iuvity Case Study

iuvity Increases Fraud Detection Rates by 200 Percent with Neo4j

Latin America is one the fastest growing regions for digital banking and payments. Banks must deliver a seamless user experience, giving them, at most, half a second to approve or deny the transaction before the experience starts to crumble. This is especially true in Latin America, where payments are typically processed by banks in real time. People expect things to happen fast.

That half second is where iuvity operates – using Neo4j’s powerful graph database as the foundation for iuviPROFILER, its own proprietary solution for assessing the risk of fraud in an instant. It’s an important part of iuvity’s offering, which handles the entire digital channel and customer experience for major banks in Latin America.

By the numbers: iuviPROFILER

• Graph scale: 250 million nodes, 2.2 billion relationships
• Transaction rate: 500 transactions per second
• Fraud detection rate: 200% improvement
• Losses prevented: $40M+ annually
• Transactions protected: 3.2B+ annually
• Neo4j query speed: 10ths of a millisecond
• Platform: Neo4j Enterprise Edition on Microsoft Azure

“Doing fraud detection in real time with a traditional, relational database is a nonstarter,” said Dr. Edgar Osuna, Chief Data and Analytics Officer at iuvity. “There are simply too many joins between too many tables to make it feasible.”

As iuvity explored an opportunity to build iuviPROFILER, it was clear to Osuna that he needed to use a new type of data structure to solve this problem. “When you’re working the problem on the whiteboard, you start drawing what happens in a fraud. You have lots of dots and lines. Those are graphs.”

“The problem was ‘screaming graphs’,” he said. The solution would be a graph application that was flexible – and extremely fast.

How Fast Can iuviPROFILER Fly?

iuviPROFILER operates adjacent to the financial services and banking sector in Latin America, providing digital solutions with a focus on the end user and ease of use. As Osuna puts it, “iuviPROFILER empowers banks to conduct business securely. In the end, we want to reduce the friction between individuals and their financial world.”

Fraud detection is something banks need. iuviPROFILER uses Neo4j as a component to process data and identify risk. This product is the in-house solution the iuvity team developed as an alternative to a commercial fraud detection tool that was no longer serving the customer’s needs.

Osuna’s financial and tech background helped him see this challenge from all angles. More than 20 years as an executive in multinational financial institutions gives him deep insights into business requirements and customer needs. He has the technical chops to match: his research was awarded the IEEE Longuet-Higgins Prize for published contributions in the field of machine learning.

Osuna said he chose Neo4j because the technology is mature enough to handle hundreds of transactions a second, reliably, and especially during periods of peak demand such as holidays and paydays. In other words, speed was of the essence. “This thing has to fly,” Osuna said.

iuviPROFILER can sustain hundreds of transactions a second, spending merely tens of milliseconds on a single query, thanks to the streamlined logic behind the Cypher query language.

The Graph Evolves

It has been more than two years since iuvity used Neo4j to launch iuviPROFILER, and over time its graph data model has evolved. Neo4j can handle as-needed changes to the data model, not just real-time deletion and appending. And the beauty of it, Osuna said, is that “the data model itself has been morphing without having to stop the service.”

“And that was one of the things that I think we saw in the beginning that was powerful,” he said. “You can always add to Neo4j.”

Cloud Portability and Ease of Migration

The iuvity team originally engineered iuviPROFILER to run on a different cloud platform, a decision made during a time of rapid technology transition at iuvity. In the end they migrated from one major cloud provider to another. The migration “went smoothly and without surprises,” he said. “It ended up not being a difference besides some fine-tuning.”

Neo4j, it so happens, works seamlessly with any major cloud hosting provider, so the switch to Azure was uneventful.

“It was a different infrastructure, for sure. But we didn’t have to recode anything or reprogram anything; everything just worked in the way it should,” he said.

Eliminate Friction for a Great Customer Experience

In our digital era, where data and analytics and information are so often stripped away and isolated from the real people behind the numbers, Osuna likes to stay focused on who he views as the true client.

“In the end we’re serving the person using the app,” said Osuna. “We want them to feel confident when they use the app.”

As cyberattacks have increased 30–40% in Latin America in recent years, Osuna relies on Neo4j to build apps that people can trust.

“We’ve all lived through a rejected transaction on a credit card or a call back on something,” he said. “It’s not just that the transfer on the phone gets executed with the right journey, but also that you don’t stop for fraud if it isn’t fraud.”

This end-user experience drives iuviPROFILER, and is why so many banks choose to adopt it as their fraud prevention solution.

iuvity replaced a well-known fraud detection product with its own application written on Neo4j and was able to compare results side by side.

“I think because we were able to look at the same data, the same banks, the same everything, and compare them bid to bid, we were able to show internally and show the banks that we were better,” said Osuna.

“For the same false positive rate, we’re able to achieve twice the detection rate,” he said. “And that in the end is less friction for the customer, less losses for the bank, and a better feeling in terms of protection for their customers.”

The Futureproof Database Is Neo4j

As iuvity looks forward, Osuna said he sees Neo4j as a “crucial part” of delivering new features in its products.

That relationship is critical, he said, “Because from my side, it means that in terms of storing information, retrieving information, getting that data model to evolve, I don’t need to look for another vendor in terms of databases, for another vendor to fill in a gap for features A, B, C, and D that are in my roadmap.” As a key commitment to this relationship, Neo4j provides 24/7 support for Enterprise customers, as well as an extensive community and knowledge base to ensure ongoing success.

“And the way we see it is just as I’m delivering, let’s say, these four things now out of Neo4j into my solution, then the fifth, the sixth, and whatever else is going to be coming from the same database engine. It’s going to feed into our enhanced, bigger, and more sophisticated models if and when it needs to,” said Osuna.