This is why I almost stopped reading when the article suggested a "normal statistician" would just use the mean score (it does later address this, for those who haven't read it yet).

IMO, NPS doesn't provide a good enough signal to be "the one number" you should care about, but it recognizes the actual numbers don't matter all that much. I've seen the same question used with a 5 point Likert scale which I imagine is easier to work with. You'd still have to deal with people responding like "0/10 can't think of any friend I'd recommend this to" :/