So A/B testing does not work because people using it don't understand the underlying statistics and don't have a large enough sample size. So the solution is an even more complex test with even more variable? Makes sense to me.