Frank Erdle/ July 1, 2019/ Uncategorized

The fight against fraud, money laundering and sanctions violations is becoming increasingly difficult. Machine learning enables financial institutions to detect suspicious transactions, reduce false-positives, and avoid the risk of penalties.

The risk of abuse by criminal organizations is higher than ever for banks and other financial service providers. Tax evasion and bribery are committed primarily in addition to embezzlement of assets and terrorist financing. The compliance of politically motivated sanctions is another challenge for the financial sector. The use of Artificial Intelligence (AI) can provide valuable help to accomplish those tasks. Unlike classic filtering systems, where experts dictate rules based on known patterns, an AI-based system uses example data to solve complex problems. With AI support, it may even be possible to discover a money laundering intention before it happens.

Current strategies are unsatisfying

According to a study, banks in Germany spend more than $46 billion a year on compliance designed to prevent money laundering. Nonetheless, the industry must cope with high false positive rates in its fight against criminals. At this point, AI comes into play. Abnormalities can be identified with pattern recognition in the course of business. For this purpose, customers, who are suspected of money laundering and show a similar behavior, are grouped into digital clusters.

Another promising application is Alerts Management: Employees who evaluate business anomalies are supported by an AI-based system so they can focus on the most important cases. In the third step, rule-based alert generation is taken over by an adaptive AI variant. With this method, the number of critical cases that must be processed manually can be reduced by up to 80 percent!

Companies increasingly using machine-based learning methods such as predictive analytics for plausibility checks. For the scope of compliance, areas such as fraud prevention, anti-abuse and credit scoring are best suited. Initially, computational models are generated from existing data, with which the amount of future transactions becomes predictable. Activities that differ significantly from this forecast are marked. As a result, abnormalities can be identified early.

Hybrid models meet the requirements of BaFin

In a position paper, BaFin demands that decisions based on AI must be comprehensible and reproducible at all times. Decision makers should therefore rely on a hybrid model of rule-based and AI-supported procedures. Another scope of application is the automation of standard processes. Significant cost savings can be achieved in embargo and sanction inspections.

What requirements do banks have to meet in order to use intelligent compliance tools? Generally, high quality data is essential to train AI systems. A technology partner with the necessary know-how and development platform for AI applications is essential. Then the customer can work productively with the new solution right from the start.

* Quelle: Studie von Autonomous Research – „Augmented Finance & Machine Intelligence“:

Frank Erdle

About Frank Erdle

Since the beginning of the Internet age Frank Erdle has followed the digitization as a journalist. His main focus is set on the topics of Digital Finance, Artificial Intelligence, security and sustainability. Erdle is not only interested in economic opportunities that arise from new technological concepts, but also examines their effects on society.