@vrhax: Interesting stuff. Thanks for it. Couple of comments/questions.
I'm a big fan of providing good feedback, but I can also see an argument for being a little circumspect with information about why, exactly, a flag was triggered. This of course because it would be useful information for someone who is genuinely guilty of attempted fraud to adjust their methods. How is this normally dealt with? Is there a sort of a financial industry standard?
I suggest that it might be difficult to pick an all around number for false positives because some environments have a higher incidence of fraud and generally more persons who are proficient at it. Bitcoin is likely one of these. I could see it being the case that a system which got, say, 5% false positives might let an excessive number of false negatives through. Thoughts?
For starters, Oracle (
yes, I know, teh big evul! lol) has published a nice overview wrt AML (see
Best Practices for Anti Money Laundering (AML): System Selection and Implementation). In this day of fast paced technology, it is a bit outdated, but still a comprehensive and well thought out piece.
As for the feedback bit? I think the rather poorly written "
high risk: your transaction has been reversed" notification, should be rewritten. For starters, imho, it comes across as neither professional or knowledgeable. Yes, it is important not to "
give away the farm" so to speak. Even so, there are ways to avoid revealing "
trade secrets," while still helping the customer understand what triggered their system. But that would require them to understand how their system is implemented. And honestly, that email does not lead me to believe the writer actually understands that bit.
As for how it's normally dealt with? It varies within and between industries, of course. Take credit card companies, for example. If your spending behavior deviates from "
your norm" you may receive a friendly (
usually an automated voice system [AVS]) call from the credit card company asking you to confirm recent purchases. If not, your card will be frozen, and you will be asked to speak to a real person who will basically do the same as the AVS, with the additional benefit of actually being able to ask them what triggered the process. At which point, they will likely explain that, for example, the sudden flurry of charges and diverse locations, triggered their fraud management system [
FMS].
In the case of bitcoin transactions, my guess is that extremely large transactions could potentially trigger their seemingly "
trigger happy" system. But that is primarily a guess based upon the various reports I've read. Even so, they do face a challenge that most banks do not. And that is the irreversible nature of bitcoin along with its distributed p2p transaction implementation. In fact, I strongly suspect that is one reason they've implemented their own "
internal" bitcoin setup, which, if executed completely within coinbase, never touches the block chain.
And finally, as for false positives? This is where the AML/fraud department comes in. Their job is to investigate flagged transactions to determine whether it really meets the risk criteria. Importantly an experienced and well-trained AML/fraud team will be able to weed out most false positives, leaving the overall stats at under 5%.
As for false negatives? There really is no such thing. If a customer manages to complete criminal transactions without being flagged by the system, that would be considered a weakness in the system. And no system is fool proof and they're always evolving... as will coinbase, as will any company dipping their toe in the bitcoin waters.
Case in point (
from the dark ages). A group of convicts figured out the credit card numbering scheme, proceeded to create fake credit card numbers, and made several mail order purchases (
like I said, dark ages). There was not much that could be done from a prosecution pov. After all, they were already in jail. What did occur, however, is that little 3 digit number that is now printed on the back of cards... used to verify that yes, it is a valid card. That, along with the expiration date, and, in some cases, the customer's address, helps to reduce fraud.