Blog Posts Process Analysis

We’re Going to Science the $#!& out of AML

Blog: Enterprise Decision Management Blog


The Martian, starring Matt Damon, was one of the best movies I’d seen in a long time: space travel, the triumph of the human spirit and a whole lot of science wrapped up in a two-hour thriller. You know the punch line — and, I have to say, that same can-do spirit inspires me when tackling new analytic challenges.

Anti-money laundering (AML) is a good example. My FICO colleague TJ Horan recently blogged about the skyrocketing compliance costs that banks face in their fight against money laundering and other financial crimes. TJ wrote:

In 2015 we acquired TONBELLER, an innovator in risk-based financial crime prevention and compliance. Like finding the relative handful of fraudulent credit card transactions in the billions that are processed daily, predictive analytics are enhancing FICO TONBELLER’s solution portfolio, which runs the gamut from anti-money laundering, to know your customer (KYC), to sanctions compliance. Through these enhancements, we are finely honing the portfolio’s efficacy in fighting financial crime.

As TJ says, we are infusing FICO TONBELLER’s solution portfolio with analytic innovation. Here are two examples of technology FICO has developed, and for which we’ve made patent applications:

Machine learning (ML) for improved AML monitoring

Machine learning is at the core of our vision for AML, migrating from transactional, rules-based approaches to solutions built on self-learning models. We’re using variants of technology proven effective in other FICO products such as FICO® Falcon Fraud Manager.

The first place we are applying machine learning is in know your customer (KYC) practices. A lot of legacy AML systems are built around KYC, which segments customers into different risk categories and applies regulatory rules, in order to be compliant with AML monitoring requirements.

The challenge here is that rules-based systems don’t offer a comprehensive approach to detecting and stopping money laundering; they simply yield a greater emphasis on demonstrating compliance. In addition, a rules-based approach makes it too easy for criminals to fake their way through the KYC process; what the bank knows about the customer may not be accurate as to who the customer really is or what the person’s intentions really are.

FICO’s KYC innovation uses collaborative profiling technology based on Bayesian modeling techniques, taking transactions in the aggregate and producing archetypes of transaction behavior. What is powerful about this analytic technology is that, with each transaction, it automatically updates archetype allocations associated with a customer. So, for example, let’s say that Customer 1 falls into Customer Segment A. That customer’s archetype distribution is constantly updating with each transaction, and being compared to all the rest of the people in this segment: what are their archetype distributions, and are the current archetype loadings of Customer 1 similar or dissimilar to the others in the cluster?

The more that Customer 1 behaves similarly to clusters of archetypes in Customer Segment A, the more confidence we have that the rules applied are germane. The more that Customer 1’s behavior is different from clusters of archetypes, or changing from where it lies in the archetype space, the higher the risk that Customer 1 is incorrectly classified (in a KYC sense) and should be treated with an enhanced customer due diligence (EDD)..

This approach allows FICO to come up with an automatic indication of abnormality of customers from others in that segment, based on constantly evolving transaction data . This is the key to pinpointing potential money laundering and where rules applied to customers break down; with self-learning models we can monitor:

In this way, we can detect through transaction history whether specific behaviors are typical for customers within a particular segmentation, or not. This gives us a new and very effective approach to re-evaluating KYC and AML threat with each transaction. When used to enhance traditional AML monitoring, this layer of machine learning “frosting” helps banks to ensure that transaction rules are applied to the right customers.

The use of scoring to rank-order AML cases

The second patent applies more FICO payment card fraud detection technology to the financial crime arena. Here, we build a model indicative of normal or abnormal behavior of the banking customer. It can incorporate the archetype analysis discussed above, in the first patent, but it also applies machine learning to the transaction behavior. This allows us to differentiate between certain activities that are typical vs. abnormal for a customer, and customers who are outliers.

For example, a customer wiring money to an account in a conflict-stricken region of the world might raise a flag in a rules-based AML system. But if the sender and receiver accounts are trusted, this behavior might be typical and legitimate for the customer in question. Like payment card purchases, it comes down to the AML system answering the question, “What is typical behavior for this customer, and customers that share same archetype transaction patterns?” That way, we don’t stop legitimate transactions for good customers.

Similarly, this approach alerts us to changes in behavior. It will indicate when a customer has tricked the bank with behaviors that appear legitimate, to gain trust, and then switched to behaviors that are more risky. These outlier models allow us to come up with an analytically derived score that indicates a high risk of money laundering activity, thus allowing cases to be rank-ordered in severity and prioritized for investigation. Leveraging these AML analytic scores equates to faster and more accurate detection and much reduced false positives, a problem that plagues the AML space today.

As TJ discussed in his blog post, banks are buried under an enormous volume of suspicious activity reports. Both of the pending FICO patents will help financial institutions to focus their resources far more effectively on the cases that matter most.

Interestingly, FICO’s analytics approach aligns with what other industry thought leaders are saying. This 2014 report by Booz Allen declares, “The next big shift in the fight against financial crime and money laundering is advanced machine learning and sophisticated data science.” Somehow, I know that astronaut Mark Watney would agree.

Follow me on Twitter @ScottZoldi

The post We’re Going to Science the $#!& out of AML appeared first on FICO.

Leave a Comment

Get the BPI Web Feed

Using the HTML code below, you can display this Business Process Incubator page content with the current filter and sorting inside your web site for FREE.

Copy/Paste this code in your website html code:

<iframe src="" frameborder="0" scrolling="auto" width="100%" height="700">

Customizing your BPI Web Feed

You can click on the Get the BPI Web Feed link on any of our page to create the best possible feed for your site. Here are a few tips to customize your BPI Web Feed.

Customizing the Content Filter
On any page, you can add filter criteria using the MORE FILTERS interface:

Customizing the Content Filter

Customizing the Content Sorting
Clicking on the sorting options will also change the way your BPI Web Feed will be ordered on your site:

Get the BPI Web Feed

Some integration examples