The current way banks are handling money laundering alerts is not sustainable. If they do not want to continue rapidly expanding compliance teams each year, they need to look towards automation technology.
Quantifind was founded over a decade ago by Ari Tuchman and John Stockton. The pair met as fellow quantum physicists with a passion for extracting signals from data. The aim of their new company was to bring hardcore machine learning to unstructured data. Over the course of the past ten years, Tuchman and Stockton have focused on building best-in-class technology models, tech stacks and distributed database solutions aimed at the unstructured data world. During the past six years, the company has had a laser focus on anti-money laundering and anti-fraud.
The motivation to focus on the money laundering problem is based on the fact so many incidents are not caught. Tuchman said, “Different people quote different stats, but everyone agrees that it’s a low number.” It is tough to assess the exact scale and impact of anti-money laundering, but a report from the United Nations claimed the measures have minimal impact. Based on data from 2009, it claims criminals keep around 99.8% of their money.
A reason for this is partly due to a resource problem for banks. Banks are businesses and not law enforcement organisations, Tuchman stated. Resources are finite, but the number of investigators needed to meet the need is growing every year.
Another aspect of the challenge is how fines are levied. “Fines on large organisations tend to become just a cost of doing business to them; another risk among others that they manage.” However, Tuchman explained things are changing, such as with some fines targeting individuals. Fining members of the leadership is a whole other level of deterrent.
To that extent, regulators are becoming more demanding of banks and increasing the amount of regulation. Operational costs on these compliance teams within big banks are increasing around 20% year-over-year, making it the largest increase cost of any department, Tuchman explained. A recent study from Duff & Phelps claimed that a third of financial services spend more than 5% of their revenues on compliance.
Tuchman added, “When you’re in a low-interest rate economic environment, spending that much on compliance becomes something that’s just not tenable.” Banks can’t just keep hiring more and more staff; they are already inundated with alert backlogs that they struggle to get through. The only viable solution is to leverage AI to help automate the process. While humans are going to be more insightful at finding things, you need thousands of people around the world to do this, and turnover is high. It’s pretty far from the core business of a bank.
The number of transactions is exploding, and with it are the number of risks. FinCen and OCC have made it clear they don’t want the number of alerts diminished. Banks need to somehow process the influx of alerts coming from these transactions, and ideally to prioritize them by risk. The way banks traditionally handled their backlogs would be by hiring consultancies for millions of dollars and have that team manually go through everything. However, it still takes six months to complete and the backlog just builds up again. “That is not sustainable. Technology is needed to clear that backlog.” Quantifind, for example, is capable of clearing 50,000 alerts overnight. This allows the team to then focus on the most challenging and risky alerts.
How does adverse media screening come into play?
Another area where regulations are somewhat ambiguous is with adverse media screenings. This process involves searching for negative news about a person or business. Tuchman stated it is not clear whether there is a legal requirement to conduct this search. On one hand, the AML legislation does not specifically state a bank must carry out these types of searches. On the other hand, banks have been flagged for doing business with some obviously bad actors, where an adverse media search would have made it clear they were not someone a bank would want to have as a client, Tuchman stated.
“Banks are being implicitly told that you need to know what’s going on in the outside world. Even if someone is not flagrantly suspicious, with funky cash deposits and bags that smell like marijuana, they need to know who their customer is. There’s no practical way to do that without adverse media. So I do think there needs to be more clarity.”
Banks currently do not have the risk assessment accuracy to monitor their millions of customers every day, so they focus on the biggest risks. However, there is also obscurity of what classifies someone as a high-risk individual. If a company only has time to complete these searches on high-risk people, they need to know what that term means. It is obvious that if someone is involved with terrorism, they are high risk. But those caught of tax evasion also fall into the high-risk category.
Technology can ease this customer surveillance by helping to assess the alerts that come through and only escalating ones that need attention. This can reduce the daily alerts to a couple hundred, compared to hundreds of thousands. This boosts the operations of the team and can even reduce the amount of false positives being generated.
Trusting the technology
Whenever using technology to complete a task, there is a big question mark placed over trust. How can the bank be sure the technology is getting things right and not making major mistakes? Quantifind builds trust through random sampling testing. This enables a team to take 1,000 alerts out of that pool of 50,000 to monitor accuracy. In the example of clearing backlogs, this shrinks the six months of work screening alerts down to just a few days’ worth of work. Through the sampling, the team can also see if the AI platform is generating 95% accuracy, like it should.
Gaining trust in technology is a process. However, even companies that are hesitant to embrace automation can implement it for low-risk tasks and still gain a lot of operational efficiency boosts. For example, a bank can use automated decisioning to say if an L1 investigator should escalate an alert. This removes a seven to 12 minute process. It might not be a lot, but it adds up over the course of a week. If this is done across the whole team, it can save days’ worth of working hours.
This is a low risk way to save time by intelligently fast tracking alerts. This allows all businesses to find a level of automation they are comfortable with. Obviously, manual oversight will be needed with a lot of investigations; looking into network effects, link analysis and more. But this technology helps them spend more time on those tasks. Tuchman said, “if everyone is consumed working 10 minutes at a time on all the low-risk alerts, they won’t have the time or focus to uncover the most challenging cases; the real human trafficking rings or the real shell companies that are proliferating terrorism.”
Tuchman stated Quantifind helps individuals increase efficiency by 40%, which spread across a team of 1,000 is a substantial increase in productivity. Implementing a platform like Quantifind does not mean banks will shrink the size of their compliance teams, it simply means they don’t have to keep increasing the department’s resources by 20% each year.
The Anti-Money Laundering Act 2020, which many have considered to be the US’ biggest change in the space since 2001, is recognising the need for automation. The regulation is also encouraging the adoption of machine learning and AI, as well as supporting the technology by establishing a subcommittee on innovation and technology in the Bank Secrecy Act Advisory Group. While this is the right direction, Tuchman sees opportunities to provide additional clarity. For example, it does not outline what machine learning technologies are good and what the new subcommittee will do exactly.
Furthermore, some banks are taking a conservative approach as they onboard machine learning, due to the complexity and burden of model governance. Tuchman explained regulators expect a bank to spend a year undertaking the model governance review, which can be expensive to run and there is no guarantee the bank will pass. Regulators are moving in the right direction, but there is still more work to be done.
To hear more insights from Tuchman, he will be speaking at the Global RegTech Summit.
Copyright © 2021 FinTech Global