# Creative testing framework for small brands on Meta and TikTok

> A creative testing framework for small brands on Meta and TikTok: one variable at a time, test structure, budgets, naming, calling winners and iterating.

Published 2026-10-09 · eroq.ai — canonical: https://eroq.ai/blog/creative-testing-framework-small-brands


Big advertisers test by volume: a steady stream of new ads, and the budget to let the numbers sort them out. A small brand cannot buy its way to an answer, so every test has to be designed to answer one question. That is what a creative testing framework is for: fewer tests, each one readable, each one feeding the next.

This framework works on Meta and TikTok with a small budget, in five steps, followed by a month of [creative testing](/use-cases/creative-testing) as a worked example. A word on roles first: the eroq [Ads Studio](/studio/ads) makes the creatives (static ads, video ads and variants). It does not run, publish or read your ads. Launching tests and reading results happen in Meta Ads Manager and TikTok Ads Manager.

## Step 1: test one variable at a time

An [ad creative](/glossary/ad-creative) bundles several decisions. Test them in this order, biggest first:

1. **Angle:** the reason to buy the ad leads with. See [ten selling angles to test](/blog/selling-angles-to-test).
2. **Hook:** the first second, in picture, voice and text. See [ad hooks that stop the scroll](/blog/ad-hooks-that-stop-the-scroll).
3. **Format:** a static, a UGC testimonial, an unboxing, a product hero shot.
4. **Visual:** the scene, setting, colors and framing.
5. **Presenter:** who is on screen.

Why biggest first: a bigger difference between two ads tends to produce a bigger gap in results, and a small budget can only detect big gaps. A test of two background shades needs a lot of traffic before the gap means anything; two different angles can separate with far less. Spend your small budget on the questions it can answer.

Keep everything else fixed during a test: the audience, the placements, the budget, the landing page, the dates. If two things change, you will not know which one did the work.

## Step 2: structure the test

There are three common structures, each with a trade-off:

- **Separate ads in one ad set (Meta) or ad group (TikTok).** Each creative is its own ad, so results map one to one. The catch: delivery tends to concentrate on the early favorites, and the ads that barely spent are unproven rather than beaten. Simple, and where most small brands start.
- **The platforms' split-testing tools.** Meta's A/B test and TikTok's split test divide the audience so each version gets a comparable share. Cleaner comparisons, but more budget per version. Keep them for big questions, such as one angle against another.
- **Dynamic or automated creative.** You upload several images, headlines and texts, and the platform assembles the combinations. Useful for scaling pieces you already trust; poor for learning, because reporting by asset is thinner than reporting by ad. Use it once you know your winners, not to find them.

Budgets, as rules of thumb rather than laws:

- **Let each ad spend two to three times your target cost per purchase** before you judge it. An ad with no purchase by then is unlikely to be your winner.
- **Run each test across a full week**, so weekday and weekend behavior both count.
- **Do not edit mid-test.** Big changes (budget jumps, new targeting, adding ads) can send a Meta ad set back into its learning phase and blur the comparison.
- **Few conversions?** Use earlier signals (cost per click, adds to cart, early video views) to prune obvious losers, but call winners on the result you pay for whenever you can.

The arithmetic decides the size of a round: if your test budget only buys a handful of purchases per ad, test three ads, not eight.

## Step 3: name everything so results map back

Results you cannot trace back to a creative are wasted. Pick one naming template and never break it. A template that encodes the variables:

`date_product_angle_format_hook_presenter_ratio`

For example `1009_AMBRE_Gift_Static_H1_none_3x4` and `1009_AMBRE_Gift_UGC_H2_Mara_9x16`. In the ads manager table, sorting by name then reads like the experiment: every Gift ad together, every H2 hook together.

In eroq, every render of a brand lands in that brand's own Library folder, and the brand's History lists its ads, so the source of each uploaded file is easy to find. Name the ad in your ads manager when you upload it (or rename the file first) with the template above.

Then keep a test log: the date, the question, the ads, the result, the decision. One line per test. The log is what turns a year of small tests into knowledge about your customers.

