Experiments
Build a creative testing matrix before making more ads
Separate angles, hooks, proof and format so your next ad experiment teaches you something you can reuse.
Ten versions of the same ad do not necessarily create ten useful experiments. If all ten repeat the same promise with a different background, you may learn about visual preference while learning little about why someone buys.
A creative testing matrix begins with a question. For a new product, that could be: does the audience respond more strongly to a practical problem, an emotional outcome or credible proof? Build the first concepts around that question before asking a generator for variants.
Separate four decisions
| Component | Question | Example for a travel organiser |
|---|---|---|
| Angle | Why should this person care? | Find essentials without unpacking |
| Hook | Why should they keep watching? | Show a frantic search before a flight |
| Proof | Why should they believe the promise? | Demonstrate a fully packed organiser |
| Format | How will the idea be delivered? | Short demonstration video |
The example is illustrative. Replace it with a real customer problem and evidence your product can support. Do not invent reviews, before-and-after results or product claims to make a concept more persuasive.
Choose a manageable first batch
Start with three angles and one clear execution for each. Keep the audience, offer and destination consistent enough to interpret the test. Write down what would make each concept successful and what evidence would cause you to reject it.
After you identify a promising concept, explore its hooks or formats. This sequence gives creative production a reason to exist: the next batch answers a narrower question from the previous batch.
Record the hypothesis beside the asset
Use a naming pattern such as product_angle_hook_format_version. A filename should help you connect performance with the idea being tested. Keep the actual brief in your working document; do not rely on a long campaign name to hold all the context.
For each concept, record the hypothesis, intended audience, promise, supporting proof, primary metric and review date. Also record placement constraints. A mobile vertical video and a desktop display asset need different compositions even when the underlying idea is the same.
Avoid declaring a winner too early
Set a review window based on your buying cycle and available volume. A concept with a few clicks has weak evidence about conversion. Do not keep changing the hypothesis after seeing a favourable metric.
When comparing results, distinguish a delivery difference from a persuasion difference. Uneven spend, different placements or a changed offer can make one creative look stronger without answering the original question. Use an appropriate platform experiment when you need a controlled comparison.
A brief you can reuse
- Customer problem and buying moment.
- One specific promise the product can support.
- Proof available to demonstrate that promise.
- Opening hook and intended placement.
- Offer and destination page.
- Metric, review window and decision rule.
AI can help draft concepts and adapt assets, but a test still needs a human-defined question and credible proof. See ten distinct angles for one product when the concept pool is narrow, and creative fatigue when the existing pool starts weakening.
Source
Shopify's creative testing overview describes creative experiments as a way to learn and allocate resources. The organiser example and matrix above are our own working framework.