Of course a big part of what the NSA does is manage the scale. To do that they need metadata both collected with the calls (e.g. what numbers the call was from and to, if available who or what that number belonged to at the time) and assembled by analysing the audio ("male speaker", "Russian language", phrases or words maybe phonetically)

After all it's all very well having every phone call, but you can't possibly listen to them all, so you need a way to find those that you do want to listen to.

"Pervasive Monitoring is an Attack" is current policy of the IETF and accordingly new protocols are intended to combat such monitoring, but there is a lot of metadata revealed by your activities and masking it takes considerable effort, while outfits like the NSA are very enthusiastic about gathering whatever metadata they can find and discovering new ways to collect more.