machine learning for financial institutions

Financial crime is in a surge and in this scenario no doubt that financial institutes can predict the chances of crime in advance that can reduce their exposure to it. It is easier to say but we must not forget that cybercriminals are also very advanced nowadays and they apply new technologies. This means banks need innovative techniques to stop them. Additionally, there are regulatory changes that banks must comply with while preventing cybercrimes.

Machine learning is an excellent way to provide flexibility that also manages operational costs of fraud and money laundering detection and prevention. To understand the benefits machine learning provides, it is necessary to understand the particular challenges that fraud and money laundering detection present to financial institutions.

Related post – What are the applications of artificial intelligence in finance?

Detecting crime in a proactive mode instead of reactive action

In recent days we have witnessed rapid change in fraud and money laundering patterns that financial institutions often face. As a result, they frequently struggle to adapt to new threats.  Financial institutes usually use rules-based approaches for fraud and money laundering detection by identifying patterns of known schemes and addressing them head-on. In this approach financial institutes may prevent yesterday’s threat but they leave the open threat of today and tomorrow. While criminals can adapt their tactics every day it takes months for financial institutes to catch up.

If we can apply machine learning in this context that helps to recognize suspicious fraud and money laundering patterns instantly, even when the context changes. In case of fraud and money laundering it is important to understand what the financial entity does, its operational details like where it operates and with whom, its need for the financial institute, its product details, and channels of use. Artificial intelligence in general, and machine learning in particular, play a pivotal role in characterizing identities through multiple techniques such as segment-of-one profiling, graph-based features, fuzzy matching, entity resolution, or even Natural Language Processing (NLP). After knowing the identities in play, machine learning models will learn how to, for instance, react to signals when an identity (A) is behaving differently than its peer group, or (B) is shifting its behavior unexpectedly.

Chasing the Long Tail

Another biggest problem in traditional rule-based approaches is they can face too many unique fraud cases. Fraud schemes can differ based on various criteria like –

  • Geography
  • Merchant category group
  • Payment method

The same applies to money laundering. Additionally, despite the high financial and reputational impact of fraud and money laundering, the numerical incidence per scheme is relatively low. Also, it is noticed that the incidence rate is also spiky across time, with sporadic and seemingly unpredictable variations. Money laundering patterns are not as dynamic as fraud and it is harder to confirm due to lengthy investigations. One of the potential reasons behind it is it involves multiple regions (or jurisdictions), participants, channels, and products. Furthermore, each unique case may not share a consistent set of characteristics. So, in this case, the identification often fails. Consequently, financial institutions apply manual review processes which are almost 25% of the total cost of financial crime.

Machine learning models are able to automatically learn these real-world complexities of fraud and money laundering, and greatly improve our ability to detect and prevent financial crime. They also improve operational efficiencies by reducing the number of manual alert reviews and providing KPI insights so managers can best deploy their resources efficiently and effectively. This provides a significant advantage when you consider the cost and scaling problems associated with traditional approaches. At the same time, a rule will also miss positive cases by misreading them as nonsuspicious. Machine learning can dive into big data sets with a scalpel, not a sledgehammer. Machine learning models enable a significant reduction in false positives while increasing the detection of fraudulent or suspicious activity.

Minimizing Disruption for Legitimate Customers

Though clients of financial institutions want protection against fraudulent activities, at the same time they do not want fraud countermeasures to interfere with their daily activities. It is not expected that when a customer is traveling abroad for vacation, their transactions would be blocked because they were in a different country. However, this is not unusual with rules-based systems. On the other hand, machine learning leverages tens to hundreds of attributes of each transaction and can, in turn, identify more subtleties in the data. For example, despite the fact that the customer is in an unusual location, he is staying in a hotel from a hotel chain that he favors, he continues to have lunch around noon and his average daily spending is only 20% above her typical daily average. The model will pick up this information and more when making a decision. This ensures that scenarios like the one mentioned previously happen significantly less frequently.

Minimizing the Burden of Case Analysis

When reducing money laundering, financial institutions must meet regulatory requirements while discovering new patterns of suspicious activity. At the end of the day, financial institutions want to reduce their exposure to activity resulting from unwanted actions such as terrorist financing, human trafficking, or supporting drug cartels. The goal is not just to obey regulators, but also to mitigate the risk of transacting with inherently risky sources. Traditional systems err on the side of regulations, typically overloading analysts with alerts that haven’t been prioritized. This means analysts spend as much time investigating weak alerts as they do strong alerts. This system is inefficient and unnecessarily burdensome to analysts.

However, with machine learning, not only can the algorithm prioritize rules-based alerts in terms of suspiciousness, but, thanks to unsupervised anomaly detection applied to specific customer segments, it can also recognize new, emerging suspicious activity. These models can combine hundreds of data points trained on “known good activity” to then highlight when unexpected activity happens. Furthermore, all of this can be combined with other AI-based techniques, such as graph mining, which allow both unsupervised and supervised algorithms to uncover not just transactional patterns, but also network-based patterns.

Leave a comment