Turn the product benefit into a hypothesis
Decide what you need to learn about the app's message before requesting another set of assets. A useful hypothesis connects a product benefit, an audience need, and an observable action. Specify what someone should understand from the ad and what evidence would justify continuing with that message. This makes the creative brief a decision document as well as a production request.
Start with a benefit that the current app can deliver. Check the actual feature, any access conditions, and the point at which a new user experiences it. If the feature requires a paid plan, the creative and destination should make that condition understandable. Keep claims approved by the product owner together with the brief so revisions do not quietly introduce a different promise.
For a proposed test, you might compare a demonstration of the useful outcome with an explanation of how the feature works. That is a hypothetical creative question, not evidence of either approach performing better. Write the expected behavior and the reason it would matter to the buyer. An attractive opening has limited decision value if the test cannot connect it to the intended app action.
Use the creative testing workflow to connect the brief, production, launch, and review. Assign an owner to each stage. The buyer should approve the question and measurement basis before production begins, while the producer should confirm which differences can be made and reviewed consistently.

Change one meaningful variable
Define the contrast you want to interpret. A hook test can compare two opening ideas while retaining the product demonstration and destination. A format test can keep the benefit consistent while comparing different ways of showing it. Write down the elements held constant and the elements deliberately changed.
Do not call every new export a new concept. A crop adjustment may be a delivery adaptation; a different benefit may require a new hypothesis. Keep that distinction in the log so production volume does not become a substitute for learning. If localization changes the message substantially, describe the local hypothesis explicitly rather than presenting it as a literal duplicate.
Creative test matrix
Use one row per planned variant. The placeholders are fields to complete, not campaign observations.
| Hypothesis | Hook | Format | Market | Variant | Decision |
|---|---|---|---|---|---|
| State the audience need and expected action | Describe the opening idea | Name the production format | Record market and language | Enter the asset ID and version | State what would justify continuing |
| State the competing explanation | Describe the alternate opening | Keep or deliberately change the format | Preserve the comparison scope | Enter the comparison asset ID | State the next action if unresolved |
Confirm that the comparison is possible in the buying setup. Google explains that App campaign ads combine supplied assets. A platform assembling and distributing combinations does not automatically create a controlled experiment between two files. Record the campaign arrangement, exposure differences, and any available experimental controls before making a causal claim.
For bundled changes, describe what the bundle contains. A new hook, voice, and screenshot sequence may be a sensible production direction, but its result cannot identify which individual element caused the difference. The next test can isolate that question if the decision warrants it.
Name the variant before it ships
Assign a stable ID before uploading an asset. Keep the hypothesis, benefit, hook, format, language, aspect ratio, version, and destination in the record. Use a readable file name for everyday work and a structured matrix for the complete specification. The file name alone should not carry every detail required for review.
Save the approved script or storyboard next to the delivered file and upload reference. Record substitutions, rejected assets, and any platform edits. If a replacement changes the message, issue a new version and note its first delivery time. A reviewer should be able to match an exported reporting row to the exact creative people could have seen.
Check the production scope before treating assets as equivalent. Video, static, and interactive creative require different QA questions. For video, verify captions, safe areas, audio assumptions, and the final action. For static images, review readability at placement size and whether the benefit is understandable without supporting motion.
For a playable, define what counts as starting the interaction, completing the intended action, and leaving for the destination. Use the playable evaluation definitions to separate those interactions from an app install or paid event. An interaction inside an ad is supporting evidence about the ad experience; it is not itself proof of useful activity in the installed app.
Record the person who approved each language and the device or preview used for QA. Keep the destination version with the variant. If the store page promises something different, the test has changed the journey as well as the ad.
Read creative signals with post-install quality
Choose metrics that fit the question. Click-through rate describes clicks relative to impressions under the report's counting rules. Cost per install connects spend to the defined install count. Trial and purchase evidence address later actions. Keep those stages separate so a promising attention signal is not presented as a subscription result.
Google's App campaign asset reporting guide notes that an ad can contain multiple assets, so some impressions appear in more than one asset row. It also describes performance labels as relative comparisons. Use the underlying report definition before adding rows or comparing labels. An asset label is not a standalone estimate of the causal effect of that asset.
Compare variants within the intended market, source, platform, and observation basis. Record how much delivery each received, whether the same event was measured, and whether later outcomes had time to develop. If a variant has little exposure, write insufficient evidence instead of treating the absence of purchases as a settled rejection.
Our team's Android photo-app approach separated US and UK campaigns and tested feature angles around avatars, filters, and photo enhancement. The published work covers images, video, and copy. It supports a concrete example of organizing feature-led creative; it does not report playable results or isolate the effect of an individual creative variant.
When signals conflict, formulate the next check. Strong clicks with weak trial starts could justify reviewing message continuity, event collection, or onboarding. Weak clicks with useful downstream activity might justify revising the opening while keeping the benefit. These are proposed investigations. Establish the explanation from the app's evidence before changing the whole production direction.
Keep the decision and next test together
Review the hypothesis rather than selecting the most flattering column. State whether the evidence supports continuing, revising, pausing, or collecting more information. Attach the reporting basis and important limitations to the decision. A later reviewer needs to understand what was known at the time, including defects that affected interpretation.
Review log
| Review field | What to record |
|---|---|
| Hypothesis and variant | Original question, asset IDs, and delivered versions |
| Evidence basis | Report name, source, market, event, date range, and export cutoff |
| Exposure and maturity | Delivery coverage and equal observation time for later events |
| Interpretation | Supported finding, competing explanation, and missing evidence |
| Decision | Continue, revise, pause, or defer, with the decision owner's rationale |
| Next action | Specific change or check, owner, dependency, and review date |
Keep the previous entry when a decision changes. Add a new review with the additional evidence and explain why the action changed. This preserves the difference between an incorrect hypothesis, an incomplete read, and a later shift in delivery conditions.
Creative review checklist
- Confirm the benefit shown is available in the current app.
- Check message continuity across the ad, store listing, and first useful action.
- Verify localization, captions, and any conditions attached to the claim.
- Complete format QA on the intended devices and placement previews.
- Match every reporting row to an approved asset ID and version.
- Record audience, destination, and delivery differences that affect comparison.
- Check event maturity and distinguish missing data from a measured zero.
- Separate platform asset labels from controlled test findings.
- Assign the next action and keep it beside the decision evidence.
End the review with an executable brief: what changes, what stays constant, and what evidence will settle the remaining question. If the current evidence cannot support a production choice, record that boundary and specify the check needed before another variant is commissioned.


