As economics change in the world, whether political or through major events such as the COVID pandemic, they will always bring new challenges in combatting financial crime. The industry must adapt to all these challenges, reducing new opportunities for criminals as quickly as possible.
Traditional methods and legacy technologies, however, can be big inhibitors to progress.
AI-driven regtech is now providing banks with a faster, smarter way to gain a more holistic view of customer behaviour, reduce false positives, and reduce costs in the process.
Starting with AI
A great use of artificial intelligence is to identify patterns within large sets of data – way beyond human capability, or that of legacy technologies – and then use the knowledge that it has learnt from looking at these patterns to make recommendations. This can be easily translated onto the role of AML investigations, making AI an important ally in the fight against financial crime.
In this case, AML investigations are required to look at large amounts of transactional data to try and identify and analyse patterns. From these patterns they must look for the anomalies that could be representative of criminal activity, and then make recommendations to mitigate the risks.
In this way, AI can carry the burden of triage in AML, helping the human analyst to then focus their attention on only the pertinent and true risks.
There are many useful applications of this technology in regtech. Three that we will look at in more detail are: client activity reviews; transaction monitoring systems; and the reduction of false positives in client screening.
Client activity reviews
Banks and financial institutions need to review customer activity on an ongoing basis to ensure that what the customers say they are doing corresponds to what they are actually doing. This applies to all transactions, activities, and events throughout the lifecycle of a customer within any particular financial institution.
The challenge analysts face is that there are huge amounts of data required for client activity reviews and it is often split across different silos with a lack of cohesion between systems and departments. This makes it very difficult to consolidate all the information into a single view and present it in a way that can help analysts make quick and accurate decisions about any suspicious activity.
An AI engine can look at the transactional activities of customers much more effectively than a human. It will use machine learning to analyse historical records to identify the patterns that represent the typical activity of a customer. From this, it can also learn what good looks like and what bad looks like, based on any anomalies in the data. This information is presented back in a simple summary, including specific details on why the data are unusual. This information enables the human analyst to make a decision on which are the key data points that require a deeper investigation.
It’s an approach that helps to increase the automation of customer reviews, speeding up risk detection and reducing the overall cost.
Transaction monitoring systems
Transaction monitoring systems are required to detect potential suspicious activity or money laundering risks. Many of the legacy platforms are built on static rules that can only analyse slices of data to see if they are consistent against manually set thresholds, e.g. if the amount of the transaction is higher than the set threshold, it requires further investigation.
However, what happens when the amount is very close to the threshold? Or if there is unusual activity but with lower amounts? With so many different scenarios and combinations, a simple set of rules may not detect all the risks.
Machine learning techniques can be used to augment the process and create a more robust and efficient system. The AI will analyse all historical data to identify patterns of behaviour that can better predict or detect risks beyond what might be identified with the simple rule system.
The rule-based system will set a threshold, typically using a scoring system to determine alerts, while the AI will add an extra layer of analysis to produce a challenger score. It will report the probability that a group of transactions may be inconsistent against patterns of previous activity, helping to detect the unknown unknowns.
False positives reduction in client screening
One of the most common issues with client screening is the huge number of false positives that are generated using processes that are often reliant on legacy technology. These processes might involve things such as checking the names of customers against sanctions lists, for example. There might be multiple policies and processes, or multiple matching algorithms, and with all of these at play it makes it very difficult to explain eventual outcomes.
By incorporating machine learning into the client’s screening process, agents can be trained to use previous information to learn why a specific risk should have been discounted. Using all the data points the machine can extract from the false positives, it can then train itself to predict the probability that similar false positives in the future should be automatically discounted.
So, again, gains are made with increased automation, more effective triage of actual risks, and a more robust process that reaches solutions much quicker than we have been able to before.
About Luca Primerano
Luca is the Chief AI Officer at Napier and has extensive experience in decision automation and digital transformation. He specialises in anomaly detection, data correlation, and pattern identification. Luca has worked previously with Goldman Sachs, Deutsche Bank, and Deloitte.
About Napier
Every day we send out a free e-mail with the most important headlines of the last 24 hours.
Subscribe now