We recently explored the idea that the allure of zombie SKU Performance Max (PMax) campaigns may tempt ecommerce merchants to make risky changes to their ad campaigns. We presented our theories about why proponents of zombie SKU campaigns are falling prey to some statistical fallacies. This article moves out of theory and offers historical evidence that zombie SKU PMax campaigns don’t give the results people say they do.
Quick Context: What Is a Zombie SKU Campaign?
There is a current trend in recommendations from ad agencies and Google reps that if you have low-performing products in Performance Max campaigns, you can push them into a “zombie SKU” campaign so the new campaign will try them out again and find winning products that otherwise would have been relegated to oblivion. You should be very skeptical of this idea.
The reasoning behind this is incredibly suspect. Among proponents of zombie SKU campaigns, there appears to be some confusion around how products actually win auctions and get clicks. There’s a type of magical thinking at play that suggests the underlying machine learning models must be going haywire. But there are far more obvious explanations for why your products are not performing well (bad data, bad prices, competitors who are more aggressive, etc.).
For more of the background on what it actually takes for products to show up in Google Shopping Ads and perform well, and background on the biases that probably make the idea of zombie SKU Performance Max campaigns bunk, check out these links:
After exploring the biases and the problems with the idea of zombie SKU campaigns, now we’re going to leave the realm of theory and dive into some actual data.
Past vs. Future
The key problem with how most people evaluate the performance of zombie SKU campaigns is one of statistical bias. If you simply kick items that have zero clicks into a new campaign, and the new campaign starts to perform, can you really be sure that it’s because you made a new campaign? How do you know those products wouldn’t have increased their performance anyway? You don’t.
But this is the justification that everyone uses when trying to sell you on zombie SKU campaigns. “Look, I set up this new campaign, and it lifted revenue by X% over our previous campaigns!” How do we actually know? Is this just a statistical bias? I think so. Today we’ll provide solid evidence.
One way to consider what is really going on here is to go back into the past where we did not make a change, and see what the data looked like going forward after doing nothing. Did those products continue to get no clicks and conversions? Or did some of them start to perform again after that point in time?
So today, let’s explore the counterfactual. What happened with zombie SKUs in the past with one of our clients, after we did not make a zombie SKU campaign?
Setting Up the Experiment
We went back to Q4 of 2023 and pulled reports of all products on the client website that existed during that period but received no clicks in Google Shopping Ads running out of the Performance Max campaigns. Since this client has several campaigns and we want to compare apples to apples, we decided to restrict this analysis to a single campaign with a lot of products in it and a lot of data to analyze.
These are all products which, if we bought into the idea of zombie SKU campaigns, we might have kicked out of the campaign they were in and set up a new zombie SKU campaign back in Q1 of 2024. If we had done that, we’d be evaluating their performance in this new campaign and comparing it to the original campaign. Obviously we did nothing to them, and they remained in the original campaign, but we can still split campaign performance between the zombie and non-zombie SKUs from Q3 2023 and see how these two groups did in 2024.
This is a great experiment. We can back-test the data and imagine these two buckets of data, the zombie and non-zombie SKUs, and know that in fact, they had precisely the same campaign settings with no outside confounding factors. It’s an A/B test of what actually happened.
Zombie SKU Performance Increased
As you might expect if you read our previous article exploring biases in how zombie SKU campaigns are evaluated, the zombie SKUs that didn’t get any clicks in Q3 2023 went on to get lots of clicks in 2024. They also got conversions.
If we had split these SKUs out into a separate campaign and compared their statistics back to the main campaign, this is what the data would look like:
Impressions – 20.7% lift from the zombie SKU campaign
Clicks – 12.4% lift from the zombie SKU campaign
Conversions – 11.1% lift from the zombie SKU campaign
Conversion Value – 8.2% lift from the zombie SKU campaign
If all else had been equal, but these were two campaigns, we would conclude that the zombie SKU campaign had lifted our revenue by 8.2%!
But that’s not what happened. We did nothing. We left those products in the campaign they were already in. And the “zombie” SKUs from Q3 2023 went on to generate that revenue with no special treatment. We didn’t have to do anything. We didn’t set up a zombie SKU campaign. That’s just what happens in the real world. Sometimes products don’t get clicks for a while. Later on they bounce back.
How Do You Interpret That?
It’s good to stay humble. We still don’t know what would have happened if we had set up a zombie SKU campaign. Perhaps we would have gotten even more conversions from those products, right?
But the opposite is also possible. Perhaps if we had set up a new campaign, the products shifted into the new campaign that had no performance history might have overspent and underperformed for a while and results might have been worse.
We simply have no way of knowing.
What we can say with confidence at this point is that there are a lot of reasons that products that are receiving no clicks are still likely to go on to receive clicks and make sales in the future. So anyone who just makes a zombie SKU campaign and then points to it as the cause of those SKUs getting more sales in the future is making a basic mistake in reasoning.
Do Zombie SKU PMax Campaigns Work?
So far we have seen no evidence that Zombie SKU Performance Max campaigns actually work, and we have solid theory as well as some historical data that suggest their purported results are largely or entirely fake.
They might work, but if so, they still definitely have less impact than most people seem to think. They might also reduce performance. How could we determine that?
Well, first we would need to control for a few basic biases. Like we might want to exclude from our zombie SKU list any products that had recently been added to a website. We might want to exclude high-ticket items that get low conversions since those could easily skew the results by random chance.
But what we really need after kicking some items out of the test is a basic A/B test or randomized controlled trial. We need products randomly assigned to group A and group B. Then we’d need leave one group in the campaign they were already in, while we exclude products from the other group and move them into a new Zombie SKU PMax campaign with precisely the same settings as the main campaign.
Even this wouldn’t be a perfect test with no biases. But it would be a lot better than what most people are attempting, which is not to test anything at all. So we’ll be testing this on a real ad account soon.
Doing Nothing Might Be the Best Strategy
We know that the idea of Zombie SKU Performance Max campaigns has theoretical problems. And we can see from historical data that their purported success could also be seen in sets of zombie SKUs if we had done nothing.
But there is also a solid theoretical explanation for how zombie SKU campaigns might decrease performance by fragmenting data into smaller and newer campaigns (see How to Structure Ad Campaigns for Better Results). There is real risk of dropping performance with a zombie SKU strategy.
So what do we do? In the absence of more evidence, doing nothing might be the best strategy. Going back to the key question of this article, what would happen if we do nothing? If we do nothing, the performance of our “zombie SKUs” will increase.
Do nothing. Performance will rise.