You launched four ads last week.
One is doing fine. Two are underperforming. One is doing something weird: high CTR, no conversions. It’s Monday, and someone on the team asks the question that always comes up at this point:
“So… what do we do next?”
Someone suggests killing the low performers. Someone else thinks the audience is off. Someone wants to try a new hook.
Every suggestion is reasonable. None of them are based on the same read of what actually happened.
It’s not even the underperformers that sting the most. It’s the one ad from three months ago that actually worked, the one everyone still points to, and the fact that no one can tell you exactly why it worked well enough to build another one like it.
By Thursday, your team has thought up a new batch, shaped less by what the last round taught you and more by a general sense that something needs to be different this time.
We see this exact moment play out across almost every account we work on. And it’s rarely the account that’s short on ideas. It’s the account with no system for turning last round’s results into this round’s answers.
“The hard truth is that 75%+ of ads won’t work.”
– Taylor Holiday, CEO, Common Thread Collective [Linkedin]
Ads fail for a lot of reasons. That’s exactly why a system matters.
Some of it is the auction: plenty of ads lose on cost-efficiency alone in a crowded bidding market, regardless of how good the creative is. Some of it is the offer or funnel behind the click. Some of it is just fatigue, an audience that’s seen too much too fast. And some of it, of course, is the creative itself.
Taylor Holiday, CEO of Common Thread Collective (one of the largest performance marketing agencies in ecommerce), has pointed to a version of this from his own agency’s data. Tracking one of his brands, he’s noted that over 75% of the ads they’ve run haven’t outperformed the account mean. His conclusion: you need enough volume in the pipeline to reliably produce winners anyway.
He’s right, and it’s exactly why a system matters more than any single ad. With that many variables in play and that much of your output destined not to be a winner, you cannot afford to treat every ad as an isolated guess. Every ad, win or lose, has to tell you something you can use. Otherwise volume just means guessing faster, and more expensively.
What a real learning system actually looks like
This is the system we build into every account: four shifts that turn each round of ads into a sharper version of the one before it, instead of a fresh guess:
- One variable at a time. Instead of changing everything between versions, change one thing at a time, starting with the biggest lever: the message itself, then narrowing to smaller variables like hook, format, proof, or CTA. That’s the only way to know what actually moved the needle. Change three things and win, and you’ve learned nothing you can repeat. In practice, this isn’t always possible. The production timeline or budget constraints sometimes limit how isolated a test can be, but it’s the standard we build toward on every account, because the closer you get to it, the more repeatable your wins become.
- Volume as a condition, not a goal. One concept produces a set of variants from the start. A test with a single version has a sample size of one. Volume isn’t a vanity metric; it’s what makes the signal readable at all.
- Metrics beyond ROAS. Hook rate. Hold rate. Click-through by variant. Fatigue timing. Each one points to where the ad broke down: the first three seconds, the middle, the offer, or the handoff to the landing page. ROAS collapses all of that into a single number and hides where things actually went wrong.
- Briefs that inherit what the last round learned. This is the step most teams skip. The data from your last batch should be written into the next brief before production starts, not discovered after the ad is already spent.
None of this eliminates underperformance. Most individual ads still won’t beat the best one in a batch, and that’s structural to how testing and the auction work, not a flaw in the process. What changes is what the underperformance produces.
Without a learning layer, a failed ad produces frustration and a scramble. With one, the same ad produces a smaller, sharper next brief. Over a quarter, the difference compounds: one account keeps guessing, the other keeps getting more informed.
The bottom line
Ads will keep failing. That part isn’t fixable, and no amount of talent changes it. What separates accounts that improve over time from accounts that plateau isn’t the ideas. It’s whether there’s a system turning every result, win or lose, into direction for the next one.
That’s the system we run for every brand we work with. If your team can’t agree on why your last batch underperformed, that’s usually the actual gap, and it’s exactly the kind of problem BTS Studio was built to solve. We turn ad performance into direction instead of debate, so every round gets sharper, not louder.