How a Little Data Science & Machine Learning Was Used to Win a Fraud Case
From 2007 through Spring of 2012, a company I founded, Canadian Venture Partners (CVP) filed a legal claim against a Canadian POS services provider which I’ll refer to as Company X or CX for short to protect their anonymity.
In June of 2007, CVP obtained a judgment against CX in the amount of $687,190 CAD. During the collections process, CX claimed that it had no money or assets to pay the judgment. Upon analysis of records obtained during the judgment debtor examination (JDE) of the president of CX (Mr. CX), I concluded that CX had transferred its merchant client base to another company controlled by Mr. CX. In Canada (and I suspect similarly here in the US), if a transfer of assets is done without consideration for creditors, it is referred to as Fraudulent Conveyance. This initial analysis was compelling enough for me and CVP’s legal council (LC) to file a fraudulent conveyance action (FCA) against Mr. CX. and his accomplices in the summer of 2007.
In order to successfully litigate a FCA as a plaintiff, two important criteria need to be met:
- First, plaintiffs must prove that assets were fraudulently conveyed by the defendant in order to unlawfully avoid payment of obligations owed the plaintiff(s).
- Second, after establishing that fraudulent conveyance had taken place, the amount of the fraud must be quantified. This second criteria is necessary in order to establish a basis (referred to as the quanta in legal parlance) from which the judge can decide on the amount to award the plaintiff(s).
This initial analysis was focused on the first criteria of building the evidence base establishing a solid argument that fraudulent conveyance had been committed while accomplishing two important complimentary objectives. The first was to convince CVP's LC to take CVP’s case based on contingency. This was necessary because neither CVP, myself, or any of the co-plaintiffs had the resources to finance a multi-year lawsuit conducted in a foreign country (Canada). The second was to establish a solid base from which a future analysis primarily focused on the second criteria of quantifying the amount of the fraud could be conducted.
This project was initially done in Java, but is reworked here in Python. It is used to demonstrate a complete analysis pipeline from the data cleaning phase, through the analysis phase and finally to the results and conclusions phase. The second criteria of establishing a quanta described above was actually a lot more work, but it was not as interesting from a data science perspective. As a result of this work, we were able to settle with the defendants for an amount much larger than the initial $687k judgment amount and I have been left with a deeper appreciation for the work regulators conduct to prosecute corporate (e.g. Enron) and Medicare fraud.