I first look at comparing two offers to pick which one has better CR.
Then I optimize the banners, and then I optimize the landers.
I always question is my method is ok, or not. Is it, or do I have it all mixed up in terms of optimization flow.
ATM our optimization process is:
Offers, Angles, Landers, Banners, Carriers/os/handsets, site/publisher Id's.
We noticed that in this hierarchy that waiting to optimize the bottom 3 variables has helped a lot, because swapping the higher up in the food chain variables around can make the lower on the food chain variables profitable without changing them. IE... site Id's might be negative at first, with bad landers, but when you change the lander up, they then become profitable... or a lander might have a bad CVR %, but changing the Angle or the Offer will improve it etc...
I would go offer/lander/banners 
Offer is always the most important.
I follow that up by finding the most suitable angle.
A better outline of the process here: http://stmforum.com/forum/showthread...icient-Testing
Thanks Finch for referrencing a great post of yours. I get it lander matters after offer. The picture is clear. Stm rocks!