When the computer power of artificial intelligence meets the transactional complexity of the AML, good things happen. I have recently had the opportunity to talk about it during the annual conference assessment and credit control of Edinburgh . This is a subject that absolutely fascinates me because I firmly believe that the elimination of the scourge of money laundering could make the world a much better place for life.
Money laundering consists of giving the impression that the illicit funds obtained by the illegal activity came from legitimate sources. He is responsible for the supply of narcotics, trafficking in human beings and terrorism. According to the UN Office on Drugs and Crime, the estimated amount of money laundered in one year equates to 5% of global GDP, or $ 2 trillion. That's almost the size of the annual GDP of the United Kingdom.
To deal with this threat, regulators have fined banks that do not stop money laundering. The fines have multiplied by 500 in the space of a decade and now rise to nearly $ 10 billion a year.
Finding a needle in a haystack
It is not that the banks have not put in place controls to counter this threat. Each bank has dedicated large teams whose sole purpose is to monitor financial and non-financial transactions and to identify and create suspicious activity reports, or SARs. They rely on a Know-Your-Customer process and use transaction rule gaggles to identify SARs. These processes are subject to scrutiny to ensure that banks comply with regulators' guidelines regarding supervisory practices.
But it is becoming increasingly clear that the rules are often not sufficient to detect money laundering cases, and regulators have begun to pressure the banks to adopt them. sophisticated analyzes in their workflows. This is not surprising because rule-based systems typically generate large amounts of false positives, resulting in the study of thousands of false leads. In addition, the rules reflect previous expert knowledge, but may not reveal new sophisticated money laundering systems designed to circumvent the rules in place.
My talk focused on the use of machine learning to combat money laundering. We built these analyzes using the advanced patented artificial intelligence systems proven by FICO. We have combined our practical and operational expertise in machine learning – acquired over the past 25 years using AI to protect about two-thirds of worldwide payment card transactions – with the power of our FICO TONBELLER Anti-Financial Crime Solutions This allows us to focus on injecting new machine learning into world-class AML detection methodologies.
Behavioral Misalignment Score in Gentle Clustering
One of the abilities I mentioned was the score of behavioral misalignment in soft clusters . This is what this drunk phrase means.
Instead of using client-based hard segmentation based on KYC data or the pattern sequence of behavior, we use collaborative profiling based on flexible clustering. We use collaborative profiling, an unsupervised Bayesian learning technique, to comprehensively analyze customer banking transactions and generate "archetypes" of customer behavior. Each client is then represented as a mixture of these archetypes (rather than as a single archetype) and these archetypes are adjusted in real time with the client's financial and non-financial activities.
Customers with a similar archetype distribution are grouped into peer groups. Different clusters have different risks, and customers who are not part of any cluster are suspect. In addition, these behavioral software clusters can accurately determine whether a client's behavior begins to deviate from what is customary for their behavioral peer group. This diversion informs the soft clustering misalignment score, which can be used to prioritize alerts-based cases or generate entirely new SAR events not detected by traditional, rule-based KYC systems.
<img class="alignleft size-full wp-image-30159" src="https://mdthinks.com/wp-content/uploads/2017/12/using-ai-and-machine-learning-to-improve-aml.png" alt=" Chart showing grouping and misalignment "width =" 1151 "height =" 674 "/>
Autoencoders for unsupervised anomaly logging
We increase the Soft Clustering misalignment score with another powerful self-learning ability called auto-encoders . These are powerful deep neural networks formed to compress transaction data and rebuild data with minimal error. In doing so, the automatic coders learn the main latent factors representing the distribution of expected data / legitimate characteristics.
When we process a new record through such an auto-encoder, the reconstruction error is calculated. The lower the reconstruction error, the more the recording conforms to the expected normal behavior and distribution. Conversely, a major reconstruction error provides a very powerful means of identifying behaviors that have not been seen before or rarely seen in historical historical data.
<img class="wp-image-32822 size-large" src="https://mdthinks.com/wp-content/uploads/2017/12/using-ai-and-machine-learning-to-improve-aml.jpg" alt=" Chart of auto-encoders "width =" 474 "height =" 179 "/>
This is essentially a powerful magnet for the surface needles in the haystack. We train these auto-encoders on cluster collaborative profiling archetypes and client behavior profile variables. When the auto-encoder finds a client with a large reconstruction error, it indicates it with a high score. Such high-scoring cases are abnormal and corresponding SARs are generated, which can display subtle behaviors from the bank's daily transaction masses.
AI is the future of the LMA
It was quite satisfying to be able to talk about our sophisticated system of machine learning and artificial intelligence analysis to identify SARs going beyond what systems based on the rules can do. The response to the interview was extremely positive. The audience was very excited about the technologies I described. This is a good sign for the future of AI in AML!