How to make your rules engine more powerful by using advanced features

When it comes to fraud, rules are easy, convenient and quick to implement. The transparency in creating rules makes explainability easier. While the ease of use of rules is well understood and accepted, so are the limitations. With the increase in the volume and variety of digital data, it becomes difficult if not impossible to stop fraud with just simple statement rules. Moreover, as the rules base grows in size, it becomes cumbersome to identify which rules are being fired and which are effective, not to mention the overhead of managing these rules. Many of the rules created are complex and involve computation that cannot stand the stress of a real-time production environment.

So, how do you continue to leverage the simplicity of rules while increasing their effectiveness? DataVisor’s answer to this is simple, effective and efficient: A centralized library of fraud detection features that can be used within a rules engine or machine learning models to dramatically increase detection and efficiency without increasing complexity.

What is a feature?

Simply put, a feature is an attribute or characteristic we want to describe – it can use complex mathematical features or statistical functions and can be used to enhance detection. Features can be used in multiple use cases – whether it is for fraud detection or for customer segmentation.

Examples of Features:

Let’s assume that you have been asked to define a condition for “does this account belong to a high-value customer”? This involves specific business logic and data analytics (e.g., if the account was created more than a year ago, and the customer has spent X dollars on the platform or conducted N transactions in the last three months, or opened our promotion emails > 60% of the time). 

This definition may vary across the rules created by different teams, and can change over time. The rules themselves may be very complex, involving several (even dozens of) conditions, making the rule difficult to write, interpret and maintain.

What if you could define “account characteristics of a good customer” centrally in a feature, and use the feature as a whitelist for fraud or as a criteria for triggering marketing promotions? Doing so would help align different functions within the company, improve consistency for definitions of key predictive indicators, simplify maintenance and even streamline compliance.

How do DataVisor’s out-of-box features work?

DataVisor’s Features Platform makes using complex features easier with a library of  out-of-box features. This library contains fraud detection features that have been developed over the years using fraud domain expertise. These features are highly scalable in a production, real-time environment and can be deployed quickly and easily through a rules engine or used in machine learning models.

DataVisor offers out-of -the-box features specific to each fraud use case — from account opening to transactions. It also enables users to look at features by tags — such as IP address, phone numbers and so on, ,in addition to advanced deep learning features and features based on our Global Intelligence Network.

The highly scalable, real-time feature library can compute features in production environments with low latency. This means rules with these features can be quickly applied to emerging fraud patterns to proactively stop fraud before loss occurs.

How can an out of box feature library help power rules engines?

By using pre-built features that are designed for specific fraud use cases, fraud operation teams can dramatically increase detection. Advanced deep learning features or features that are derived from the DataVisor Global Intelligence Network are based on fraud domain knowledge and have been applied across 4B user accounts, worldwide. These proven fraud detection features can accelerate detection without increasing the complexity of rules whose performance may be unproven.

Moreover, a centralized library of features makes compliance, control and maintenance a lot easier. When a feature needs to be changed – for example: the definition of a high worth customer – these can be consistent and managed so all relevant applications adopt the same definition across the organization. It also ensures the efficacy of these definitions when resources who create and update these definitions move on.

And more importantly, they enable fraud operations teams to truly be nimble and adapt to fast emerging fraud without any latency and dependence on internal teams. For an leading insurance client, DataVisor was able to accelerate implementation of features in production from weeks to hours, rapidly increasing the response to fraud.

Want to know how features can be used within the rules engine to help boost fraud detection? Request a 1:1 demo today.