Creative

You test creatives, but afterwards nobody knows what won

A test only holds when it is settled beforehand what gets compared and what decides. In practice this fails on three things: several things get changed at once, the runtime is too short for a statement, and there is no rule for when a difference counts as a difference. The platforms deliver the volume once enough variants exist; Meta explicitly recommends a wide variety of ad creative meta-adv. The method you have to bring yourself.

What we do: We ask one question per test, calculate volume and runtime beforehand, agree the decision rule and record every result; production effort becomes knowledge that counts again at the next creative.

By Robin Kardinal, Chief Operation Officer (COO)

Last updated on 5 October 2026

What it is about and when it becomes relevant

What this is about

A creative test answers a question about a creative, not about a campaign. The question can be: does the message about the price work better than the one about delivery time? Everything that varies alongside makes the answer unusable. Where platforms reward many variants meta-adv, discipline in the setup becomes more important, not less.

When it becomes relevant

As soon as budget sits in channels where the creative carries the performance. Then the quality of the tests decides whether production effort turns into insight or just into material.

What a test is not

Running two creatives one after the other is not a test, it is a time series. In between, the market, the weather and the competition change. Comparison happens simultaneously or not at all.

Common mistakes and how you spot them

Several changes at once

How you spot itThe new variant has a different image, different copy and a different landing page. If it wins, nobody knows why, and the next test starts from zero again.

Test stopped early

How you spot itAfter three days a decision is made because one variant leads. At small volumes that lead is usually chance.

No threshold agreed

How you spot itAfter the test, people argue whether the difference is big enough. That discussion belongs before the test, not after.

Results land nowhere

How you spot itThere is no record of past tests. The same question comes up again half a year later and gets tested again.

How we go about it

One question per test
We write down what gets compared and what stays the same. One variable, one result.
Calculate volumes and runtime beforehand
We estimate before the start how much delivery is needed for a difference to become visible at all, and set the runtime from that.
Fix the decision rule
From which margin a variant wins is agreed in advance, so the outcome does not get negotiated.
Record the results
Every test lands in a register with question, setup and result. Over months that becomes a body of knowledge instead of a string of one-off attempts.

Frequently asked questions

How many variants do I need for a test?

As many as the question demands, usually two to four. For delivery as such, Meta recommends a wide variety of ad creative meta-adv; but that is feeding the automation, not test design.

How long does a creative test have to run?

Long enough for the expected volume to accumulate. Any fixed duration would be a guess, because it hangs on budget and audience size. The calculation belongs before the start.

Can I just run creatives one after the other?

That produces a time series, not a comparison. Between the periods, market and competition change, and the difference cannot be pinned on the creative.

Who decides in the end?

The rule agreed beforehand. If the rule only emerges after the result, the variant somebody wanted anyway always wins.

Is this exactly where you are stuck?

Write to us about where it is stuck. Robin Kardinal and the team will look at what is concretely behind it in your case, and tell you where the biggest lever sits.

Book a meeting

Sources

Every statement about ad products comes from the official documentation of the respective platform, with the date we retrieved it. Products change; if something looks outdated to you, write to us and we will check it.