4 of the top 5 global banks in the world are fighting back with a new weapon in the war against fraud: advanced graph analytics. Within this piece, Dr. Yu Xu explains how financial institutions must escalate their fraud detection efforts to stay one step ahead of fraudsters.
Banks and financial institutions have always been on high alert when it comes to detecting financial fraud and COVID-19 has escalated these efforts.
In fact, U.S. banks have reported a record spike in fraud this year. Financial institutions filed almost 2,000 suspicious activity reportsOpens a new window involving business loan fraud in August 2020, according to the Treasury Department’s Financial Crimes Enforcement Network. Fraudsters, meanwhile, continue to adjust their tactics involving various intricate scam techniques.
4 of the top 5 global banks in the world are fighting back with a new weapon in the war against fraud: advanced graph analytics. Advanced data analytics in graph databases can detect suspicious patterns of online payment activity in ways that other database systems cannot – preempting and preventing potential fraud.
Learn More: Can Advanced Behavioral Analytics Put an End to Financial Frauds?
Fraud: An Increasingly Complex Problem
Fraud detection systems tend to rely on examining transactions that exceed preset levels or identifying people who attempt to max out a credit card with no intention of paying it off. These types of suspicious transactions are easy to detect because they rely primarily on the information in the transaction itself, looking at the amount, the destination or other properties that might trigger a warning.Â
However, fraud has become more complex as fraudsters have learned to work across multiple accounts, including mule accounts, making individual transactions look ordinary. Money “mules†are people or accounts that receive money from a third party and then illegally transfer the funds to another account in return for a fee.Â
Fraudsters can employ hundreds of accounts to launder money in a technique known as “smurfing.†Here, money is disbursed to accounts in quantities small enough to avoid triggering automated reporting limits. The money is then “layeredâ€, mixed, divided, and transferred from these accounts to other accounts, with the history of the money becoming more difficult to trace with every transfer. Ultimately it is transferred to the final destination, its true origin lost in a long and complex audit trail. Layering requires a fraudster to set up all of the accounts they need with the banks, but once that is complete, it becomes easy to move money quickly thanks to electronic banking APIs.Â
Graph: Real-Time Help for Fraud Detection
How does the graph help banks pinpoint and intercept complex fraud trails?
Using deep-link analysis, graphs can analyze thousands of customer data points – and the crucial relationships between them – to deliver fraud alerts in real-time.Â
Graph techniques can be used for fighting financial fraud by analyzing the links between people, phones, and bank accounts (among other things) to reveal indicators of fraudulent behavior, not only helping banks pinpoint suspicious activity but also giving them the tools to explain what’s going on.Â
Graphs have the ability to perform at speed, especially compared to relational database solutions such as SQL. While banks have had fraud detection systems in place for years, graphs add speed and analytical depth to the equation. While SQL depends on bulky table joins, graph is less memory-intensive and able to handle a greater query load.Â
Learn More: Open Banking Leads the Next Generation of Financial Services
Graph in Action
One of the ways to detect fraud is to find groups of transactions or persons that have an unusually high number of interconnections. Let’s consider an example. Suppose Alice wants to send $1 million to Bob without setting off internal bank alerts. She divides the money into 110 transactions of around $9,100 each to stay below the mandatory reporting threshold of $10,000. These are then deposited into 110 separate bank accounts. Then, each of these accounts divides its money into transactions of between $100 and $1000 each and sends them to other intermediary accounts. Then repeat several more times. Each step is another layer and adds another hop to the depth of transactions that would have to be analyzed to uncover this fraud.Â
After ten layers, there have been thousands of transactions and the source of the money in each account (which never exceeded $10,000) is truly obscured. Then all the accounts disburse their money to Bob. A cursory examination of the bank records shows Alice’s company paying a lot of suppliers and Bob’s company being paid by a lot of customers. No red flags because these scenarios fit the pattern of normal business behavior.Â
However, if you could map it visually, it would show the money flowing out from a single artery into a lot of capillaries and then back into one main artery, revealing that there was a transaction for $1 million between Alice and Bob.Â
Graph is optimal for running algorithms for finding connections. Community detection algorithms, for instance, would help to detect the fact that Alice was using a host of mule accounts to wash her money. Path detection would quickly identify that Bob was frequently the final destination for her money. And in the case where you found a community but didn’t know who was running it, centrality algorithms would help identify Alice as the kingpin.Â
Learn More: Open Banking Isn’t Without Its Cybersecurity Risks. Here’s How To Overcome Them.
Graph and Machine Learning: Looking Forward
Many financial institutions are now looking at coupling graphs with machine learning to further boost results. Graph is excellent at generating data for training ML systems because it can produce explainable models of what it has detected. Rather than simply giving something a score based on heuristics, the graph generates data on the links between different objects in the database, which can be fed into ML systems for further analysis. Graph databases are good at showing results graphically, making explainability a key strength. This, combined with the ability to explore contextual data, makes it an asset in fraud detection.
Today, financial institutions must escalate their fraud detection efforts to stay one step ahead of fraudsters. In doing so, company leaders should work to identify, evaluate and implement specific technology that will help protect their customers — and more specifically, their customers’ hard-earned dollars.
Let us know if you liked this article or tell us on LinkedInOpens a new window , TwitterOpens a new window , or FacebookOpens a new window . We would love to hear from you!