Zombie SKU Performance Max campaigns are a bad idea. This type of campaign is currently en vogue with many advertising agencies and we’ve heard reports that Google Ads representatives are pushing this idea on advertisers as well. In this article, we’ll explore the shaky foundations of this idea and consider why these campaigns don’t work the way they’re being presented to you.
What is a Zombie SKU PMax Campaign?
The basic idea is that sometimes Performance Max campaigns make mistakes and stop advertising products which could be making money. These products end up not getting any clicks and rarely enter ad auctions. Since the Performance Max campaign is not pushing these products, the idea is that the items should be kicked out of the campaign they are in, and put into another campaign. Then, that new campaign pushes only those “zombie SKUs”. And perhaps we set more aggressive targets in order to kick start the process, spending more than we did in the past on these items.
Is There Evidence That This Works?
If you perform a search for “zombie sku pmax” you‘ll find many articles promoting this idea. What they have in common is that an advertiser set up a zombie SKU campaign and then watched it over time, and the products which previously did not receive any clicks were put in a campaign that went on to perform moderately okay to actually pretty well, boosting sales by some amount. Some articles say that by getting zombie products to finally sell, these campaigns lifted sales by 5% or some other amount over a few months or a year.
We’re going to spend the rest of this article exploring why simplistic explanations like this are probably wrong. In some future articles we’ll explore some real data and see if we can make a solid case that zombie SKU campaigns definitely don’t work. Let’s dig in!
Why We Think Zombie SKU Campaigns Probably Don’t Work
In a nutshell, the arguments for zombie SKU campaigns appear to be based on data evaluated with a lot of statistical biases. Nobody we’ve run across makes a solid case that these campaigns achieve better results. Given how modern AI-based campaigns actually function, it is risky for advertisers to separate their products into more campaigns with smaller amounts of data. Smaller campaigns with more fragmented data tend to perform worse than larger consolidated campaigns. As long as data drives the machine learning systems that power ad results, more data in a campaign is better.
Those who push this idea aren’t really digging into questions around why products don’t get clicks in Google Ads. They aren’t exploring issues around the distribution of clicks across populations of products (80/20 rule). They aren’t considering selection biases in how they’re interpreting the data. I haven’t seen a single article consider how new products on a site might skew the results or how competitor behavior could cause some products to lose ad auctions some of the time but bounce back later—entirely outside of your control.
So let’s explore some of these biases and how they might impact the zombie SKU campaign idea.
Why Shopping Ads Don’t Get Clicks
If a product isn’t getting clicks, that could be due to several factors. It’s good to review how Google Ads auctions work, and what it takes to win them. The highest Ad Rank wins. Basically, it’s about the quality of your ad for a given search, coupled with how much you’re willing to pay for that click.
The simplest explanation for why you have “zombie SKUs” is not that Google Performance Max campaigns are making a mistake. It is that you have bad product feed data and that you aren’t bidding aggressively. You have competitors with better product data feeds who are bidding more aggressively than you are. You may simply be losing on data quality and financial issues, not because the campaign is broken.
For a deeper dive into these issues, check out Why Isn’t My Product Showing on Google Shopping Ads? The key idea to take away is that how well your ads stack up against your competitors’ is what controls whether your ads show up. This is a competitive system. If products aren’t winning auctions and getting clicks, you might have some work you need to do on your product data, landing pages, pricing, etc.
80/20 Rule – The Pareto Principle
Across any business, a tiny number of SKUs are going to account for most of the sales and revenue. And this is not unique to digital advertising. There are a lot of products in the long tail of searches that simply aren’t as popular.
A popular search term could have many thousands of possible SKUs across hundreds of advertisers with a reasonable ad to show. If you have quality data and bids that are similar to your competitors, your products are going to show up often.
But what about the long tail searches? There are searches that only happen maybe a hundred times per month on Google in the USA. And there may be dozens of companies competing for those searches, each with several products that would be reasonable. In these cases, only some advertisers are going to win each month. Only some of their products will show. Most won’t get enough ad impressions to get very many ad clicks, and only a few of those clicks will convert.
A lot of these products, for a lot of these advertisers, are simply not going to get any clicks or conversions some months. Other months you might get a conversion. The fact that a product didn’t get any clicks in January really isn’t important in cases like this. You’ll get clicks and sales other months if you’re competitive.
