The 80% figure is a rough estimate based on personal experience and borrowed from the Pareto rule, of which one of the interpretations is "80% of everything is crap". In fact, the original crap rate (introduced by Sturgeon in his law) was 90%, which is probably more accurate.
Ged replied with some actual numbers where the success rate for software projects is 16.2% rather than 20%. Now any statistical data can be disputed ad infinitum. Good luck even defining what constitutes "success"!
But let's say that success largely means achieving the originally set objectives. Of course, Columbus failed miserably in finding a path to India when he actually discovered the New World (discovered for Europeans, that is :)). Now how often the original plan works out as intended? And how often the planners are reasonably happy with how it did work out?