Run a clean A/B test, not a muddy one

Run a clean A/B test, not a muddy one
3am, want to test a headline. Do it clean or learn nothing:

— Change ONE variable. One. Headline OR image, never both
— Same audience, same budget, same start time
— Let it run to 50 conv per arm before calling it
— Stat tie? Keep the cheaper-CPM winner

Change two things, learn zero. Lock one variable on your next test.