As Dakoda indicated, the entire sub tree has to be walked and tallied up. With rate limits, especially the upcoming rate limits with Helix, that is pretty easy for channels with few subscribers - however, when you get to channels with thousands and thousands of subscribers, only 100 users may be queried at once. Additionally, a cursor has to be used and during large queries you will never be guaranteed to have correct data as it can change during the query (which becomes what is called a Dirty Read) and will potentially may never be correct - again, that only impacts very large subscription bases. It cannot be constantly queried so it has to be timed to run on a limited basis, meaning that it might not be up to date.
There is also a possibility that they are using an undocumented endpoint. There are some data sets that be queried, for example, with a different type of query, however, Twitch has indicated that endpoint set is not to be used and may result in a ban. Perhaps they look the other way for larger customers, but I doubt they would look the other way for individuals.