Creative testing is the process of running ad variations against each other under fair conditions so you learn which concept, hook or format lowers your cost per result.
A test gives a real answer when you change one thing at a time, give each ad enough spend to reach a meaningful number of conversions, and judge it on CPA or ROAS instead of clicks alone.
The rest of this guide shows how to set that up on Meta and TikTok, using real results from our database.
Every number below comes from a result in the ROASArchive, credited to the person or company that posted it, or from a source linked inline.
What a creative test is (and is not)
A creative test answers one question: does ad B get you customers more cheaply than ad A, with everything else held equal?
The "everything else" part matters most. That means the same audience, the same optimization event, the same budget logic and the same landing page.
A creative test is not launching ten new ads into your main campaign and seeing which one Meta spends on.
That shows you what the algorithm prefers in the first few days, and that is useful.
But it is a delivery signal, not a controlled comparison.
The ad that wins spend early is often the one with the cheapest clicks, not the cheapest purchases.
It is also not a landing page test, an offer test or a bidding test, even though those can move CPA more than any ad.
If you change the price and the ad in the same week, you will not know which one did the work.
Our archive shows how big non-creative changes can be: @SamPiliero cut cost per purchase by 45% just by lowering the product price by €10, with the same ads and budget. Run that kind of test separately.
A good creative testing framework has three layers:
- Concept or angle: the core reason someone should buy (health benefit, price, status, convenience).
- Execution: format and style (UGC, founder talking to camera, static, carousel).
- Iteration: hooks, first frames, captions and edits on a concept that already works.
Test from the top down. A new hook on a weak angle rarely saves it.
A new angle can change the whole account.
Test structures: ABO test campaign, CBO, Advantage+ creative testing and the A/B test tool
There is no single correct Facebook ad testing framework.
There are four common setups, and each trades control for speed in a different way.
1. ABO testing campaign
You create a separate campaign with ad set budget optimization (ABO).
Each ad set holds one concept, often with 3-5 variations of that concept, and gets a fixed daily budget.
Because the budget is set per ad set, every concept gets spend whether or not Meta "likes" it early.
This is the classic media buyer setup. It is fair to each concept, easy to read, and costs more, because you are forcing spend into ads that may be losers.
Winners then graduate to your main scaling campaign.
2. CBO testing campaign
With campaign budget optimization (CBO), Meta moves budget between ad sets on its own.
You get faster answers at lower cost, but weak-looking concepts can get almost no spend, so you never learn whether they could have worked.
How you split ads inside CBO matters. @AshvinMelwani took a CBO with 50 ads and split them into 5 ad sets, and new-customer ROAS rose 55.65% to 1.79. Packing too many ads into one ad set tends to concentrate spend on a few of them.
Our page on CBO vs ABO results collects more of these comparisons.
Scale changes the picture too. In another result in our archive, @SamPiliero ran "broad packs" of creative across 64 ad sets and got 1,238 purchases on $42.5k, roughly a $34 CPA.
That is a testing structure built for volume, not for a small account.
3. Meta's built-in creative testing tool
Meta now has a native creative testing option inside the ad editor.
According to Jon Loomer's walkthrough, you can test two to five ads inside an existing ad set.
Meta gives each test ad a near-equal share of a test budget and recommends spending no more than 20% of the budget on the test.
The limits matter. It requires the Highest Volume bid strategy, so you cannot use cost per result goals, bid caps or ROAS goals.
Tests run for seven days by default. And it creates duplicates rather than testing your existing ad, so the original ad is not part of the test.
This option suits accounts that run Advantage+ or consolidated campaigns and do not want a separate testing campaign.
It is the easiest way to give new ads a fair amount of spend without restarting learning on your live ads.
4. The A/B test tool (Experiments)
Meta's A/B test in Experiments splits the audience randomly so each person only sees one variation.
Per Jon Loomer's guide, you can compare two to five campaigns or ad sets that should be identical except for one variable.
Meta reports a winner on the key metric you pick and shows how confident it is that the same result would repeat.
This is the most rigorous option for Facebook ad split testing, and the slowest and most expensive.
Use it for big decisions, like UGC vs studio video as your main format, not for weekly hook tests.
