Averages lie so one of the best ways to improve the efficiency of a PPC account is to use more granular segmentation.
In this example the average overvalues segment one and undervalues segments two and three. By restructuring the account to allow targeting on segments two and three separately from segment one efficiency can be improved.
In 2012 I did this for some of my travel clients; separating London out by creating geo-targeted campaigns. The value of London traffic is normally higher because London has wealthier people but in this case the effect was exaggerated because of the Olympics. Running ads like "escape the city this August" was a highly targeted message for this audience and the campaigns performed well.
So now I have a lot of London specific campaigns kicking around. And they still perform better than the overall average, but the delta is much smaller now. If London were still grouped in with everything else I certainly wouldn't segment it into a separate campaign now; the difference in performance isn't big enough. But it feels wrong to merge campaigns together again; I spend so much time thinking about segmentation that I've never considered when to aggregate things back together again.
This is something that I should give more thought to; PPC is a dynamic enviroment. I have appreciated for some time that dimensions occasionally perform differently enough that segmentation is worth the extra effort. But I've never thought about when to go the other way.