October 11, 2016
Why fighting fraud and financial crime isn’t just another machine learning problem
Most of us don’t think too much about the numerous automated decisions made for us every day - every Google search, every Uber request, every Facebook post you see - these are all machines making decisions based on your data to find the most relevant results, the best route or the most entertaining post possible.
Machine Learning and Circle
At Circle we take a similar approach in developing our risk and compliance controls. We leverage state-of-the-art machine learning algorithms that take into account millions of data points every hour to provide probability scores that are used in order to decide in real-time whether to open an account, allow a transaction go through or certain activity to continue. These models aim to detect the outlier population - those fraudsters, money launderers and other actors with malicious intent, and separate them from our perfectly legit customers.
At this point you might ask - ok, sounds nice but not very different than what many other tech companies are doing, everyone is using machine learning these days, what’s so special about Circle’s models?
Risk Detection - a Unique Machine Learning Domain
When others such as Google predict based on your historical data which search results will be most relevant to you, you and they both aim for a similar goal - getting the most suitable results. Sure, you might delete cookies because you don’t want such companies to have so much information about you, and many people use proxies and such, all of these elements might impact search or marketing models and cause results to be a bit off. Still, in general no one has a strong incentive to make a Google model take the wrong decision by deliberately behaving in a certain way.
This is very different in our case of course.
Circle risk models have to be sophisticated to understand that a person, doing whatever he can to seem like a legit customer and behaving like any other, has actually stolen this credit card. Circle compliance risk models have to be sensitive enough to the fact that a certain transaction or behavioural pattern, which on the surface seems normal, is most probably a financial crime scenario. Our models need to make the right decisions for our customers, whilst those outlier folks do whatever they can in order to try and mislead us and them. It’s a tough job. Ever evolving, improving and reengineering our models is imperative in order to stay ahead of those bad actors, who, just like us, use data for decision making and learn from previous mistakes - making their behaviors and patterns harder and harder to detect.
This means that risk and compliance functions are traditionally in an ongoing fast-paced game of cat and mouse, whac-A-mole or attack and defense, choose your favorite metaphor. At Circle, we aim to be one step ahead of fraudsters, scammers and money launderers at all times, and take an active approach toward building risk and compliance tools, processes and infrastructure in order to ensure that. Quite a challenge, but one that excites us and helps us make Circle stronger and safer every day.
Join the Team
We are always expanding our Circle and are looking for bright team players that will help us stay ahead. If you’re excited and energised about leveraging data in smart way and fighting fraud and financial crime, why not take a look at our current vacancies? https://www.circle.com/en/careers