It's not, really. You can extrapolate based on achievements. Specifically you can follow known trends such as how many people clear a normal raid.
For example - if, in X.0 10,000 guilds cleared normal and then in X.1 8,500 cleared normal and then in Y.0 8,000 cleared normal - we can easily presume the subscription numbers are going down at a specific rate.
You may not know exact numbers but by following overall trends from the years you can get a very good pictures of the health of the game.
So no, it's not all speculative. It can be heavily inferred based off of previous data.
You can even track how many guilds clear normal and how long it took them. So this will allow you to take into account the difficulty of the raid itself. With Mythic Plus data you can track how many are playing regularly now and extrapolate the health of the game from that.
The thing about WoW, specifically, is it's health relies on the snowball effect. They got lucky. In much the same way Facebook got lucky and people don't just jump to the next big thing. It's why people in the EU aren't jumping to signal overnight but, instead, are still using WhatsApp. Timing is important as well as having a critical mass / momentum going.
This means if trends start showing the player count as slowing - that's bad. Very bad.
This is why they need to focus on new blood. I seem to recall most people quit the game before even reaching max level where the "real" game begins. This could be for a variety of reasons - none of which I'll ever trust Ion's responses nor will I ever trust Ghostcrawler's previous responses about it.