“Make a better video” is a production request. “Show the setup process because buyers are unsure how much work installation requires” is the beginning of a hypothesis. The second statement gives the team a reason to create the asset and a way to learn from its performance.
A creative hypothesis does not have to sound academic. It should be short enough for a writer, designer, buyer, and analyst to use together. The detail belongs in the supporting brief, where the comparison and evidence are made explicit.
Start with a real customer uncertainty
Use customer interviews, sales notes, support questions, product reviews, or observed behavior that the business is authorized to use. Record the source and distinguish a recurring pattern from one person's unusual experience.
For an illustrative home-organization product, customers might ask whether it fits in a narrow cabinet. That concern suggests a demonstration of dimensions and use. It does not automatically justify a claim that the product fits every cabinet.
If the concern is only an internal guess, label it as a hypothesis about the audience. The test can still be useful, but the team should not describe invented customer research as an established insight.
Describe the mechanism
Explain why the proposed creative could change the customer's decision. A demonstration may reduce uncertainty. A comparison may clarify a tradeoff. A credible proof point may address skepticism. A clearer qualification may help unsuitable buyers self-select out.
The mechanism connects the creative choice to behavior. Without it, the team may conclude only that a particular color or format “worked,” even when the meaningful change was the information communicated.
Keep the mechanism specific to the offering. Generic claims about human attention can be too broad to tell the designer what belongs in the actual ad.
Use a one-sentence template
Write: “For this audience and situation, showing this message or evidence will change this behavior because it addresses this uncertainty, compared with this control.”
An illustrative version could read: “For first-time buyers considering the storage insert, showing it measured inside a narrow cabinet will improve qualified purchase response because it resolves fit uncertainty, compared with the current lifestyle-only creative.”
This is a proposed explanation, not a result. Keep the expected direction separate from the observed outcome after the test.
Add the evidence record
| Brief field | What it should answer |
|---|---|
| Audience context | Which customer situation makes the concern relevant? |
| Source evidence | What observation suggested the concern? |
| Creative change | What will the viewer see or understand differently? |
| Control | What is the actual comparison asset? |
| Primary outcome | Which business response will guide the decision? |
| Diagnostic metrics | What helps explain how the response changed? |
| Falsifier | What evidence would weaken the proposed explanation? |
| Next action | What will the team do for each plausible outcome? |
Use the control-selection guide to avoid choosing a comparison that differs in several unrelated ways.
Define a falsifier before the result
A hypothesis should be capable of being unsupported. If every result can be explained as a hidden success, it will not improve decisions.
For the cabinet example, stronger early engagement with no improvement in qualified purchase behavior may weaken the claim that the demonstration resolved a buying barrier. Customer feedback showing that the demonstration remained unclear may suggest an execution problem rather than a completely wrong concern.
Write the distinction in advance where possible. It helps the team decide whether to revise the asset, abandon the mechanism, or run a clearer comparison.
Match the design to the strength of the claim
NIST's introduction to completely randomized designs explains the role of assigning treatments to experimental units. Ordinary ad delivery is not automatically that kind of experiment.
Use a supported experiment when a causal comparison is needed and feasible. If the test is an operational rollout with unequal delivery, keep that limitation in the result. The hypothesis can still organize learning, but the report should not claim more isolation than the design provides.
Define what stays stable: offer, destination, audience context, conversion goal, and relevant campaign settings. If those also change, describe the test as a broader package.
Choose diagnostics that follow the mechanism
An opening-message hypothesis may need attention and early engagement diagnostics, but the business outcome still matters. The hook test guide separates the opening's role from the rest of the story.
A qualification hypothesis may intentionally reduce click volume while improving lead usefulness. A product demonstration may affect purchase confidence more than immediate engagement. Select metrics that fit the proposed mechanism rather than using the same scorecard for every creative idea.
Keep definitions consistent and wait for the relevant outcome to mature. A short-term click improvement should not be rewritten as a confirmed revenue improvement.
Make production serve the question
Give the creator the evidence to include, the claim boundaries, the control asset, and the exact variable being explored. Allow creative judgment within that scope.
If production reveals that the intended demonstration cannot be made truthfully, revise the hypothesis or the asset. Do not substitute a synthetic depiction that implies evidence the product does not have.
Before launch, review whether the finished ad still tests the original idea. A visually attractive asset can drift away from the customer uncertainty that justified it.
Save the learning in context
After review, record the hypothesis, actual asset, delivery context, result, uncertainty, and next decision in the creative learning repository.
Retain unsupported ideas as well as successful ones. They help the team avoid repeating the same test under a new filename. A useful hypothesis system turns creative production into a sequence of better questions and better-supported decisions.
Download a working test brief
The creative testing template turns the hypothesis into an editable plan with assignment, outcomes, timing and a decision rule. Before launch, use the A/B sample-size planner to check whether a two-arm binary-outcome experiment is feasible.