## Step 4: call the winner

Rules of thumb for calling it:

- **Enough spend** on each ad, as in Step 2.
- **Enough results** that one lucky purchase cannot decide it.
- **A clear gap on the metric you pay for**: cost per purchase or return on ad spend, not clicks alone. The ad with the best click-through rate and the worst cost per purchase is clickbait, not a winner.
- **Consistency across days**, not one spike.

If the gap is small, call it a tie: keep the version that is cheaper to produce, and test a bigger difference next round. A losing angle is a result too: write it in the log.

## Step 5: iterate on the winner

A winner is a starting point. The next round builds on it, and this is where eroq makes the creatives fast:

1. **More of the winning angle, as statics.** Run a new static batch for the winning product: Clap plans it for free across the selling angles, each ad with its own scene, headline and call to action, in Square (1:1), Portrait (3:4) or Story (9:16). Keep the ads that land on your winning angle. On Eroq One, which holds the product exactly to its photo, statics cost 20 credits / image. More in [static ads in batches](/blog/static-ads-in-batches).
2. **A video ad on that angle.** Create, Video ad: the format that fits the angle (Unboxing for Gift, Luxury macro for Ingredient), a presenter from your own [characters](/studio/characters/new), and a spoken line. Clap drafts four lines on different angles: pick the closest to your winner, or edit it to fit. Up to four takes per click.
3. **The Multiplier on the winning video.** One ad becomes ×3, ×6, ×9 or ×12 independent [variations](/glossary/ad-variations) with a new character, outfit, location or product, while the motion, the cut, the timing and the original sound stay. Switch on only the change you want to test. On Replace, 8 credits / second of clip: six variants of a 10-second ad cost 480 credits. The Multiplier invents new faces and does not write new words: to test your own characters, or a new hook, make a new video ad. See [one winning ad into 12 variants](/blog/ad-multiplier-one-winning-ad-into-12-variants).
4. **Test the new round** against the old winner, with the same framework.

The variations are what keep a winner alive when [creative fatigue](/blog/creative-fatigue-how-many-ad-variations) sets in; the new concepts are what find the next one. eroq does not publish to Meta or TikTok yet: download the files, or fetch them through the [API](/docs), and upload them in your ads manager.

## A month of testing, as an example

Here is the framework applied to AMBRE, an amber perfume, by a brand with a small budget. The number of ads per round is an example: size yours to your budget, as in Step 2.

| Week | The question | What you upload | Made in eroq with |
|---|---|---|---|
| 1 | Which angle sells? | Three statics in the same format: Gift, Ingredient or material, Lifestyle moment | A static batch of ten, keeping one ad for each angle you test |
| 2 | Which hook, on the winning angle? | Three video ads, same format and presenter, three different opening lines | Create, Video ad, once per line |
| 3 | Which presenter? | The winning video and three variants with new faces | The Multiplier at ×3, only Character switched on |
| 4 | Is the winner tiring? | Fresh variants of the winner, plus one new concept on the next angle | The Multiplier again, and a new static batch |

At the end of the month you know which angle, which hook and which kind of face work for AMBRE, and you have the next round ready. A batch of ten statics on Eroq One costs 200 credits; Clap's planning and lines are free. [Start the first round](/studio/ads) in the Ads Studio.

## FAQ

### What is creative testing?

Creative testing is running different versions of an ad against each other, under the same conditions, to learn which angle, hook, format, visual or presenter brings results at the lowest cost. For a small brand, the rule is one variable per test, so each result means something.

### How much budget do I need to test ad creatives?

Enough for each ad to produce comparable results. A rule of thumb is to let each ad spend two to three times your target cost per purchase before judging it, over a full week. If your budget cannot do that for many ads, test fewer.

### Should I use dynamic creative or separate ads for testing?

Separate ads, or the platform's split-testing tool, when you want to learn which creative works, because results map to one ad each. Dynamic or automated creative suits scaling pieces you already trust, since its reporting by asset is thinner.

### When should I call a winner?

When each ad has had enough spend, enough results that one purchase cannot swing it, a clear gap on cost per purchase or return on ad spend, and a lead that holds across several days. A small gap is a tie.

### Does eroq run or analyze my creative tests?

No. eroq makes the creatives: static ad batches, video ads and Multiplier variants. It does not publish to Meta or TikTok and reads no spend, clicks or results. You launch the test and read it in your ads manager.