The sense I've gotten so far is that given a causal model with non-controversial causal assumptions, you can do algebra in some cases to come up with conclusions that otherwise (in the absence of do-calculus) would have required experimentation. And in other cases, you're still stuck.