But the deeper issue is that any large Google Ads account will have some products that simply perform poorly some of the time but bounce back other times. The fact you didn’t get a click last month doesn’t mean this is a “zombie SKU” that you need to bring back from the dead. It may just be a long-tail SKU that is profitable, just at a low volume, and some of the time.
Competitors Making Poor Decisions
It gets even more interesting in search ads. Demand for some products ebb and flow over time due to seasonality. Our key competitors might be brilliant or they might not be.
Strange effects happen when we have competitors with flat monthly budgets throughout the year, but demand for the products we both sell rise and fall in some seasonal pattern. This can cause situations where competitors are overbidding for results some months of the year but underbidding other months of the year. If we’re more enlightened and bid towards actual profit targets, we might stop winning auctions on some products during certain times of the year due to competitors wasting money on unprofitable bids that we aren’t willing to match, because spending at the level would be unprofitable for us. But at other times we might win a lot of auctions at a reasonable cost per click and make profitable sales because our competitors are underbidding due to their financial mistakes.
So, the reason we have some products not getting any clicks from time to time may just be due to fluctuations in opportunity coupled with the dynamics of how ad auctions work when competitors spend fixed budgets that cause their actual bids to spike up and down over time.
Bid Strategy Optimization Biases
We are not optimizing for clicks. In ecommerce advertising we are generally optimizing for return on ad spend (ROAS) goals or seeking to maximize conversion value (MCV). How do we get conversion value that drives MCV or tROAS bid strategies? We need to get an ad displayed, by winning an auction. Then a user needs to click our ad rather than a competitor’s. Then that person needs to make a purchase so we can measure some conversion value.
So our campaign bid strategy controlling the actual bids in ad auctions plays a huge part in whether our ads show up or not. If we get clicks at a high cost, but our conversion rates and average order value are low, we don’t meet our target ROAS goals and our bids get turned down. When they get turned down, we win fewer auctions and our ads don’t appear as often. There’s a snowball effect up or down based on whether we’re meeting our ROAS targets or not meeting them.
If we’re not getting clicks on some products when we’re using a Smart Bidding strategy like tROAS or MCV, that’s a clue that those products simply aren’t capable of delivering the same financial results as other products. This is not a problem with Performance Max campaigns. This is exactly how they are supposed to work–delivering financial results and reducing costs on what is not working. If a PMax campaign stops advertising some items, it’s because advertising those items is not meeting our conversion goal requirements, so they should stop being advertised.
But remember, as competitor behavior and consumer demand shifts, this may change. Then you can again see results from products that previously weren’t winning auctions, getting clicks, and making sales. Products that are on the edge of meeting your performance goals can bounce in and out of acceptable performance. It is still better to keep your products grouped as a larger campaign than to fragment that data.
Rescuing Disapproved Products in Google Merchant Center
If you have products that for any reason got disapproved in GMC, any analysis you do of zombie SKUs will include those items, because they aren’t getting clicks. But some of those products might be viable with a little attention. For example, GMC may disapprove some very popular products for a problem like tax calculations or shipping rate discrepancies. Once you resolve that technical issue, they can bounce out of non-performance back to high performance.
This is one of the obvious oversights in typical articles online about zombie SKU PMax campaigns. People advise regularly reviewing all items receiving zero clicks in a recent time period and shifting them into the zombie SKU PMax campaign. If you aren’t watching for this, you may inadvertently shift high performing products that GMC simply disapproved into the new campaign. There, they will revert to being high performing products and you will mistakenly think that you resurrected a zombie SKU. In fact, you just diluted the effectiveness of your original campaign by fragmenting your data.
New Product Bias
Any product freshly added to your feed today will of course not have received any impressions, clicks, or conversions in the period of time leading up to today. This is a huge blind spot with people promoting the idea of zombie SKUs.
Some of the results they claim to get are due to new products, recently added to the catalog, now being added to the zombie SKU campaign since they didn’t “perform well” in the past simply because they didnt’ actually exist in the past. Some of these new products will go on to perform great in the zombie SKU campaign, but they would have performed great in the original campaign as well.
Google Ads Shopping Reports Biases
To evaluate the SKU-level performance of items in Google Ads, we generally pop into the Google Ads Report Editor and pull down a Shopping Report by Item ID. But these reports are not what they seem.
