Back
aml machine learning

Could ML-Powered Risk Models Really Spot OZARK’s Schemes?

July 14, 2020 Joseph Sibony AI Education

It takes Marty Byrde less than three episodes of television to go from white-collar embezzler at a major bank to blue-collar money launderer for the Mexican drug cartels. And boy, does he make it look easy. The money laundering schemes at the heart of Netflix’s hit show OZARK are slick and surprisingly low-tech. While today’s machine learning-powered anti-money laundering tools are increasingly effective at finding even the smallest trails to follow, Marty goes the old-school way and somehow manages to elude the law (for three seasons and counting now). 

In some ways, OZARK’s main plot seems too good to be true, and it might actually be. Marty Byrde’s money-laundering operation is actually fairly based in reality — at least as a foundation. Compared to some of the most advanced schemes out there, Marty’s low-tech approach might seem like a safer bet. However, TV rarely accounts for the powerful anti-money laundering (AML) tools banks have at their disposal to weed out crimes and risk

Risk Models

In fact, although law enforcement may have trouble spotting the crime, most machine-learning (ML)-powered AML systems can likely detect it. Let’s see why Marty’s schemes would likely not stand up to the best ML systems. 

(Disclaimer: OZARK is fictional, and we in no way condone or advocate you attempting to launder money. In fact, if you do so, you’re likely to get caught and do some hard time…so don’t.) 

Marty Byrde — money launderer extraordinaire?

Without getting into the reeds, it’s worth mentioning that while the dirty money remains in cash form, it’s hard for banks to really detect what’s going on. However, because most banks have instituted rigid limits on the amount of cash that can be deposited at once, and making massive purchases with cash raise serious questions, Marty’s — and really, any money launderer’s — problem becomes concealing the cash’s origins, converting it into clean money, and making sure it can safely enter the banking system as legitimate funds. 

It’s at this point — the entry into banking systems — that money laundering becomes an increasingly dicey proposition. With that in mind, let’s see the basic claims OZARK implies, and why they don’t stand up to the scrutiny of powerful AML and risk models

Buying cash businesses to place the money is an effective strategy 

This is one of the oldest money laundering strategies (fun fact, the term “laundering” comes from Al Capone’s purchasing of laundromats to funnel his illegal cash). It’s true that placement in a cash business makes it easy to filter in small amounts of cash surreptitiously. However, it’s a major flag that AML models detect. The problem is that when you go from no cash to lots of cash — and then to purchasing businesses — machine learning algorithms can start spotting patterns and anomalies even in the smallest, seemingly innocuous transactions. 

Think about it this way: let’s say you have a history of making standard purchases from your account, with maybe a car loan here and a mortgage there. Suddenly, you add a restaurant, or a laundromat, or a bowling alley to your portfolio, purchased in a single payment without a loan or any debt. There might be an explanation for your sudden cash influx, but an anomaly detection model will immediately flag you for further inspection. These types of abnormal expenditures rarely go undetected. 

Filtering laundered cash into regular transactions is undetectable  

Again, we run into the problem of the “eyeball test” against the power of highly trained machine learning algorithms. In the old days of laundering, you could simply spread out your laundered money by having runners deposit small amounts at separate bank branches, but this became untenable as detection methods became more sophisticated, and the rules tightened to prevent exactly this sort of laundering. 

Runners gave way to small, fake cash transactions layered in with real ones. Marty employs this latter method, using his Blue Cat Lodge (a hotel and bar) and a few other cash enterprises to quickly filter the cartel’s money through legitimate businesses. 

The idea is that as long as you don’t get greedy, and you keep transactions small, they’ll be virtually untraceable. However, an ML-powered anomaly detection system would pick up on the same transaction happening daily, with the same exact amount, in a business where other transactions rarely match up so perfectly. $500 a day, every day, for a store that handles transactions that include cents, different amounts, and more, becomes suspicious after a week. 

Moving cash around businesses and shell companies can hide it from authorities

For law enforcement, this is a real issue, as once money is caught in this web of anonymous companies and becomes layered into other organizations, it’s hard to trace back to a point of origin, especially with so many jurisdictions. So if all you want is to evade the law, then it definitely helps. However, banks are not restricted by jurisdiction — if anything, they’re in a more delicate place, since they must work with multiple jurisdictions and their AML regulations. 

For a bank’s anomaly detection system, a transaction that’s been routed through multiple other banks and shell companies in a matter of seconds or minutes will immediately raise suspicion and red flags. The problem is that money that follows unusual paths creates an anomalous flow that ML-powered detection systems flag for further review, meaning that once a machine sees the pattern, a human will as well. 

Another issue here is the use of shell companies that invoice the launderer for “services” to make the movement of cash seem legitimate. In the show, Marty buys more AC units than he needs, switches food providers to buy significantly more expensive inventory, buys more carpeting than he can use to remodel the Blue Cat Lodge and more. 

One of the biggest red flags here would be the sudden change of providers and expenses. The volume of transactions, as well as the amounts, would raise red flags when compared to historical, and even comparative data. A hotel (such as the Blue Cat Lodge) of that size spending well outside its weight class will ping most ML-powered risk models. 

Final word: Marty may be in trouble in the real world 

As clever as Marty is at evading the law on TV, he may not be able to beat the machines in the real world. It’s worth noting that law enforcement and banks have different tools, and look for different things when it comes to money laundering. Banks are often looking for these smaller signals that law enforcement won’t or can’t pursue, meaning that they can spot it in smaller signs. Moreover, banks often have to deal with multiple countries’ AML policies in order to do business, so they’re hyper-alert to the problem, and they employ machine learning tools to their fullest to help prevent it from slipping by.

As slick as Marty is in OZARK, it seems that if he attempted his schemes in the real world, a bank using advanced machine learning-powered tools could potentially spot them rather quickly, landing him in some hot water.

Risk Models

Subscribe Today! Get the latest updates with our newsletter.
We promise you'll love it.

Follow us

Just announced! Explorium Announces $31M in Series B Funding to Accelerate Growth Read more