In previous posts we explored the theoretical reasons why ”zombie SKU” PMax campaigns won’t work, how Google Shopping Ads actually select what products to show, and then looked at historical data about how zombie SKUs actually just come back to life on their own. But what about a real test? This post will go over the details of a live test of a zombie SKU Performance Max campaign and how it didn’t just achieve zero results – it dropped account performance.
The Experiment
We wanted to run a randomized controlled trial of what would happen to zombie SKU products if we put them into a new PMax campaign versus if we left them alone in the campaign they were already running in.
Our definition of “zombie SKU” was any product that had received 0 clicks in shopping ads in a specific campaign in Google Ads during the previous calendar quarter. We picked a good-sized campaign with lots of conversion data, good budgets, and about 18,000 products. Of those, about 12,000 products matched our criteria as being a zombie SKU.
We then randomized those into two groups of products and assigned a custom label in Google Merchant Center to denote which groups each product was in. The control group was left in the existing PMax campaign.
The test group was assigned to a new PMax campaign that was configured to only show Shopping Ads (it didn’t have other assets that would allow it to run Search and Display Ads). All of the settings that would impact Shopping Ad performance were set precisely the same as in the original PMax campaign. Both campaigns had the same target ROAS goal, unconstrained budgets, and were allowed to spend any amount while attempting to achieve their ROAS goal.
After 28 days, we compared the results of the zombie SKU campaign to our control group to see which group of products did better. We did this using data from Shopping Ad performance in the Google Ads Report Editor for each group of products.
Avoiding Biases
Before getting to the results, I want to point out that we designed our test to eliminate the most misleading biases plaguing the poor advice about why to create zombie SKU campaigns.
First off, we compared apples to apples. Most advice out there has you filter out zombie SKUs using data in the Report Editor of products that did not get any results in a prior period of time, but then they compare the overall results of a new PMax campaign against those previous product-level results. This is crazy, because PMax campaigns also run text Search Ads and various Display Ads, in addition to Shopping Ads. A zombie SKU campaign, depending on how it is set up, could actually get all of its results from text ads showing up for brand keyword searches. You can’t compare campaign data to product data!
While generating our list of products to test, we tried to eliminate other situations where the products would be expected to improve in their performance for reasons other than being in a zombie SKU campaign. We eliminated products that had been recently added to the feed, which simply might not have had time to get their first clicks yet. We also eliminated any products that were disapproved in Google Merchant Center, knowing that many of those would get approved in the future and start getting clicks again regardless.
We randomized the 12,000 zombie SKU products into two groups so we could test half of them while leaving the other half as an untreated control group. This is very important. Most posts evangelizing zombie SKU campaigns fail to try something like this, and just move all of the zombie SKUs into a new campaign and then when it starts achieving results there is no way to know what would have happened if nothing special had been done.
Finally, to eliminate another source of data inconsistency, we calculated conversion value and ROAS using the “by conv. time” metrics in Google Ads rather than the normal conversion value metrics. So we only compared results that actually occurred during the experimental period and not future sales that were attributed back to the campaign after the experiment ended.
The Experimental Results
The Zombie SKU PMax campaign products:
- 59,635 Impressions
- 0.36% CTR
- 216 Clicks
- $480.05 Cost
- 19.34 Conversions (by conv. time)
- $2662.83 Conversion Value (by conv. time)
- 5.55 ROAS (by conv. time)
The Original PMax campaign’s control group of products:
- 39,753 Impressions
- 0.66% CTR
- 263 Clicks
- $654.08 Cost
- 20.78 Conversions (by conv. time)
- $4,185.91 Conversion Value (by conv. time)
- 6.40 ROAS (by conv. time)
At a glance, the original PMax campaign with half of the zombie SKUs in it won the test. The zombie SKU campaign was not able to keep up with the performance of zombie SKUs left in the original campaign.
But why? Well, I’d encourage you to go back and read the previous posts in this series, collected under Don’t Fall for Zombie SKU PMax Campaigns.
There are a lot of reasons we expected this to be the case (and I wrote those posts 1-3 months before starting this experiment, so this wasn’t a set up!). New campaigns with less data perform more poorly. Smaller campaigns perform more poorly. And we expect that some products that didn’t perform well in the past will bounce back on their own.
The way to interpret these results is simply that. Some products that don’t perform well in the past will bounce back into performing better on their own, due to all sorts of factors both in and out of your control. If you move those products out into a new campaign, it starts from scratch testing everything and showing poorer search results to users as it starts to build up data. That’s why the zombie SKU PMax campaign had much higher impressions and such a low CTR. It tried a lot of things that didn’t work.
If we had just put all of the products into a zombie SKU PMax campaign and not run a test like this, we would have seen those campaign results and tricked our minds into thinking that we had brought those zombie SKUs back to life, and it was a result of our new campaign. But in fact, we would have been completely wrong.
Zombie SKU Campaigns Don’t Work
The bottom line is that zombie SKU PMax campaigns don’t work. We have explored some theoretical reasons why. We have explored how Google Shopping Ads actually work and what you should be paying attention to instead. We have explored historical data to show that zombie SKUs come back to life on their own. And we have now run an experiment that controlled for a lot of the errors in judgment that people generally make around these campaigns.
But it’s even worse than concluding that zombie SKU PMax campaigns don’t work. The data suggests they reduce performance! And if you factor in the wasted labor costs of setting them up, managing them, and reporting on them, they are likely a substantial waste of your company’s resources.
Zombie SKU campaigns destroy value. Don’t make one.