Problem is, the latter strategy is rarely used (telemetry is cheaper!), and as far as I can see the consequences, telemetry data is hard to use correctly. In particular, it's prone to become a mirror of your design, creating a feedback loop. Telemetry will show people will use more of the things you've exposed more, and less of the things you've exposed less, so if you take that at face value, then you're just amplifying your own initial design bias.