TikTok has its own version. TikTok's split testing help page says the tool splits audiences so each group sees only one ad group, supports creative as a test variable, and is designed for a 90% confidence rate before it declares a winner.
For TikTok creative testing on a budget, many advertisers still use a simple ad group with several ads and judge on CPA.
Which structure to pick
| Structure | Fairness | Speed | Cost | Best for |
|---|---|---|---|---|
| ABO testing campaign | High | Medium | High | Testing new concepts and angles |
| CBO testing campaign | Low to medium | Fast | Low | High-volume iteration |
| Meta creative testing tool | High | 7 days by default | Capped by test budget | Consolidated or Advantage+ accounts |
| A/B test (Experiments) | Highest | Slow | Highest | Big format or strategy decisions |
To see how working media buyers set these up in practice, our list of media buyers to follow includes several who post their testing structures and the results.
How much to spend per creative
The honest answer is "enough to get conversions, not clicks."
A practical rule many buyers use is to give each ad or concept a budget of at least one to three times your target CPA before you judge it.
If your target CPA is $40, that is $40-$120 per ad before you call it.
The reason is simple math. With one or two purchases, the result is mostly luck.
An ad that gets 0 sales on $40 and an ad that gets 1 sale are not proven different.
Below roughly 10 conversions per variation, treat any "winner" as a hunch to confirm, not a fact.
Meta's own delivery system has a similar threshold.
An ad set generally exits the learning phase after about 50 optimization events in a 7-day window.
If your test ad sets will never get near that, consider optimizing for a higher-funnel event (add to cart) for early reads, then confirming on purchases.
A worked example: you have $1,500 a month for testing and a $50 target CPA.
At two times CPA per concept, that is $100 per concept, or about 15 concepts a month.
That is a realistic pace for a small brand.
If you want to test 40 ads a month, you either need more budget or a cheaper proxy metric.
Volume does matter once the budget supports it.
@SamPiliero cut cost per order from €84.38 to €58.63 (−30.5%) by adding better ads on a flat budget.
A study in our archive from NP Digital found that campaigns with 20+ ads had a 6.8% lower CPA than those with 1-2 ads.
Our page of creative volume results has the full set.
Note what these results do not say. More ads helped when the ads were different and good.
Ten small edits of one weak ad are not volume. They are the same test ten times.
What to test first (ranked by what moved results in the database)
We tag every result in the ROASArchive with the lever that changed.
At the time of writing, among creative levers, visual treatment appears most often (125 results), then creative volume (117), format switches (92), angles (63), ad copy (38) and hooks (29).
See what advertisers test most for the full breakdown, and our analysis of what moves CPA most for the size of each effect.
Here is the order we suggest, based on how big the documented wins were.
1. Angles: the reason to buy
Angle tests produced some of the largest CPA drops in the archive.
inBeat's work for Hurom replaced discount messaging with health-benefit creator videos and reported a 65% lower CPA and 300% higher ROAS. inBeat's case study says it also tested hooks, calls to action and creator personas along the way.
On a tired account, @SamPiliero launched about 20 new angles and cost per purchase fell from $101.29 to $60.50 (−40.27%).
Also from @SamPiliero, a Gruns breakdown shows ROAS ranging from 2x to 5x depending on which customer avatar the ad spoke to.
The same product, sold to different people for different reasons, performs very differently.
2. Format and execution
Once you have angles that work, test how they are delivered.
MHI Media (Kamal Razzak) reported a 62% lower CPA from a simple founder video shot in a car, which beat polished UGC and studio ads. @alexgoughcooper's archive entries include a B-roll mashup of old footage that ran at a $58.61 CPA on $107k spend.
Platform-native formats also show up. In vendor-reported stories, TikTok says TANP got a 39% lower cost per install with creator Spark Ads, and Instagram says Prose cut CPA 23% with vertical Reels-first video.
Treat those as directional, since the platforms are reporting on their own features.
3. Hooks and first frames
Hooks are the cheapest thing to iterate and a good use of a winning concept.
A Janky Patel newsletter case study in our archive tested problem-first hooks on broad targeting for an eight-figure DTC brand and reported 10.5% lower CAC and 20.5% higher ROAS over 75 days.
