The software analogy is invalid, because those people are already your users or customers. In the case of reading the article, think of a popup as a criterion filter in much the same way as 419 scammers use poor english.
Just because it's mildly annoying, doesn't make them any less effective either. There wouldn't be millions spent on Super Bowl slots if they weren't effective (although the SB is an outlier with regards to adverts).
Except for really bad popups, I can find an x, click it and move on - or ever so occasionally actually sign up for something...
It hasn't been until recently that you could timeshift or torrent TV shows consistently to not see the commercials. They do show that mildly annoying can still be effective and profitable.
To me this does not make sense, and is exactly the reason I avoid annoying users in my applications (advertising or otherwise). Because you definitely have numbers to back up the effectiveness of the popup (although I would argue it is not effective). What you don't have is of the 98% you annoyed, how many will not return because of the annoyance (by choice anyway).
Or to put in another way,
Would you implement a feature in your software that only benefitted 2% and annoyed the 98%?
Seems like this kind of approach hints at a deeper issue here and that's effective targeting.