The summary seems to indicate that the value of "Stellarwind" wasn't clear because it was one of many sources and few had access to it, not that all NSA spying was seen as ineffective.
The NSA does so much spying that it seems like it would be hard to ever calculate the marginal value of each additional unit of spying. Probably more so because of the fragmentary and unreliable nature of clandestine information and the need to develop multiple sources to achieve any kind of confidence about a particular conclusion or piece of information.
The latter bit is probably what leads to never-ending development of new data sources and methods, especially as each new spying method becomes less and less specific and requires more and more analysis to tease out information. Call metadata doesn't tell you what was discussed or necessarily who was called. You need parallel data from some other source to tell you who is associated with those numbers, where they were, etc.