In my post “Banners versus Search“, I talked about the problem with view-through measurements that so many folks use to justify banner buys. Kevin Lee has written a short piece on ClickZ worth checking out for those interested in more on this issue and all the pitfalls of the view-through measurement. He also includes a proposal for a new measurement called “view around”, which sounds to me like a classic controlled test / regression model scenario.
This is a much better approach; after all, do you really care what the actual behavior behind the banner / search combo is (measuring accurately) or do you care about how to optimize the combo and make more money (measuring precisely)?
Since you can’t really control the way traffic interacts with a combined banner / search campaign anyway, attempting to measure something like “view-through” isn’t actionable; there are too many uncontrolled events. Measuring the effectiveness of various combinations – regardless of which is viewed / clicked first, last, how many times, etc. is a much more actionable approach.