Fortune 500 Financial Services Company Case Study

Real-Time Graph Analysis Creates Potential for Millions in Fraud Detection Savings

“No one knew a word about graph technology, but Neo4j was very supportive. They gave us the software to play with while the discussions were going on and provided our teams with in-depth training on its fraud detection capabilities.”

–       Fraud Detection Product Manager, Fortune 500 Financial Services Company

Analysts at this Fortune 500 financial services company rely on large amounts of data to make fast, accurate decisions related to potentially fraudulent activity.

Partnering with Neo4j and AWS enabled this company to use real-time data analysis and visualization to identify fraud patterns more accurately. Additionally, the firm significantly reduced manual review times, increasing the total number of reviewed transactions. The firm now stops fraudulent transactions worth millions per year.

The Company

This Fortune 500 financial services company delivers cross-border, cross-currency money movement, payments, and digital financial services to empower consumers, businesses, financial institutions, and governments – across more than 200 countries and territories and nearly 130 currencies – to connect with billions of bank accounts, millions of digital wallets and cards, and a global footprint of hundreds of thousands of retail locations.

The Challenge

The financial services company collects a huge amount of data provided by its customers as well as enriched data from outside vendors – all of which needs to be analyzed in real time before a transaction can be approved. While the majority of these requests are approved or denied instantly through an automated fraud detection system, potentially fraudulent requests are also submitted to an analyst for manual review.

Each analyst has a dedicated transaction review tool to manage relevant third-party data. However, they also had to query a Microsoft SQL Server database to review the customer’s history and look for any association with known fraudsters. These queries could require four or more levels of JOINs, a time-consuming, painstaking process that often exceeded SLAs related to transaction-review times. 

“It was taking five minutes or more to run a query,” said a product manager within the company’s fraud detection solutions department. “Since our analysts were reviewing up to 10,000 transactions each day, this just wasn’t a sustainable practice. So many link-analysis queries were putting a huge burden on our database, so we also realized that a relational database was not a good long-term fit for our objectives.” 

The final straw: the fact that the queries would return extremely complicated data that analysts had to review in a matter of minutes. Other processes didn’t work either. For example, data streamed from Splunk into their SQL-based reporting database wasn’t available for hours, yet meanwhile, fraudsters were completing their work in mere minutes.  

Facing so many challenges and limitations, the company quickly realized it needed to find a more efficient way to analyze data to reduce the time analysts spent on queries and ultimately, provide faster, better service to its customers. 

The Strategy

As the company began its evaluation process, it established two clear goals. First, decrease the amount of time it took to complete fraud-detection queries and provide analysts with simple data visualizations. In addition, the company needed a database technology that was infinitely scalable. 

The search team explored several data package solutions and data visualization tools. While there were a number of available graph visualization tools, they were limited in scale and performance by the underlying data store. To achieve the real-time results they needed, the team would need a database optimized for storing and traversing several levels of relationships. 

That’s when they turned to Neo4j.

The Solution

While the company initially sought a turnkey solution to avoid a drawn-out development process, the actual development time with Neo4j wasn’t too complex or time-consuming with a pilot project being easily completed during spare time. 

Once live, the Neo4j solution provided both real-time results with connected data and data visualization that allowed analysts to make faster, more accurate decisions. This opened the door for newer, more extensive searches, which the company hopes to expand from four to 10 degrees of separation. 

Also, the company analysts began noticing clusters and relationships in the data that revealed new, previously unnoticed potential fraud connections. Each customer can be represented by up to 30 nodes, each with up to 60 properties, totaling over 4.8 billion nodes and 14.2 billion relationships, and a 20% annual growth in database size.

AWS Deployment and Usage

Neo4j Enterprise Edition on AWS gives the company a powerful, flexible foundation for fraud detection, with a number of benefits:

1. Scalability and Performance: AWS hosts Neo4j datasets of approximately 2+ terabytes, spanning billions of nodes and relationships.
2. Enhanced AI and Machine Learning: The company has enriched its fraud detection models with graph-based features by integrating Neo4j with Amazon SageMaker, leading to more accurate identification of suspicious patterns.
3. Multi-Region Disaster Recovery: The company implemented a multi-region disaster recovery setup built on AWS infrastructure, ensuring high availability and business continuity for their critical fraud detection systems.
4. Security and Compliance: Using Neo4j’s Role-Based Access Control in conjunction with AWS’s security features, the company ensured that their sensitive financial data remained protected while still accessible to authorized analysts.

Above: Fortune 500 Financial Services Company’s Neo4j AWS Architecture

Looking ahead, the company is exploring Amazon Bedrock to ground their AI models in the knowledge graphs they’ve built, opening new avenues for identifying fraud schemes.

The Results

The data visualizations made possible by Neo4j cut analysts manual review times in half. This improvement now allows analysts to review twice the number of transactions on a daily basis – stopping many more fraudulent transactions much sooner and significantly reducing wait times for customers. 

Reviewing the data in Neo4j also uncovered suspicious activity that previously went undetected. With more insight into clusters and relationships, the company improved its ability to perform faster, more accurate fraud detection. 

In addition, the company began to see new possibilities beyond the original data visualization use case. The company is also integrating Neo4j with their decision platform to instantly stop fraudulent transactions and save the company thousands of dollars per day.

Get in Touch

Curious about the kinds of insights graph technology could unlock for your organization? Chat with a graph expert today.