September 19, 2018 | Written by: Sam Kalyanam
Categorized: IBM RegTech Innovations
Share this post:
In 2010, I was traveling with my son, when his name was flagged for a partial match to a suspicious person by the US Transportation Security Administration (TSA). The TSA agent, dutifully asked if he could have a word with my son to sort out this issue, to which I replied, “Be my guest.”
My son was just 18-months old at the time, so the conversation didn’t go too far.
Another relatable example of the same phenomenon is search results. If you ask Google or Bing to show you “images of all animals except elephants”, you will be shown exactly the results you are not looking for. Don’t believe me? Check it out here.
Elephant in the room
These situations are fairly common, and for those who share a name similar to someone flagged by a government agency, like in my son’s case, constant revaluation is required – no matter how many times those same agencies clear you.
The point here is that, despite the advanced technologies available today, we still rely on blunt methods to solve complex problems, gain information, assess risk and make informed decisions. And because we accept the limitations of these tools, especially free-to-use platforms like search engines, we as users do the work of adapting our query to improve the accuracy of the output, rather than expecting the system to adapt and learn over time.
You can certainly live with a bunch of elephant images that are returned wrongly for the query you made. But if you are applying the same methodology to your current Anti-Money Laundering (AML), Know Your Customer (KYC) and Customer Due Diligence (CDD) reviews across millions of customers, you are perpetuating a vicious cycle of wasted effort and wasted resources for your financial institution. Even worse, you are damaging the morale of your analysts by forcing them to make sense of this highly inefficient and ineffective system.
Lack of learning hurts customers, banks and birthdays
As you have seen in the examples above, little is done to “learn” from past decisions. Instead, the same logic is applied again and again generating, unsurprisingly, the same inaccurate results.
One simple example of this lack of learning is looking at behavior that seems risky on paper, but is not only perfectly normal in context, but also quite predictable. Let’s say you send your mother in Nigeria $200 every year on her birthday. Looking at your transaction history, a human could easily spot that pattern and, after some initial questioning, establish an exception for that repeated action in the years to come.
But many existing systems have rules that they apply without exception and often without context. As a result, your transaction gets blocked until you, once again, explain that you are just being a generous son or daughter. In the meantime, you miss her birthday and one of your siblings gets all the credit for just mailing a postcard. Your disappointment eventually translates into frustration towards your financial institution mired in outdated approaches to financial crimes management. Time to look for another bank.
Replace the vicious AML cycle with a virtuous one
The projected growth in transaction volume over the next few years and tightening of risk thresholds by regulators will continue to force financial institutions into trying to “do more with less” – a practice that has been widely used since the recession a decade ago. But using the same outdated systems and getting the same subpar results is not sustainable.
Techniques such as entity link analysis, peer group segmentation, robotic process automation and machine learning each provide a modest improvement in AML investigation accuracy and efficiency. But even when used individually, these techniques do not yield the comprehensive insight promised by artificial intelligence and learning from past outcomes.
Instead, these techniques should be combined into a layered analytical approach, each providing additional insight and context, to help financial institutions resolve unwieldy volume of AML alerts with precision and confidence.
UK bank uses layered approach to curb false positives
One client we worked with had this same issue. They had grown significantly through a series of acquisitions, creating a large, complex institution with an overlapping customer base (i.e. a customer could have accounts in multiple historical institutions.) This combined institution had inconsistent data and little understanding of the connection between customer relationships, for example, making an innocent transfer of funds between an individual’s savings accounts look like potentially suspicious. The result was 99% false positive alerts, duplicated investigation effort and simply wasted time.
Using a variety of strategies, like entity and network analytics to understand customer identity and relationships, as well as machine learning and statistical analysis of previous outcomes, we worked with the bank to prioritize legitimate risks and reduce redundant effort.
The outcome was a 70% reduction in false positives and 50% reduction in false negatives, meaning higher quality alerts and time savings by avoiding duplicate investigations.
Reversing the decision funnel
These results are typical of the new direction in managing AML and other financial crime programs. Instead of systems making uninformed decisions and analysts spending time collecting information to support them, now systems are gathering corroborating or exonerating information to provide analysts with greater context in final decision making.
This is augmented intelligence, where technology does the heavy lifting and mundane tasks of data gathering, enabling analysts to focus on higher value tasks that they perform well, like investigations and evidence-based decision making.
The growth in transaction volumes and push for greater customer scrutiny by regulators should be a wakeup call to the financial services industry that this new approach to AML is not just a “nice-to-have”. Institutions can no longer keep repeating the same actions and wasting efforts. They need tools and processes that can adapt, learn and accelerate the ability to make the right decisions. Whether or not you have already made investments in AI and cognitive learning, practical application of those tools is taking flight. You can either get on-board with those now or watch your peers accelerate on their journey.
Want to learn more about the future of AML? Join IBM experts at the Sibos 2018 Conference in Sydney, Australia and visit the IBM stand (#F22). Or attend one of our various speaking sessions to hear from and chat with our experts.