Experiments and Drafts

Experiments and drafts let you clone a campaign, change one thing, and run it as a randomized split test against the original so you learn whether the change actually helped.

You have a hunch. Maybe a new bid strategy will lower your cost per lead, or a rewritten set of headlines will pull better. The lazy way is to just flip the switch on your live campaign, wait a week, and compare last week's numbers to this week's. The problem is that the week itself changed too, so you never really know if it was your edit or the weather. An experiment fixes that. You make a draft, a working copy of the campaign, adjust the one thing you want to test, and then launch it as an experiment that splits traffic evenly with the original.

Now both versions run at the same time, against the same market, with people randomly sent to one or the other. When the results come in, the difference between them is a fair, apples-to-apples read on your change instead of a guess dressed up as a conclusion. You keep the winner and discard the rest, and you did it without betting the whole account on a hunch.

The catch is that this honest method still needs fuel. Google recommends letting an experiment run about four to six weeks to gather enough data, and that assumes you have the traffic to fill both halves. On a low-volume account, each side collects a trickle of clicks and the gap between them is mostly random noise. The tool will still show you a number, and it will look like a verdict, but it is not one. If you cannot feed both arms enough data, decide by judgment instead.

Test when you have the traffic to prove it, and decide by judgment when you do not.