They don't have to be mutually exclusive, do they? You can surely learn some things from large-scale telemetry collection that you can't learn from user studies (e.g., semi-representative "in the wild" performance metrics).