Anti-Money Laundering using Graph Analytics

What is money laundering?

The technique of transforming huge monetary gains from illicit activity into legal assets while hiding their real origins is known as money laundering. To combat such acts, governments all over the world have been increasingly tightening AML policies. Financial institutions are now obligated to adhere to strong anti-money laundering rules and to disclose any suspicions of money laundering activities. Money laundering has a significant societal impact since it fuels terrorism, trafficking, drug dealing, and other criminal activities.

The problems and challenges with money laundering starts with:

• Rising AML operational costs: it is pushing financial institutions to seek alternatives to their present tools and technologies in order to avoid fines and penalties.

• The rise in false positives: it keeps compliance personnel distracted and as a result, resources are spread thin across all phases of the AML process.

• The prevalence of false negatives: sophisticated criminals who are able to circumvent AML protocols in order to perpetrate crimes.

• Difficulty is locating money laundering practices: Every year, money launderers become more skilled, establishing an elaborate network of identities and accounts through which to channel their illicit activities which makes locating the false negatives hidden deep inside the mountain of valid transactions very difficult and time-consuming.

The graph approach

A graph or network is a collection of nodes and connections (also called edges). Graph analytics is a collection of analytic tools that enable the investigation of links between items of interest such as companies, individuals, and transactions. It assists data and analytics executives in analysing linkages in data and reviewing data that is difficult to evaluate using standard analytics.

In the field of anti-money laundering systems, the concept of networks and connection analysis is fundamental because it helps expose hidden aspects of transactions that are not discoverable by any other means. When paired with ML algorithms, these technologies have the potential to trawl through hundreds of data sources and documents, allowing financial institutions and AML specialists to quickly uncover hidden patterns and relationships in transactions.

Graph analytics is essentially a set of analytic tools that allow you to "dig down" into complicated interrelationships between businesses, individuals, and transactions. For example, a major international investment bank in the United States is utilising sophisticated graph analytics to strengthen its fraud prevention activities, especially fraud detection for debit and credit cards. The organisation is integrating graph analytics into its machine learning system to discover data links between “known fraud” credit card applications and fresh ones. As a consequence, the bank can discover more suspicious trends, reveal fraud rings, and close down fraudulent cards more quickly. The bank will save millions of dollars each year as a result.

Graphs may be used to detect anomalous patterns, which can aid in the prevention of fraudulent transactions. Terrorist activity has been found in certain cases by examining the flow of money across interconnected banking networks.

Fig 1 : Fraud detection with regular analytics and with advanced graph analytics can be visualised from Fig. 1. The use of graph analytics allows for the dynamic study of relationships within a huge dataset. It is possible to investigate and visualised who and what a client is linked to using data as diverse as an email, a phone number, a device, transactions, and so on. The detection of accomplices becomes very rapid

A regular fraud detection case

A tip or a detection system may occasionally flag a client or a transaction as suspicious. In this circumstance, it is vital to determine whether or not this particular questionable circumstance is isolated. The customer might be a member of a larger criminal ring, or the transaction might be part of a broader operation. In the absence of more information, it is critical to pursue as many leads as possible. This necessitates investigating what the customer or transaction are related to.

Consider a simple payment made using a digital payment provider such as PayTM, PayPal, Google Pay, Amazon Pay, or Razor Pay to see an example of possible fraud and why it is so hard to identify using standard analytics. A user has opened a new account that is connected to their Bank X credit card. They have connected their phone number and email address to their account as part of the setup and two-factor authentication. The user uses an Apple iPhone X with the registered phone number as their device and starts a payment of Rs. 5000 to another account. Because the user is a new user with a new phone number and email address, there are no red lights or alerts in a standard financial services fraud detection solution at this stage (none of these have been associated with any fraudulent transactions in past). Regular analytics does not uncover anything strange or suspect and the payment passes through without being reported or refused.

Use of Graph analytics on the case

Deeper analysis with a native parallel graph analytics technology, on the other hand, offers a different image. There is a fraudulent activity related with a gadget, phone number, and stolen credit card six levels inside. Here's how it goes down: The payment's recipient account belongs to a user who authenticated the account with a Phone Number as part of the account registration procedure, and that phone number is used with a different device Apple iPhone Y. As the deep link analysis searches the history of previous fraudulent transactions for devices linked with those transactions, it discovers that this Device was used last year with a different Phone Number to set up a separate Account. This account initiated a payment that was subsequently discovered to be fraudulent since it was paid using a stolen credit card.

Advanced analytics using graph analytics may go deep into the related data, in this case six links deeper, to uncover the link to earlier fraud in real time, and the payment transaction is refused as a consequence. As you can see, advanced graph analytics is required for real-time payment fraud detection — and this analytics identifies fraud three layers "deeper" than normal analytics. This disparity between normal and advanced graph analytics can result in hundreds of millions of dollars in fraud losses. Advanced graph analytics with real-time processing can process the payment transaction in under a second and then perform the multi-connections query on the related dataset. In other words, the system must check every connection along the path from the person initiating the payment to the ultimate receiver, the one involved in fraudulent Payment.

Clearly, fraud detection is at the top of every financial services organisation's priority list – and this is unlikely to change. As fraudsters grow increasingly tech-savvy, it is critical for businesses to keep one step ahead of them. These deeper insights are enabled by advanced graph analytics, which complements conventional BI technologies and powers AI and machine learning. Hence, as a consequence, firms can anticipate and avoid possible fraud while also safeguarding their consumers.