That's the whole concept before something can be concluded thus it's complicated, but once it is concluded you might be surprise that it's required only less than 100 lines of code for the logic (not counting those supporting things for the bot to work i.e: talk with server, catch errors, etc..) and if it work for freebitco.in with 5% edge, it will work on any site since 1% or 2% is max.
For this case it includes both: logical calculation & personal experience. His bot initially perform not very effective (hit a rare streak once every two or three weeks) but after sometimes play and together with input from other users it achieve a near perfect result (hit a rare streak once every two or three months). So it can be concluded: research to have logical calculation (betsize, odds, iteration of random events/sets), resolve the "spike" or "skewness" of distribution of numbers (rolled results), test as much as possible to have accurate statistic. Sum of few dollars I would say it is a no, the minimum balance at least should be 0.5BTC (total willing to lose) for around 70% win chance but for better chance to win in the long term you need 1 -> 2BTC (almost you will never lose 24/7 plays with minimal/safest settings).
Otherwise as the site mod said, majority of them lose, that's because of "not enough balance". In any case, play with test mode first to understand how it work, if you managed to have enough patent to run the test 24/7 for few months or even half a year you will understand much of it and have enough confident on playing for real (required balance/ total earning / chance of bursting at specified balance). The reason for test that long is to find out how long it is for the question "you will lose in the long run".