I would actually put it 60-80%. As there is plenty enough of ideas that should work. It is not really random search, but applying existing knowledge or expectation to some different area. If doing one thing with one thing works, I would fully expect in many cases that doing same thing with this other thing relatively closely related to also work.

Lot of research is iterative process and there you can do iterations that you know could work and already ignore things that certainly won't.