On July 1st 2021, Joost van Houten, Sentinels’ CEO presented at the AI Online Seminar organised by the Nederlands Compliance Instituut. The interactive session was moderated by Eric Schuiling and covered the future of artificial intelligence and its impact on the financial industry. 

July 10, 2021 1 minute read

AI Online Seminar - The Impact of AI on the Financial Industry

On July 1st 2021, Joost van Houten, Sentinels’ CEO presented at the AI Online Seminar organised by the Nederlands Compliance Instituut. The interactive session was moderated by Eric Schuiling and covered the future of artificial intelligence and its impact on the financial industry. 

As an introduction to the subject, Joost explained the definition of artificial intelligence and touched on common misconceptions. The discussion centred on three statements that were presented to the audience.

  • The lack of innovation is the biggest threat to compliance in the financial sector.
  • Is it possible to fully automate transaction monitoring with ML?
  • In five years from now, AI knowledge will be crucial to keep our jobs.

White paper: The use of AI in AML Transaction Monitoring

The lack of innovation is the biggest threat to compliance in the financial sector 

Joost Van Houten stated that the traditional compliance methods that characterise the financial sector won’t open the doors to innovation and by simply applying the same solutions for different pain points in different industries we won’t achieve much. He noted that, although there is a degree of social fear related to AI, most people don’t realise how it’s already being used to improve our day-to-day lives. From Spotify’s weekly discovery playlists, to average expenses being calculated based on spending behaviour, AI methods are used everywhere. 

The need

Joost van Houten highlighted that an estimated 2% to 5% of global GDP is laundered. The money-laundering losses do not come from white-collar offences, but high-calibre criminals involved in human trafficking, drug dealing and terrorism. Whilst governments, authorities and financial institutions do their best to catch the people standing behind these crimes, unfortunately, too often their approaches are outdated and inefficient.

Since inception, Sentinels has been on a mission to optimise global efforts in countering money laundering. By engaging industry experts and conducting in-depth market research, Sentinels identified the key pain points the sector faces.

A key finding centred on the problems associated with case management, which forms a significant component of a financial institution’s workload. For example, 20% of the compliance employees of ABN are responsible for monitoring transactions. These professionals are frustrated in their roles as their expectations don’t match the reality of the work. Sentinels learned that a vast majority of compliance experts initially pursued their careers believing they would play an active role in combatting financial crime. Whereas the reality is, they are too often overwhelmed by reviewing hundreds of false positive alerts on a daily basis. This has led to many compliance specialists to feel inefficient and useless, leading to a staggering turnover rate: between 20% and 30% of people analysing transactions quit their job during their first year. In response to this research, Sentinels developed a goal to improve compliance workflows to improve suspicious behaviour detection.

New call-to-action

AI and ML

Joost van Houten emphasised that methods such as machine learning can significantly improve AML and transaction monitoring. He noted that Sentinels takes a pragmatic approach to develop their AI-powered software by replacing the black box models with ethical technology. 

The difference between AI and ML:

AI: “Any technique that enables computers to mimic humans intelligence. This definition includes machine learning.”

ML: “A part of AI that includes techniques that enable machines to improve at performing tasks by learning from data.”

Joost van Houten summarised three methods Sentinels uses to detect suspicious transactions:

1. Using AI to strengthen business rules

  • Cleaning and analysing unstructured data.
  • Clustering peer groups based on their behaviour. 
  • Improving thresholds based on historical data and feedback

2. ML to decrease unknown risks by using anomaly detection models

  • Supervised learning - humans classify certain categories and the machines learn from then on.
  • Unsupervised learning - a computer will look for clusters and differences in a stack of data.

Joost van Houten explained that Sentinels uses unsupervised learning to detect money laundering. By analysing large amounts of data, it is possible to highlight new patterns and detect anomalies. This is why unsupervised learning works so well in transaction monitoring. Whereas supervised learning doesn’t react to unknown data, so in the case of a new anomaly that hasn’t been classified by analysts, it will not trigger an alert. 

