Tips for Google Adwords Testing

One of the reasons (aside from great performance) that Google Adwords and other search advertising has become so popular is the ability to optimize campaigns “on the fly”. With unprecedented control in a self-service interface, the user controls budget, pricing, targeting and creative within a simple, easy-to-learn interface. But with all this control comes the downside of “over-optimization”.

Have you ever been too hot and turned down your air-conditioner only to find that you are too cold later? And then you turn the AC back up only to find that you are too hot later? The difficulty getting it just right is, in effect, what happens with over-optimizing search: The user logs in, changes all the parameters (keywords, bids, creative, etc.) to improve results and then finds that they did too much. So they log back in and make a number of similar changes to correct the campaign; except the campaign quickly goes back the other way. This happens to too many search campaigns and the following is intended to provide a brief guide on how to avoid over-optimization:

Scientific Method: This holds true for Adwords as much as science: Before taking action, create a hypothesis and construct the campaign in a way that lets you know if your hypothesis was right or wrong.

Learning: Whether a test succeeds or fails, knowing what does not work is just as valuable as knowing what does. Ensure that variables are isolated so that results – positive or negative – are definitive.

Historical Data: Do not continue to rely on results and assumptions that are more than 3-6 months old. Continue to re-test former winners and losers to ensure that optimization takes into account changes in the landscape that may impact current results.

Limit Variables: Tests should not overlap with each other. All testing variables should be isolated to eliminate the chance of misreading results and to drive learning more quickly. For example, a change in bid strategy should not be executed at the same time as a creative test if at all possible because it will be difficult to attribute a change in performance to the correct cause.

Limit Risk: Do not run tests on all keywords or adgroups at the same time because results will be unpredictable and there is always a chance that tests will yield poor results. Segment a portion of the campaign to test and run winners on the remainder of the campaign. The level of testing is dependent upon the marketer’s threshold for risk, but should be somewhere between 10-40% of budget. For example, if there are 10 adgroups spending $1000/month each, then segment 1-4 adgroups to be eligible for tests and run winners on the remaining adgroups. Rotate the 1-4 adgroups every time a new test is started to ensure that no overtesting of any particular adgroup occurs.

Click to read more posts about paid search or marketing tests.

© 2009. Thrivepoint LLC. All Rights Reserved.