I would also argue that they should augment what the government is already doing.
I would also like to point out, in the area of social impact bonds, that I think one of the original ideas was that they would work in areas where government was not already working. They're supposed to be taking areas that are currently not receiving public funding and testing out new interventions to see if they're effective. If they're effective, ideally they should continue through a more simple direct contracting method or through adoption into public policy.