Advertising creations: organize tests that answer a question

Meta Ads · 6 min read

To run useful ad creative tests, formulate a question, choose what will change between versions, and define the outcome that will guide your decision. Two different ads do not automatically constitute a reliable comparison. The test must make it possible to learn something precise, even when the data does not separate the versions.

Key points

• A test starts with a hypothesis, not a collection of visuals.
• Keep items that are not part of the question constant.
• Choose an indicator linked to the business objective.
• Conclude that the test does not decide between versions when the results are insufficient.

Transform a hesitation into a testable question

Write what you want to understand

“Which advertising works best? » is too broad. Instead, ask: does a product demonstration help drive purchases more than a product presentation alone? Or: Does an explanation of product compatibility reduce off-topic requests? These questions connect content to useful behavior.

The presentation of A/B tests of Meta recommends choosing a hypothesis before the variable. Note your expectation, but also the observation that could contradict it. Otherwise, you risk looking only for what confirms your creative preference.

Delimiting what changes

The documentation Meta on split testing recommends only one variable per test. If you change the visual, keep the text, the offer and the destination. If you are comparing two complete concepts, think of the concept as the variable: you will be able to choose between them, but not attribute the difference to a particular color or phrase.

Prepare truly comparable versions

Create a test sheet indicating the versions, the elements kept and those modified. Specify the advertising objective, audience, destination and distribution conditions. Also check the customization or automatic transformation options: if they modify the versions differently, record it and limit your interpretation.

Check the quality of each creation before distribution. Unreadable text or a video whose beginning is cut transforms the test into a comparison between a usable file and a defective file. The variants must be clearly identified, with their final exports and their validation.

Distinguish current diffusion and controlled test

Placing multiple ads in a campaign provides useful insights, but does not prove that each ad encountered comparable conditions. Do not infer general superiority from a simple discrepancy in the circulation ratio.

In its A/B device, Meta distributes versions between separate segments, so that a person does not see both versions of the test. Meta also recommends identical budgets for a fair comparison. Check the options available in your account; a manually duplicated campaign does not necessarily reproduce this feature.

Choose the indicator before seeing the results

A main criterion and useful controls

For a purchasing-oriented campaign, use a criterion linked to purchases rather than just reactions to the ad. For a commercial requests campaign, look at eligible requests and their cost, not just the total forms. Clicks or views can help explain a journey; they do not replace the desired action.

Document the counting method and keep the same measurement settings across versions. Check for cancellations, off-target requests, and technical issues. The Meta documentation recommends setting the flags before creating the test.

Predict when a decision will be possible

Set an observation period and an authorized expense adapted to your situation. Define the expected confidence level and decision criterion before launch, with competent assistance if necessary; do not choose these rules after seeing the results. Wait until the events have had time to be recorded. Do not shorten the test as soon as a version takes a momentary lead. If useful actions are too rare, conclude that the test remains undecided. If the distribution is unbalanced, check whether this imbalance was expected and consider group sizes and expense before comparing results. An incurred expense does not automatically create proof.

Fictitious example: showing the use of a bag

This illustrative example concerns a fictitious bag store and does not present any customer results. The team wants to know if showing the interior storage helps trigger a purchase more than showing the closed bag.

She prepares two photographs of the same model: one shows the closed bag, the other shows its interior with everyday objects. The text, the offer and the product pages remain identical. The main criterion concerns purchases recorded using a common method. If the data favors a version in an actionable way, the team can retain that presentation. She cannot conclude that all products must now be photographed open.

Checklist for launching a creative test

1. Write the question and hypothesis.
2. Define variable and constant elements.
3. Check the files and possible transformations.
4. Choose a suitable comparison device.
5. Fix the main indicator and counting method.
6. Plan period, budget and conditions for technical shutdown.
7. Record the conclusion, its limitations and the next decision.

Common errors and test limitations

Changing the offer during the course, comparing different commercial periods or subsequently selecting the most favorable indicator distorts the reading. A new discount can explain a gap without the creation being better. Even if correctly carried out, a test answers a question in a given context; it does not certify future performance.

Frequently asked questions

Can we test a video against an image?

Yes, if the question concerns these two propositions. If their content also differs, the conclusion will relate to the complete proposals, not to the format alone. To isolate the format, bring their message and information as close as possible.

How long should a test last?

There is no valid duration for all situations. The budget, frequency of actions and buyer decision time matter. A past period is not enough if the data remains too poor.

Is a version with more clicks a winner?

Not on their raw number: we must also compare the conditions of exposure and the expense. Clicks can guide the decision if they constitute the chosen criterion and meet the objective. To sell or get qualified inquiries, more clicks may not produce more useful results.

Next step

Choose a question to decide on your advertising creations and fill out a test form before producing the variations. If you need help framing this comparison, the services page 4SYSTEM presents paid advertising support.

Sources and resources

• Meta: presentation of A/B tests, for the hypothesis, distinct segments and comparable budgets.
• Meta for Developers: split testing, for the single variable and the indicators defined upstream.