A/B split testing is the foundation for all direct response marketing and the tools available to paid search marketers make this channel especially appealing. Creating a test takes just seconds and can yield big gains in account performance. Unfortunately, while these tests are easy to run, running them well can be challenging. Here are 3 rules to help maximize results from your current and future A/B split tests.
Every marketer has heard the story of the 300% conversion boost when they switched their Conversion Buttons from blue to red, or the double digit drop in Conversion Cost they achieved by changing their CTA from “Submit” to “Sign Up Now”. These stories are famous for two reasons:
1.They give credence to the value of testing.
2. They’re rare.
More often than not, serious lifts come only from substantial changes to product positioning or a campaign’s offer. It’s easy to create ads with minor tweaks to things like capitalization and word order, but these will rarely provide serious gains.
Instead, test big ticket items. Change the price, highlight different product attributes, and test out features vs. amenities. While none of these are guaranteed to provide a boost, they have a much better chance of doing so than playing around with formatting or details.
Eliminate Extra Variables
One of the biggest issues I see when we analyze Paid Search accounts here at Mittcom is setting up tests that aren’t isolating variables. This can be caused by different factors, but the most common is not pausing current champion ads before beginning a test on a new version
Why should I pause the original ad?
Google calculates Quality Score each and every time an ad is served, so the history of an ad will impact how that ad performs moving forward. Assuming that the goal of running an A/B split test is to find the better ad copy, we want to make sure that everything else (bid, impression share, audience & keywords) is identical. If we run an ad that has a history against one that does not, this history will muddle the results of your test.
If I pause the ad, what do I test against?
Duplicate and pause the original ad and leave the new ad active. Then test the new version against this duplicate ad. This way, both ads start with the same quality score and the true winner can emerge.
Ad Rotation Settings
Changes to Shorten the Learning Curve
Google has made several moves in the past few years to make the Adwords Platform a more turnkey experience for new users. New programs like Adwords Express, Enhanced Campaigns and changes to geographic targeting are all designed to make it easier for new users to get a campaign up and running quickly. As with any change, there are some trade offs, and these changes were no exception. Each additional feature designed to simplify the process also made advanced testing a bit harder.
One of those changes was to the ad delivery method, which was changed several times in the summer of 2012. For the sake of simplicity, we’ll just discuss what this means for ad testing. For those interested in the details, PPCHero documents many of these changes in an article located here.
Fix The Problem
The setting changes your ads from serving evenly (which is necessary for proper A/B testing) to serving ads Google believes will drive more clicks.
To test your ads, you want each version to receive a randomized portion of the ad group’s traffic. Unfortunately the default from Google doesn’t do this, but a quick change will point you in the right direction.
- Sign in to your account.
- Select the “Campaigns” tab from the green menu bar.
- Select the Camapign you’d like to edit.
- Select the “Settings” tab from the Campaign navigation.
- Click “Ad delivery: Ad rotation, frequency capping” link and select “Edit” under “Ad Rotation”.
- Select “Rotate indefinitely: Show lower performing ads…”
- Click Save.
Note: Google will show a warning message saying something to the effect of “this may harm the account’s performance, not recommended for most advertisers”. This warning is designed to prevent new users from accidentally selecting this option.