Conversion I can do.
I'm hoping to figure out the probability of a unique visitor clicking on an ad at some point during their visit. Armed with that, I can how many unique visitors will follow the ad off the site.
(As to figuring out how many of those customers come back to the site, I'm clueless and haven't even begun to tackle the problem.)
Based on that, I can show how many non-buyers do something that'll make us some money (X) and how many buyers we lost (Y).
So then if the Revenue(X) we gain is greater than the Revenue(Y) we loose, I can say Ads are good. Conversely if we loose a bucketload of sales, but don't make enough on ads to cover it, I can say "Ads is Bad."
So it's pretty elementary, other than the tricky part of figuring out how many people won't buy stuff.