expertise and training to be fully effective, especially since fraud trends evolve rapidly.

  • Can often be a black box, especially when it uses self-learning techniques. The machine can learn the wrong thing, and its decision making isn’t fully transparent.

The best option for fraud prevention is to combine automated machine learning with a rules-based approach, giving businesses more immediate control over fraud decisions.

Bringing together the best

CyberSource’s fraud management platform, Decision Manager, combines its own unique version of machine learning with its flexible rules-based engine.

Its machine learning model brings together the flexible data analysis of advanced self-learning and the best bits of the static model. It’s always learning from a huge amount of data that’s constantly updated. That makes it swift and accurate in responding to unique or emerging trends, with the right approach for each situation.

And added to this is Decision Manager’s flexible rules-based engine. Rules act as the first line of defence by applying deterministic decisions to an order; when those can’t tell a good transaction from a bad one, machine learning steps in.

The rules-based engine uses 260 anomaly detectors, and 15 region, channel and industry-specific risk models, each tuned to identify fraud in different scenarios. Not only can fraud analysts set and adjust the rules at any time, it’s easy to see which rules were applied to make a specific decision.

But core to the success of any fraud management platform is data. It’s key for greater accuracy and detection. Decision Manager’s machine learning and rules-based capability is fuelled by data that’s richer and more plentiful. And it’s more relevant because it uses more fraud detectors and actual outcomes of past transactions. Decision Manager is also uniquely enabled by more than 68 billion worldwide transactions, processed annually by Visa and CyberSource, creating the world’s largest detection radar.

The result? You can spot and handle even the latest types of fraud more efficiently. This means you can help reduce fraud losses, protect your revenue and operate more efficiently. It can also help keep customers happy and returning, making sure they stay loyal and aren’t driven away by refused orders and fraudulent use of accounts.

This editorial was first published in our Web Fraud Prevention and Online Authentication Market Guide 2017/2018. The Guide is a complete overview of the fraud management, digital identity verification and authentication ecosystem provided by thought leaders in the industry from leading solution providers (both established and new players) to associations and experts.

About CyberSource

CyberSource is a global, modular payment management platform built on secure Visa infrastructure with the benefits and insights of a vast USD 427 billion global processing network. This solution helps businesses operate with agility and reach their digital commerce goals by enhancing customer experience, growing revenues and mitigating risk. For acquirer partners, CyberSource provides a technology platform, payments expertise and support services that help them grow and manage their merchant portfolio to fulfill their brand promise. For more information, please visit www.cybersource.co.uk.

 advertisement
 advertisement
 advertisement
 advertisement