PPC Split Testing: Reducing Risk When Testing New Ads Against Control
I learned of this tip through a fantastic free video explaining the AdWords Triangulation Method (38:50 minutes through the video).
Here's the danger in testing a winning ad (as your control) against experimental copy - if your new creative is a dud, you could lose 25-50% of your click through rate potential if your ads are showing evenly, depending on how many ads you're testing.
A smarter approach recommended by Stompernet is to create 4 identical copies of your control ad, and one test ad. This ensures that your "safe" ad will show 80% of the time, and your test ad only 20%. Plus, the control ad's solid click through history will cause it to rank at higher positions than brand new ads, so creating new copies reduces this bias.
I think you could also pause the control ad, and create 4 new copies to eliminate the bias altogether. But it's nice to have one copy that's going to achieve better positions at lower click through rates, especially if you're continually testing. If you can tolerate more risk, you could create just one copy of your control so you're running 3 ads - 66% of the time you're running your control copy, 33% of the time you're getting higher positions and lower CPC from the click through history of the established ad, and 33% is your experimental ad. Then you can segment out the control ad with the history to compare performance of your control copy vs. new copy, without the ranking bias.