How We Found Yizzam's Real Buyers in a 250,000-SKU Catalog and Held a 5x ROAS for Five Years
By Kevin Veitia
- Client
- Yizzam, an online store for all-over-print apparel
- Channels
- Meta Ads (Kevin), Google Ads (Michael Reiss of Advertology), Reddit Ads as a test
- Timeframe
- 2015 to 2020
- Scale
- About $150,000 a month in marketing spend
The challenge
Yizzam sold all-over-print shirts: loud patterns of burgers, pickles, cats and fine art, many built from public-domain images. They were ugly on purpose, and that was the charm. The catalog ran past 250,000 SKUs.
A catalog like that has no single customer. The person who buys a pickle shirt is not the person who buys a cat shirt, and neither of them is the person who buys a Klimt. Spread a budget across all of it and no product gets enough spend to prove anything. You end up with a mountain of data and zero answers.
Where SUREFIRE was born
Yizzam ran from 2015 to 2020, and it's where Kevin first ran paid media as a system instead of a string of experiments. He didn't do it alone. His mentor, Michael Reiss of Advertology, ran Google Ads on the account while Kevin ran Facebook and Instagram, and together they built the testing method they called SUREFIRE. Michael took Kevin under his wing and gave him the room to prove himself on this account, and Kevin credits him with teaching him most of what he knows.
Before iOS 14.5, the biggest lever in a Meta account was the audience, not the creative. So SUREFIRE started there. We wrote one control ad, a minimum viable ad, and ran it unchanged across competing targeting models: detailed interests, lookalikes, broad and hyper-targeted. With the ad held constant, any difference in results came from the audience. Only once a winning audience emerged did creative testing begin. The same method went on to work for Canva, Ground News and Jenfi.
What the testing found
The buyers with real pull weren't who anyone would have guessed: art teachers, mostly women over 35, buying the fine-art designs. Klimt was a bestseller.
It makes sense once you see it. An art teacher wearing a Klimt isn't buying a shirt. She's buying a small piece of her identity. Demographic targeting would never have found that. A controlled test across audiences did.
Once the product and the audience matched, the creative and messaging could get specific, and the budget could go to the combination that had earned it.
Reddit, as a test
We ran Reddit the same way: the same ads, tested across two types of targeting. The winner reached roughly 2.75x ROAS. That's respectable, but well short of Meta and Google, so Reddit stayed a side channel and the budget stayed where it earned more.
The test that changed how we buy media
At one point, Yizzam's owner asked a fair question. If the art-teacher audiences were proven, why not optimize those campaigns for link clicks instead of purchases? Traffic would be cheaper, and the same people should buy at the same rate.
We cloned the campaign and ran both versions as an A/B test. In Google Analytics, the link-click version's bounce rate went from about 55% to 99.99%, and time on site fell from two and a half to three minutes to effectively zero. Same audience, same ads. The only change was what we asked Meta to find, and Meta found exactly that: something that clicks and leaves. That test is why we don't run link-click campaigns.
Results
- About 5x ROAS on average across five years, with some months reaching 7x
- $10 cost per acquisition
- Around $150,000 a month in marketing spend at scale
- Roughly 2.75x ROAS on Reddit, as a secondary channel
Out of a quarter of a million products, Yizzam ended up with something worth more than a long catalog: a proven buyer, a proven way to reach her, and a testing method that outlived the account.
What you can take from this
- When the audience is the bigger lever, find it first. Hold one ad constant and let the targeting models compete.
- Look for identity, not demographics. Art teachers bought because the designs said something about who they are.
- Test a new channel the same way you test an ad. Reddit got the same method, and the numbers said keep it small.
- Never optimize for an action a bot can perform. Switching to link clicks sent bounce rate to 99.99%.
Want a second set of eyes on your account? Book a free 30-minute consultation and we'll walk through what's working, what isn't, and where the next win is. Curious what it costs first? Our pricing is public.
Book a Free Consultation