As a research mathematician, you are supposed to have some aspect to what your are doing, however possibly narrow, but still important enough to worth noticing, which is superior over what every other mathematician in the world is doing. There is no such requirement in non-research occupations.
If you are baking bread, your bread does not necessarily need to be the best bread in the world. Not even any particular narrow aspect of it needs to be the best in the world. Your bread does not even need to be the best in your town. The producer of a better bread would just price it higher. If you are a doctor treating people's kidneys, there does not necessarily need to be an aspect of treating kidneys that you do better than every other doctor in the world. If you are a lawyer advising your clients, you do not need to better at some aspect of it than every other lawyer in you jurisdiction. Those more skillful lawyers just have their clients, and you have yours.
But as a research mathematician, you are useless if you have not excelled, literally, over everybody else in the world, at least in some respect. It is in this sense, I think, that you either succeed or fail.
Thus the name of the game in research is that you are supposed to understand something that noone else in the world understands. And betting your fortunes on the proposition that you are in some way smarter than everybody else in inherently dangerous. You take some route which everybody believes is leading nowhere, but you know better than everybody else. You tackle a problem that everybody considers to be meaningless or unsolvable, or you approach a problem in a way which everybody thinks is clearly doomed to fail, but you see things deeper than they do.
Now, what is this notion that there are some certain, sure ways to excel over everyone else in the world? It is an illusion and nothing but an illusion. A researcher taking one or another dangerous route in one's research career -- what other people would consider to be a dangerous route, based on what they know and understand -- may indeed succeed or fail, depending on the quality of his vision of things. But a researcher systematically avoiding all such dangers is literally doomed to fail. There just isn't anything he can possibly succeed in.