Two of Napier’s founding members, Nick Portalski and Luca Primerano, recently spoke to Ross Andrew Hurley of EM360 about problems in identifying money laundering, including the complex issues regulated entities are encountering.
In this blog we look at some of the big, recurring themes broached in the podcast in more depth, and discuss how Napier’s tech offers a solution.
You can listen to the podcast here.
The challenges in detecting money laundering
Organisations of all sizes face multiple challenges associated with identifying money laundering.
Lack of AI accessibility and useability
Until recently, AML systems benefiting from artificial intelligence (AI) and machine learning (ML) could only be utilised by organisations employing data scientists to help them use and make sense of their system. This was – and still is – a big issue.
Lack of AI accessibility and useability has had two main impacts: firstly, it limits the appeal of AML systems using artificial intelligence. This has the knock-on effect of ultimately limiting the number of companies deploying the most sophisticated technology to fight financial crime.
Secondly, for the pioneering companies that did adopt artificial intelligence, they needed (and often still do need) a team of suitably qualified analysts and data scientists to not only interpret the outputs but to make adjustments, such as adding new rules. This dramatically escalated the cost and complexity of compliance and forced a division in skill.
Compliance officers and data analysts must be able to understand the outputs of their AML system and take any appropriate action, from deep data dives to rule adjustments. This concept of being able to understand AI is called explainability and Napier has been working to make it available in its Intelligent Compliance Platform.
Data is integral to any successful AML system, but many organisations face significant challenges relating to its quality, quantity and storage.
Data is often scattered across multiple systems in different countries, and isolated in silos. It can be very difficult to access the right type of data at the right time. Many organisations also struggle with the sheer quantity of transactional and unstructured data that’s processed daily.
The performance of any system will only be as good as the quality of the data at its disposal. Data is the new oil. It makes sense that the more data we have on customers, the better the risk identification.
Without continuous access to good quality, up to date data, it becomes increasingly difficult to generate an accurate customer risk assessment. High risk customers can therefore inadvertently be overlooked.
Artificial intelligence in particular needs lots of good quality data. There are many factors that can affect data quality, from source to formatting to age. Poor quality data will ultimately lead to inflated levels of false positives and false negatives, which are hugely expensive to process and expose unnecessary risk.
Once an institution improves its data quality, it will then need a data platform that aggregates and pulls in data from multiple sources which would give them an ongoing 360° view of their customers.
Often there will be multiple functions in an organisation working on a single case that spans different departments. It’s not uncommon to see large financial institutions battling with hundreds of different systems.
Messy, tangled AML processes are never an ideal scenario. They’re inefficient and can lead to data loss and unnecessary risk exposure. When tech isn’t working as well as planned, more people are often thrown at the issue. Unfortunately, this isn’t the answer either. It’s estimated that the financial services sector spends $180bn on AML, more than 60% of which is on people. But with less than 1% of criminal profits being recovered, there is clearly scope to do better.
A step in the right direction would be to integrate multiple compliance solutions, such as KYC and AML, into one master dashboard. This could improve efficiency by allowing automated reporting and intuitive workflows that make better use if a compliance officers time.
Combining legacy technology with in-house software is another breaking point. Many financial institutions are using aged technology and systems which simply can’t keep up with the sophistication of today’s financial crime. And the issues not only lie with legacy systems. There has historically been a perception that AI is a panacea, but this isn’t the case – general AI systems can also be problematic.
When we yet again consider that less than 1% of money laundering is detected, we can infer that 99% of the alerts that banks are processing are false positives. There is also the colossal issue of false negatives. The majority of risk is hidden in data and processes that are overlooked by legacy systems.
Moving away from legacy technology is necessary but when it comes to adopting AI for AML, it’s important to know that a one size fits all approach is never desirable. As more organisations have become aware of this, we are now seeing AI aimed at the specific problem and implemented with greater awareness and understanding of how the AML system will deliver value.
Simple steps to improve AML systems using AI
An effective AML system will not only help to achieve better criminal profit recovery rates; it can help regulated entities to achieve a competitive advantage. Nick and Luca recommend the following:
1. Optimise transaction monitoring
Aside from getting data in order, this has to be one of the first and most basic steps in improving AML systems. Transaction monitoring needs to be of the best quality for artificial intelligence and machine learning to be able to extract value and detect risk.
2. Make AI an integral part of your compliance system
AI is no longer an elusive technology for the data scientists of the world. It can now be simple to use and understand, and it’s advanced capabilities in detecting suspicious activity means it should be integral to every AML system. AI will improve process efficiency and provide an all-important different point of view to improve customer risk assessments. Crucially, machine learning agents can run through and extract information from billions of transactions far more quickly than any individual or rule-based system.
3. Channel AI to focus on a specific problem
To work effectively, AI must be applied to solve a specific money laundering problem, pattern or concern. Not all transactions should be monitored equally; machine learning should focus only where the real risk is.
4. Ensure explainability
Whatever AI system you use, explainability is essential. Explainability means the system explains the outcomes/findings/alerts in a single, easy to understand way. This ensures all users can make an informed, fully accountable decision with a full audit history.
Future AML expectations and predictions
The amount of money that criminals are laundering is going up significantly, as is the number of organisations that fall under the ever demanding AML regulations. We expect the future to be even more risk-driven as it becomes paramount to identify which customers represent the greatest risk, and how often that risk level should be reassessed.
Ideally customer risk assessments should be continuous and pull in third party data in near real-time. Napier offers this capability with its Risk-based Scorecard.
Perpetual KYC, which builds a richer picture of customers by closing the gap between KYC processes and transaction monitoring, should also become standard practise.
Finally, if the world is serious about tackling money laundering then there will be a point when all organisations will need to feed their data into a central provider. An integrated dataset between all banks would have to become a reality. But right now, we’re a very long way from that.
Be the first to know about Napier’s podcasts