I'm not an expert in R&D but I'm going to make an educated guess that more paths lead to failure than success. So that type of hit rate probably isn't terrible.

I certainly agree with you that if the research money is being doled out by human beings instead of a non-existent, hypothetical and probably impossible AI system then political considerations and all forms of corruption are leading to a significant portion of the funds being allocated inefficiently at best and likely leading to boondoggles.