THE SUPPRESSION OF TRADE-BASED MONEY LAUNDERING THROUGH MACHINE LEARNING AND AI

Objectively speaking, trade based money laundering is only superficially understood by the global compliance community. Most compliance officers are not aware of the tradecraft involved in moving millions of dollars in criminal proceeds, in the open, within the massive amount of legitimate international trade that must be promptly paid for, in order to keep the wheels of commerce turning. If a bank does not identify a transaction as TBML, it passes muster and is approved.

The extent of general knowledge of programs to assist compliance officers in the identification and interdiction of TBML is unfortunately quite limited. The extremely useful International Trade Alert https://www.internationaltradealert.com, which identifies any transaction where the price of goods is more than 5% off current market prices, only catches first-generation TBML; it does not alert the user when any of the more advanced, and even esoteric, permutations, combinations and variations that do not need to engage in price manipulation to move the proceeds of crime abroad are employed, and these techniques are rarely known to a compliance officer without advanced training and experience.

For example, some money launderers move money under the guise of a legitimate international trade transaction by using several different banks, located in different cities, who are fooled into sending payment for the same shipment. All the trade finance institutions are ignorant of the fact that documents for a sole transaction are being proffered to them, while only shipment, not five, is actually being transmitted overseas. Since the sales price is within normal parameters, the other four banks, having received the same shipping documents, with clever modifications, do not red flag the transaction.

 If someone actually checks, they can verify that the shipment has gone out, but they do not know about the other four payments. We are simplifying the fact patterns somewhat, and omitting document alteration details and other illicit acts, but multiple payment for a single shipment, phantom shipments, and similar sophisticated  techniques, such as sending extremely valuable goods and categorizing them as low-cost goods, and then selling off the goods abroad for cash, are used every day. There are a large number of modified TMBL techniques that are currently working exceedingly well, and are not being uncovered by the platforms in common use, which are rules-based programs.

Rules-based programs, which are by definition inflexible, simply do not account for the imaginations and ability to adjust & innovate that experienced money launderers possess; they are reactive when your platform cannot be. The shortcomings of rule-based TMBL systems:

(1) Rules can only cover known money laundering typologies.

(2) Rules are, by their very nature, blunt and imprecise tools.

(3)Regarding departures from a profile, rules can only work if based upon the unusualness of a known data point at a time.

(4) Rules are rigid.

(5) Rules are notorious for producing a very percentage of false positives.

Platforms that employ supervised and unsupervised machine learning, powered by artificial intelligence, are effective in identifying, and interdicting, in real-time, TBML. Rules do exist in such programs, but it is the combination of rules and machine learning that can detect TMBL where traditional programs cannot. 

This is intended to be only an introduction to the subject. Readers who wish further information on this topic may want to read an interesting White Paper which expands upon the material covered here, entitled The Challenges of TMBL; if you email me at miamicompliance@gmail.com I will be happy to send it to you.