At that YouTube workshop I went to a month back, we actually spent a lot of time talking about subscriber purges, and the room was really split on the subject.
The older channels were concerned about the optics of having 2-3m subs but only getting 50-60k views per video, while the newer channels were more concerned about the optics of slipping below a milestone.
It was actually really informative in that the YouTube folks got into some detail about which dashboard metrics influence discovery and which are purely informational.
A few folks were concerned that accumulating inactive subscribers was hurting their channel by driving down % subscriber engagement, but we were told that subscribers only indirectly impact discovery by being an audience, that it's views and engagement in absolute that matter.
So the stats on the dashboard of "how many viewers are subscribers" is really just for channel owners to better understand who watches their stuff, and isn't being ranked in some way.
A lot of really popular channels, like tech tutorial channels, have like 1% subscriber viewership because overwhelmingly their traffic comes from people searching for answers to a problem. How the traffic gets there is incidental.
A huge chunk of the workshop was about pivoting or evolving your channel, so this was actually pretty relevant as an example, since if most of your viewers aren't subscribers then it's easier to experiment with new formats without "burning" your audience.
Where if most of your viewers *are* subscribers then changing formats or genres is more of a negotiated process between you and your audience: what do you want to do, what do they want to see, etc.
But, also, a committed audience might be really willing to trust you to show them something new and cool, while trying new things with low subscriber engagement can feel like spitting in the wind.
So there's pros and cons to both, and the point of those metrics is to help channel owners better understand what their actual people-clicking-and-watching-my-videos audience looks like.
