Fraud Prevention Best Practices

Combating Fraud with Custom Fraud Modeling & Machine Learning

May 25, 2018

With access to faster and cheaper computing, fraudsters have shifted their targets to more profitable weaker points in almost every vertical.

It’s wise to expect every part of your process to be continuously and thoroughly examined for potential weaknesses and opportunities. If, and when, weaknesses are found—you can bet word will spread like wildfire on fraudster communities and forums.

Of course, any fraud solution you choose should be adaptable for your business case. Examples include: your transaction types, customer profiles, and which regions you operate in.

But today’s world requires you to be able to quickly identify and stop risky transactions, while still approving the majority of your good volume. Here’s how you can employ custom fraud modeling do that.

The role of network intelligence

Network intelligence offers a holistic view of the space you operate within. But not all networks are equal. It’s important to differentiate siloed networks against those which rely on crowdsourced data.

Operating in a siloed network will only get you halfway there. Fraudsters rarely only focus on one merchant, industry or vertical. Instead, they hit as many as they can at once.

There are a lot of benefits in knowing what your peers are facing. Intelligence around fraud events allows you to identify risky behaviors faster. Then, you can react before they become a problem.

There’s a flip side, too: the same process works with legitimate customers. When you can identify positive behaviors, you can quickly approve those customers.

By unlocking the global intelligence in the network, our decision science team can build fraud models which are tailored to identify the latest fraud patterns across geographical markets, transaction types, card products, and segments. Models are carefully segmented to ensure that they can uniquely capture both local and regional fraud trends and transaction patterns.

Custom fraud modeling

Crowdsourced data lends a wider and more operational view. It also provides a wider variety of elements for mapping custom fraud models. Custom fraud models gives you the coverage of industry and peer trends, allowing comparison against your own transactions.

So, when choosing a provider, make sure you assess the ability to offer customized models. The next step is adding machine learning capabilities to update those models in a real time.

Machine learning uses data history to create a fraud model that is deployed against current transaction activity to highlight suspicious transactions compared to model data. This approach ensures fraud models are timely and current for optimal fraud detection — enabling you to frequently fine-tune the performance of your fraud model by adjusting rules and parameters

It’s important to establish feedback loops into the process to ensure that it can monitor model performance over time, and proactively work to refresh the fraud model as fraud patterns change.

This provides robust, real-time fraud prevention controls at both ends. You can identify (and stop) relevant trends, patterns and behaviors. As with network intelligence, this process also extends to low-risk transactions and good customers.

Specialized fraud modeling

To face today’s threats, you have to take it to the next level: specialized fraud modeling. Specialized fraud modeling is exactly what it sounds like — the process of creating unique models to score your unique transactions.

Here’s how it works: Network and industry events are inputs for customized fraud modeling. Then, with this intelligence, you can develop specialized models for your business.

For example, peak spending seasons vary by country. Data has shown that in Europe, card usage may peak in the summer. In the U.S., the peak season is centered around the months of November and December. You can tune your specialized models to expect higher transaction volume during the typical peak season period for the country or region, which helps control false positive rates

The benefits of this approach are twofold. You’re getting the best outside data and utilizing it in a way that’s specific to your business.

In conclusion

Fraud is a difficult behavior to define, and even more difficult to predict. Fraudsters are constantly changing their methods, making patterns hard to establish.

Every company has a different procedure for processing transactions, and accordingly, fraud patterns are going to be different for each. This clearly calls for customized strategies for predicting and preventing fraud.

Companies today need to build fraud solutions that maintain consistent performance over time. The solutions should include multiple types of data, including a variety of consistency checks that can work to verify digital identity.


I invite you to follow me on LinkedIn

Follow Emailage on LinkedIn and Twitter (@emailage)

Click here to discover how to get secure, intelligent risk assessment using an email address. 

close-link
close-link
close-link