Myself and quite few others who come by themselves to the gym more infrequently andwere using the machines (you need a spotter for the same exercises on the free weights). Now both groups paid the same for membership, but they essentially removed the ability for one group to train, needless to say I moved to a different gym. By blindly following the data they optised for worse customers (you want people to subscribe but not show up often).
There are other considerations. If it costs a lot more to have someone using the machines than to have someone using the free weights, you might be better off with just the free-weight crowd than you are also having the intermittent machine guys. You want customers that are profitable, which isn't the same thing as customers that only rarely show up.
If it's really what happened I'd see that as a pretty great move.
At same spaces, yes! Good quality barbells and weights, while expensive, keep for over 30 years. Bodybuilding and powerlifting doesn't take much space, so you don't need a large space, and the customer base is dedicated and can often tolerate their gyms being a bit out of the way. Machines, however, like to break and are incredibly expensive compared to barbells and weights.
When people want to join a gym they ask the fit guy which gym they recommend. It's a bit like techies recommending thinkpads or macbooks which gave those respective products massive boosts.
As a non-English computer user, I'm increasingly annoyed by Crowdin-style jumblations. Missed context is the common trope for both - whether the less-used feature do mean the feature is less important, or whether the "Play" on a button do mean "I instruct you _Play_ this media" or "I want to _Play_ this game" depends on the context(the language I mainly use do use separate words for the latter two).
Pretty much the problem in a nutshell.
Which can make their results absolute garbage.
Like data driven can tell you a button is "not used often" it can't tell you that a button is "essential to be fast available in some safety critical situations". (Or for other examples, that a feature is not unused because people don't want it but because it's hidden or bad designed.)
But somehow I still meat people which believe that poorly data driven approaches will yield the best results, which as far as I can tell is complete unrealistic. (Which doesn't mean in any way you shouldn't also use data for decisions, just be aware that data just shows a part of an picture and can often be very misleading.)