I think you would target high ranking politicians... because the option to blackmail them ensures you an uninterrupted and growing black budget.
You would target communications businesses... because compromising a properly placed employee grants you access to the sum total of customer and peer communications through their networks.
You would target media... because a jump on a popular change in public sentiment is very actionable intelligence, both to multiply funding (through investments prior to predictable market response), and to further control politics.
You would target diplomats... both for tradition's sake, and because borders are the most easily grasped us-and-them (divisive) tool in the post 20th century semantic playground, giving you options for powerful public sentiment manipulation through selective media generation. However, realistically for most embassies worth their salt you'd know these groups are largely not going to do anything remotely surprising that you can pin to them through pervasive communications intelligence gathering.
Probably also, you would target multinationals, because almost all of them are doing something dodgy, somewhere, and that gives you tremendous leeway for behavioral modification.
But in reality, the majority of these can be monitored very effectively on an automated basis with near zero effort once you have full visibility of various domestic financial networks, the SWIFT international financial network, credit and debit card networks, electronic information on intended travel (passenger name records) and border crossing (whoops! I-lost-my ... new passport number, anyone?), and the public switched telephone network.
Email, social network and general web use are cute extras, and probably greatly useful just for delving in to people's character and actions, communicative profiling (grammar, typing style, languages known), interest profiling, waking hour and social network profiling (beyond just phones), etc. But I don't think it's necessary to go to that point most of the time ... the broad metrics are already available and probably extremely reliable unless people are making a concerted effort to bypass dragnet surveillance activity. (eg. By avoiding all of the above networks... damn hard these days, it would seem, for any length of time)