Shopping Ads reports do not report on what SKU a customer actually bought. They report on ad performance based on what ad the user clicked on. If the user clicks on an ad for Product A but then navigates around the site and buys Product B, Product C, and Product D, all of the sales statistics reported are attributed to Product A. This can seriously skew results for many advertisers who misinterpret this data, because the reports don’t show the sales of Products B, C, or D, making them look dead in comparison.
Your supposed zombie SKUs might be doing just fine, but you might have some other products with better imagery and pricing that are the ones drawing people to your site. With some advertisers we see huge biases in the reporting that can really confuse analysis if you don’t keep this distinction in mind.
Another reporting bias is that these reports don’t generally show data for products that didn’t win ad auctions and get impressions. You can pull more rich reporting via the API. But just reviewing a report of products with zero clicks in a time period won’t give you the full list. You’re going to get a list of items that did win ad auctions, but subsequently weren’t clicked. Your report will miss the items that lost ad auctions so they had no chance of getting a click.
If you used one of these reports to then shift items that did receive ad auctions but didn’t get clicked into a zombie SKU campaign, it should be obvious that this could cause some statistical biases. These are actually products winning auctions and more likely to get clicks in the future.
Is This Just a Selection Bias?
If you look at products that did not receive any clicks recently, the fact is that some portion of them will actually go on to get clicks and make sales in the future, whether you change anything or not. If you just kick them all out into another campaign, then what?
It’s easy to interpret the results of the new campaign as an increase in performance when there is actually zero evidence for that. You could simply be seeing what would have happened anyway. Many products you push into a Zombie SKU campaign actually would have performed better in the future even if you hadn’t taken this action–which we will demonstrate in another article.
The analysis that matters is whether taking an action like this actually improves performance over the baseline of what would have happened if you had done nothing. That’s a harder thing to prove. To do so, you’d need to do a more controlled trial, perhaps by using an A/B split of products you think are Zombie SKUs, and seeing what happens if you leave the A group in the current campaign while you split the B group out into a Zombie SKU campaign. A randomized controlled trial like this is something that NOBODY who is promoting the idea of Zombie SKU PMax campaigns is doing. So we’re going to do it, and we’re going to report on the results. Stay tuned.
If people are telling you to pull Shopping Reports from Google Ads about *products* with zero clicks, and then telling you to put them into a new PMax campaign, and then to evaluate the performance of the new *campaign*, they are actually making a ridiculous logical leap. It makes no sense. You’re first evaluating product, not campaign, performance. And then you’re comparing it to campaign, not product, performance. Performance Max campaigns don’t simply show shopping ads. They show search ads and display ads and video ads. You could put a bunch of crappy products in a PMax campaign, then evaluate campaign performance, but the entire results of that campaign were actually from those other kinds of ads. And this, really, is where you could really go wrong. If someone convinces you to drop your target ROAS on a Zombie SKU campaign to “kick start” those products that aren’t getting clicks, what you’re really doing is over-spending on remarketing, search, and video ads in that PMax campaign that just happens to be assigned those under-performing shopping ads products.
Zombie SKU Performance Max Campaigns Are Highly Suspect
Because of these sorts of biases and misinterpretations of data, it is impossible for zombie SKU PMax campaigns to work the way many people claim. If someone says that such a campaign raised their sales by 5% or 22%, the reality is that much or all of that lift could be a misinterpretation of what was actually going on.
To be completely fair, it’s possible that there is a slight positive impact from zombie SKU PMax campaigns. It’s just not possible that the impact is as high as people make it out to be. I want to stay humble about that and test the idea, so our next couple of articles will do just that and present real data about what we observe in the real world.
But… It’s also possible that these campaigns reduce performance by shifting products into new campaigns where they actually perform worse. And Google Ads accounts with fragmented data in smaller campaigns do tend to perform poorly. There is a very real risk that zombie SKU campaigns are a terrible idea that actually reduces overall account performance. And even if Zombie SKU campaigns performed just as well not doing them, if we spend time and money making them, and they only do as well as not doing them, that is a net loss. Unless Zombie SKU campaigns are provably better than simply taking no action at all, they are a waste of time and money.
What’s Next?
So how could we prove that zombie SKU campaigns actually increase an account’s conversions and revenue? We would have to run a trial to demonstrate that these campaigns outperformed the simple act of doing nothing while trying to avoid some of the biases we explored in this article. Nobody is doing this. So we’re going to run an experiment. More soon!