A/B testing
A/B testing is running two versions of a page at the same time, sending each visitor to one of them at random, and comparing the results to see whether the difference between the versions actually changed behaviour.
The randomisation is the whole method. Comparing this month to last month compares two different sets of visitors in two different weeks; splitting one stream of visitors between two versions compares the versions and nothing else.
The result needs a significance test before it means anything. A version that is ahead 12 conversions to 9 is almost certainly ahead by chance, and acting on it is worse than not testing, because it feels like evidence.
Two practical requirements are easy to get wrong: a returning visitor must see the version they saw before, or the comparison is measuring inconsistency; and the versions must not each have their own indexable URL, or the test has quietly published duplicate content.