4. Visuals and small edits
Backgrounds, colors and text overlays come last for most accounts, though they can matter for catalog ads.
Marpipe reports (vendor-reported) that Ridge saw 53% higher ROAS with colored backgrounds on catalog ads, and that AgelessRx's top ad showed a 77% lower CPA after one week in a test of copy and value propositions.
Reading results: CPA, hook rate and significance
Pick your decision metric before the test starts, and make it the one that pays the bills.
For ecommerce that is CPA or ROAS on purchases.
For lead gen it is cost per qualified lead, not cost per form fill.
Leading metrics help you diagnose why an ad won or lost.
The most useful is hook rate: 3-second video views divided by impressions.
It tells you whether the first seconds stop the scroll.
@alexgoughcooper's makeup teacher video ran at a $47.13 CPA with a 45.9% hook rate, which shows a strong opening and a strong purchase result in the same ad.
A simple way to read a test:
- High hook rate, poor CPA: the opening works, but the middle, offer or landing page loses people. Keep the hook and rework the body.
- Low hook rate, good CPA: the few people who watch are buying. Test new hooks on the same ad.
- Low hook rate, poor CPA: move on to the next concept.
- Good CPA on low spend: confirm with more budget before you scale it.
On significance: Meta's A/B test shows the percentage chance you would get the same result again, and TikTok's split test aims for 90% confidence.
In regular campaigns you do not get that number, so use volume as your guard.
If ad A has 12 purchases at $40 and ad B has 3 at $30, ad A is the safer bet, because ad B's lower CPA rests on three sales.
Watch for two traps. First, early winners fade.
Check whether the CPA holds after the first week, when frequency rises.
Second, attribution windows can flatter some ads. If you rely on in-platform numbers, compare winners against blended results like MER over time.
Tools for creative testing and analysis
You can run a creative testing strategy with Ads Manager and a spreadsheet.
Tools help in two places: finding ideas worth testing, and reading results across hundreds of ads.
We have not used these tools ourselves; the notes below come from their public pages. Prices are as of November 2026.
- Foreplay is an ad swipe library with creative briefs and competitor tracking. Plans start at $59/month ($49/month billed annually), with a 7-day free trial. It is best for teams that need an idea and briefing workflow. It is not a replacement for your analytics.
- Atria combines an ad library, creative analytics and AI briefs. The Core plan is listed at $129/month on annual billing during a promotion (normally $219) and covers up to $500K in analyzed ad spend. It suits teams that want inspiration and reporting in one tool. Solo advertisers may find it more than they need.
- Motion is a creative analytics tool for Meta, TikTok and YouTube. Its Starter plan is $999/month for brands spending up to $50k a month on ads. That price makes it a poor fit for small accounts.
- AdCreative.ai generates ad creatives and offers a creative scoring feature. Its site lists plans from $39/month with a 7-day trial. It claims its scoring predicts which ads will perform with over 90% accuracy (vendor-reported). A prediction is not a result, so still run the ads before you trust the score.
Foreplay, Atria and AdCreative.ai have affiliate programs with us, so we may earn a commission if you sign up through our links. That does not change the downsides above.
For deeper comparisons, see our roundups of creative analytics tools for reading test results and ad spy tools for finding angles competitors are already running.
When to hire help
A testing program needs two skills that rarely sit in one person: making many distinct ads every month, and structuring and reading tests without fooling yourself.
If you are short on either, hiring can be cheaper than a year of noisy tests.
Signs it is time:
- You ship fewer than 5 new concepts a month and CPA is rising.
- Every new ad looks like the last one, so tests keep finding nothing.
- You spend enough that a 20-30% CPA drop would pay for an agency or freelancer.
- Nobody owns the testing calendar, so tests start and stop at random.
Look for people who can show results tied to specific changes, like the angle and format tests above, not just ROAS screenshots.
Our directory lists creative agencies and media buyers, and we can match you with one that fits your budget and platform.
Whoever runs it, the rules stay the same: change one thing at a time, spend enough to see conversions, judge on CPA, and test new angles before you polish hooks.