Good data-driven marketing teams work hard to establish control groups and known baselines before starting high-spend campaigns, and the ROI of those can be quantified by referrer info if they're highly specific and keyword-related.

As for more general campaigns, in my experience, it's a crapshoot, if the numbers go up everyone is happy whether or not marketing was responsible for the majority of it. I'm sure other companies have better approaches than that though.

Good question! Combination of holdout groups (split testing) based on historical keyword traffic/conversion and tagging paid and natural search entries to compare cannibalization across campaigns.