This is an area where i advised my engineers a lot. Experimentation doesnt mean we get the licensed to fly blind. Experiment design, like software needs to have carefully crafted hypothesis and some data to back that up. Having data actually helps u refine your experiment strategy and there by saving campaign cycles and dollars, in some cases.

Also, we need to learn from each experiment and use the failures to tweak the experiment strategy for the next round.

This sounds on paper a reasonable expectation and makes it sound like u didnt want to be accountable? Its always "Trust but Verify" and never "Trust me".