3. Network analysis and entity resolution - incorporating relations to identify money launderers

Network analysis is based on the principle that criminals don’t work alone. By incorporating this solution Sentinels is able to identify suspicious relations. We also know that criminals don’t just use one account. With entity resolution we are able to join additional dots that identify an actor behind several accounts. In combination, network analysis and entity resolution provides a far more detailed picture of the money laundering landscape.

New call-to-action

It is possible to fully automate transaction monitoring with ML?

According to Joost van Houten, the possibility is there. The question is: is this what we want? He noted a core approach of Sentinels is to use AI to enhance people’s work, not to replace them. The most effective solutions so far use a hybrid model (machine learning and human supervision). The efficient triaging of alerts with AI increases the analyst’s capacity to focus on suspicious transactions, whilst keeping the decision-making process in their hands.  

In 5 years from now, AI knowledge will be crucial to keep our jobs.

Joost van Houten presented, then debunked seven myths surrounding the use of AI and ML in transaction monitoring.

Myth 1.

To implement AI and ML you need to be an expert in artificial intelligence or the mathematical frameworks that are behind it. 

Joost Van Houten emphasised you don’t have to be an expert. Nowadays, with just a little knowledge of AI, you can create impact. The real experts provide toolboxes that simplify AI in order for others to use it.

Myth 2.

AI and ML deliver magical outcomes instantly.

Joost Van Houten noted that whilst AI does significantly improve the way we do things, it should be viewed more as an experiment. Over time, AI-based solutions improve and get more sophisticated.

Myth 3.

AI and ML in transaction monitoring means humans won’t be handling tasks anymore.

Joost Van Houten noted the algorithm is trained to replace some work, such as repetitive and high-volume tasks. He emphasised, ML is perfect for detecting anomalies but is cannot make the right judgement. This is what the analysts excel at –deep investigations of unusual transactions. 

Myth 4.

The more data, the better.

Joost Van Houten stated that more data brings an extra layer of complexity–not all data is relevant or can be unlocked. By analysing smaller amounts of data we can already show quick wins and results. Based on those results, we can add more data and allow the algorithm to keep learning. He noted, in a more general sense, the statement is correct, however it really depends on which phase of the development you are in.

Myth 5.

Just data and models - you don’t need anything more for AI to be successful.

Joost Van Houten noted that At the proof-of-concept stage, this statement might be true. But he emphasised, if you want to proceed further, you will need human knowledge to analyse the data, to keep it clean and up to date.

Myth 6.

Models get worse over time because of bias, which often results in blind spots.  

Joost van Houten acknowledged blind spots are a risk, and emphasised that’s why it’s so important to keep testing the models with older ones. Another option to mitigate risk is to increase unsupervised models. By doing so, you won’t steer the machine in a certain direction, but let it search for new anomalies, rather than simply the ones you told it to look for. He also noted the “four-eye principle” can also be effective in mitigating risk. 

Myth 7.

AI and ML will not be accepted by DNB as a method to detect unusual transactions.

Joost Van Houten noted that DNB (Dutch National Bank) is one of the most innovative regulators. They have dedicated teams that specifically work on creating the educational pieces and right guidelines related to AI and ML for financial institutions. In addition, DNB not only accepts but also encourages implementing ML to transaction monitoring using a hybrid model.

Q&A and closing remarks

The closing question from the audience was: How do you see the impact of AI/ML in non-financial institutions, specifically SMEs?

Joost Van Houten noted that AI and ML are applicable in many sectors. One of the examples could be logistics, where AI is applied to improve routing. 

“AI andML is not only suitable for financial institutions or big corporates. We’re a start-up, so if we can do it, why can’t SMEs?” – Joost van Houten

Joost van Houten concluded by highlighting that AI and ML are reshaping the financial industry. Although we are often unaware, the innovation is already there improving our daily activities. Implementing these new technologies in an ethical and pragmatic way has the potential to improve financial crime detection. 

White paper: The use of AI in AML Transaction Monitoring