Knowing what to test and how to interpret the results based on nuances and oddities of experiments is an important skill for people, not automations.
While there are many true-and-tried best practices in search marketing, the devil is in the details when it comes to achieving the best results. For example, it’s hard to argue with the merits of automated bidding but it’s not that hard to get bad results if you deploy it incorrectly.
Say you read that Hertz used smart bidding to reduce their CPA by 35% so you decide to deploy the same strategy in your account. If it were that simple to run a successful Google Ads account, we’d all be out of jobs. Simply knowing what feature to use isn’t enough as you also need to know the right settings that will make it work as well for you as it did for the advertiser in the case study.
And to be the best search marketers we can be, we can’t simply look at what other advertisers did. Instead, we can take hints from others and use it as the basis for honing in on what works for us. We have to discover the details of the right way ourselves.
And that’s why being really good at PPC experimentation is so important. I spoke on this topic at SMX East in the session “Awesome Tests, Profitable Results,” and here are some of the key takeaways.
The three most popular PPC testing methodologies
One of the key claims to fame of search marketing is that it’s more measurable. So whenever we try something new, we better have some numbers to back up our findings so we need to run experiments in a structured manner.
There are three ways we usually see this done.
The simplest way to start a test is to make a change in a live campaign and then compare the results from before and after the change was implemented. The beauty of this method is that you can test anything in your ads account quickly. The downside is that while setup is super quick, measurement takes more effort and you can’t have an apples-to-apples comparison because results may be impacted by external factors that change during the before- and after periods.
How much data to compare
When measuring results, allow enough time to minimize time-based factors. And while I’d love to tell you exactly how much time that is, remember the point that things differ for every advertiser, industry, etc.
But if you want some guidance, at the very least measure one full week before and after to remove the impact of weekday versus weekend performance.
If you are in a vertical where not just the day of the week but also the time of the month plays a role, e.g., in automotive, measure a full month before and after the change. In automotive, the time of the month may impact how aggressive dealers are on price as they try to hit monthly targets and consumers’ willingness to buy fluctuates along with the dates when they get their paychecks.
Lookback windows for bid management changes
Specific to bid management, if you’re using the before-and-after technique to measure impact, remember that your lookback window should be the same as the frequency of your changes. For example, if you make bid changes every day, you can’t look at the last 30 days of performance data because that might include data from 30 different bid levels, one for each day of the lookback period.
So clearly, a before-and-after testing methodology comes with some serious challenges and that’s why both Microsoft and Google have added features to run better tests in PPC accounts. While it takes a bit more time to set up the experimental campaign with all the changes to be tested, it has the benefit of removing any potential skew in results that’s common in before-and-after tests. It does this by letting the advertiser do a split test, for example, a 50-50 split where half the users are served the control and the other half the experiment.
And not only are the results more reliable, whatever time is invested to set up the experiment is easily recouped because reporting of results is baked into the product. – Read more