Pausing the ad with the worst CPA feels like straightforward account management. The harder question is what happens to the rest of the account after it disappears.
A creative might help someone understand the product before another ad gets conversion credit. It might also be wasting money. The attributed results alone do not settle that decision.
The useful operating question is: Does keeping this creative improve the economics of the set it belongs to?
Where the hypothesis comes from
Cody Schneider described an observation on X: removing two ads increased another ad’s CPA, and restoring them brought it down. He interpreted the pattern as evidence of Meta sequencing ads so that some creatives help another close the conversion.
That is a useful hypothesis, but the observation does not establish the mechanism. The account includes no spend, conversion volume, test duration, uncertainty, or total campaign results. Removing ads could also change delivery, audience composition, auction exposure, or where conversion credit lands.
There are two separate questions here:
- Can creatives influence how people respond to other creatives?
- Did these particular creatives create enough additional customer value to justify their cost?
The second question should drive the budget decision. You can investigate it without proving the exact exposure sequence.
Define the decision before designing the test
Start with a specific choice: keep a defined group of support creatives, or concentrate the budget on a conversion-focused creative.
Write down what each support creative is supposed to contribute. For example, one might explain the problem, another demonstrate the product, and another address a purchase objection. These are proposed roles, not evidence that the ads perform them.
Then specify what would make the portfolio worth keeping. If the goal is efficient acquisition, the portfolio should produce better total new-customer economics at a comparable budget. If the goal is additional volume, define how much deterioration in acquisition cost or contribution margin you are willing to accept.
This prevents a familiar escape hatch: defending every weak ad as “awareness” after the numbers disappoint.
Use this approach when the spend and decision are substantial enough to justify a controlled comparison. A tiny ad with little delivery may not warrant a dedicated experiment. A large budget allocation deserves more scrutiny than an explanation about possible assists.
Compare the portfolio with a clear alternative
A practical starting design has two groups:
- Single-creative group: a stable conversion-focused creative.
- Portfolio group: the same conversion-focused creative plus a predefined set of support creatives.
Use randomized, mutually exclusive audience groups where feasible. Keep the offer, optimization event, budget framework, placements, and measurement definitions comparable. Avoid material creative edits during the comparison.
At equal budgets, this tests whether allocating some spend to support creatives improves the overall result. It does not isolate the persuasive effect of each support ad. Delivery allocation is part of the strategy being evaluated.
Set the test duration and decision rules in advance, based on expected conversion volume, the improvement worth detecting, and the delay between exposure and purchase. Include a rule for how initial delivery instability will be handled. Extending the test until a favorable result appears weakens the comparison.
If you cannot create comparable groups, a removal-and-restoration exercise can still generate a lead. Record concurrent promotions, budget changes, and other disruptions. Treat the result as directional because the periods may differ for reasons unrelated to the creative change.
Measure the whole group, then name the metric correctly
Track total spend and new customers for each group. Calculate acquisition cost using a consistent customer definition and observation window. Where relevant, follow those customers into activation, retention, or contribution margin so a cheaper acquisition does not hide lower customer quality.
Ad-level CPA remains useful for understanding delivery and attribution. It should not be the sole verdict on the portfolio.
Be precise about incrementality. Dividing group spend by recorded new customers gives an acquisition-cost measure; it does not automatically give incremental CAC.
To estimate incremental CAC for a strategy, you need a credible estimate of how many customers would have converted without that advertising. A suitable no-ad holdout can provide that counterfactual. Incremental CAC then compares the strategy’s spend with its estimated additional customers.
A randomized portfolio-versus-single-creative comparison answers a narrower question: which strategy produces better measured outcomes under the tested conditions? It does not, by itself, establish how much either strategy adds over no advertising.
Prefer business-level customer outcomes where measurement permits. If your conclusion relies only on attributed conversions, attribution redistribution remains a limitation.
Decide from the portfolio result
Agree on the possible decisions before looking at the final result.
If the portfolio improves total new-customer economics and meets the business guardrails, keep the tested set. Describe the finding at the level the experiment supports: the portfolio performed better under those conditions. Do not claim that every support creative was necessary.
If the closing ad’s CPA improves but total outcomes do not, you have no demonstrated business benefit from that movement alone.
If the portfolio adds customers but worsens acquisition cost, evaluate the tradeoff against the volume and margin rules set before the test.
If the result is inconclusive, say so. Choose the next action using the cost of continued uncertainty, the available budget, and the feasibility of collecting more evidence. An inconclusive test is neither proof of an assist nor proof of waste.
Watch for the failure modes
Changing several things at once. A new offer, budget increase, and creative removal in the same period make the result difficult to interpret.
Assuming a sequence was delivered. Assigning educational and conversion roles to ads does not demonstrate that people saw them in that order. An exposure path can suggest a mechanism without proving its causal effect.
Protecting weak ads indefinitely. A support hypothesis needs a decision date and a measurable business outcome. Otherwise it becomes a permanent exemption from scrutiny.
Overreading the winning set. A portfolio win does not identify which support ad mattered. Follow-up removal tests can investigate that question when the potential savings justify the work.
Ignoring delivery and customer quality. Review spend, reach, frequency, delivery stability, and downstream value alongside acquisition cost. These checks help explain the result and identify limits to scaling it.
Evidence and limitations
This playbook draws on Cody Schneider’s reported removal-and-restoration observation and the hypothesis that creatives can influence one another’s results. The observation is anecdotal and lacks the campaign details needed to assess causality or effect size.
The test design and decision rules above are operator recommendations, not reported results. They do not establish that Meta universally sequences ads or that creatives with poor attributed CPA deserve protection. Their purpose is to make a specific budget decision testable while keeping attribution, mechanism, and business value separate.
Source basis
- Cody Schneider’s public X anecdote about CPA changes after removing and restoring other ads.
- Critical assessment of the anecdote’s missing evidence and alternative explanations.
- Original experimental guidance for comparing creative portfolios and interpreting acquisition outcomes.