This topic stems out of this related thread, where we're currently having a semi-heated discussion about the quality of SE's RMT-detection algorithm. I think it's an interesting topic, so I want to split off and draw in some more opinions about it. I'll post some of my own ideas, but I'd really like to see what others can come up with as well. Hopefully some good ideas will come up in here, and there's always some small chance that someone at SE will read over this information and improve their detection methods as a result.
Imagine that you have access to a database with all character data, all POL accounts linked to them (including credit card info), and logs of all actions ever taken in game. With the ability to look absolutely anything up, what do you look for to find accounts that are involved in either buying or selling gil? Don't worry about trying to find some way to hit everyone in one sweep, if you think you have a good method to catch one particular type (fish-botters, for example), that's a good thing.
What specific criteria would you search for? Does your criteria guarantee that players matching it buy or sell gil, or does it just make them suspicious? Can you combine multiple criteria to be totally sure, or at least make it very likely? Once you've found a suspicious-but-ambiguous character, how could you look deeper into that specific character to come up with a final decision?
On the opposite side of detecting RMT, it's also very useful to consider criteria for detecting legitimate players. With this sort of criteria you can exclude people, and shrink the set of "potential RMTs" to use your other criteria on.
As discussed in the original thread, "false positives" (incorrectly labeling legitimate players as RMTs) should be minimized as much as possible. Incorrectly banning a real player is a major mistake, much more major than failing to ban an RMTer. Make sure to consider special cases like linkshell bank characters that have a high gil-throughput, but are legitimate.
Thanks, looking forward to seeing what people can come up with.