I mean, seriously...the advertising industry hasn't been paying attention to the response rate of different variations on marketing campaigns until now? Really? Isn't that, like, one of the key benefits of web advertising over other mediums?
Think about this -- you run an online campaign, serving hundreds of millions to billions of impressions, with different levels of users creating different interactions with your ads, website, etc and a certain percent "converting" to the action you want. How do you reliably process hundreds of millions to billions of rows of data and tie it together? What factors are reliable predictors of future actions? If you increase one aspect does it correlate with other efforts increasing? If there is a flaw in the data collection, how do you spot it?
What about offline behavior? Can you correlate interaction with a widget with store purchase behavior? A lot of activity happens that can never be tracked and it needs to be factored into the models.
So the point being when you dig deep it is not as simple as it seems and there is still a long way to go for all of the technology to work well together and provide the analysis needed to react in real time and change human behavior for the desired result.