What would your collection proposal be then?
You can't determine data of interest until you have sufficient data to determine if it's of interest in the first place. Even the NSA doesn't have an Oracle computer that can look into the future and figure out what vanishingly small percentage of communications are just the ones they should be interested in. If they did, they could also solve the halting problem and rewrite the history of computer science and time-travel.
Remember, the standard the NSA is held up to is that it should not only be following known bad people who are/might be doing bad things, but to ferret out the unknown bad people. Every time there's a Boston or similar, everybody goes and climbs all over the NSA for "not knowing about these guys"
You can't do that until you have a sufficiently large enough collection of unevaluated data to start looking through.
I'm not saying it's right or wrong, only that it's the reality of the task spy agencies have before them.