To a large extent I agree with you. Differences creep in when there is ambiguity and judgement involved on what is that metric that the org wants to optimize.

To a large extent I agree with you. Differences creep in when there is ambiguity and judgement involved on what is that metric that the org wants to optimize. This is fairly common. Typically, in these situations its the PMs who make the final call. There the goal of the experiment protocol is glean as much knowledge as possible, and present it to the PM. The thinking there is -- if that comes at the cost of exposing some customers to bad choices, so be it.