The short version
- Launch as many new ads as your testing budget can give a verdict to: each ad needs roughly 2–3× your target cost per purchase in spend.
- Put 3–6 genuinely different ads in each test. Small accounts can judge 2–5 new ads a week, large ones 20 or more.
- Test different ideas first, then make variations of the ones that sell.
- Meta's current delivery rewards variety, so count distinct concepts, not near-identical copies.
- Stop an ad once it has a verdict: cut it at about 2× target cost per purchase with no sale, and scale it in steps once it sells at target.
The short answer
Test as many genuinely different ads as your budget can give a verdict to — and no more. For most brands that means 3–6 ads in each test, and somewhere between 2–5 new ads a week on a small budget and 20 or more on a large one.
That number isn't a matter of taste. It falls out of one sum:
Launch fewer and you're guessing: two ads can't tell you much about what your customers respond to. Launch more and most ads get only a little spend each, nothing reaches a verdict, and you end up picking "winners" out of noise. The rest of this guide shows where the sum comes from and how to use it.
Working out your number
1. Start from your target cost per purchase
Your target cost per purchase (CPA) is the most you can pay for a sale and still be happy with the result. If you haven't set one, start from your margin on a first order. It doesn't need to be perfect to be useful here.
2. Work out what a verdict costs
An ad has had a fair chance when one of two things has happened:
- It's a clear loser. It has spent about 2× your target CPA without a purchase. It would need to be far better than it has looked so far to come good.
- It's a likely winner. It has three or more purchases at or under target — which, by definition, takes up to 3× target CPA of spend.
So budget 2–3× target CPA per ad. Treat it as a first filter, not proof: three sales can be luck. That's fine, because a winner's next step is a bigger budget, where it has to keep performing to keep the money.
3. Divide
Many media buyers set aside somewhere between a tenth and a third of their spend for new creative — more in a young account with no proven ads, less once a few winners carry the account. The example below uses a quarter.
One catch: Meta doesn't share an ad set's budget evenly between its ads. It pushes spend towards the ads its models expect to do best, so in any week some ads get far more than 2–3× CPA and others almost nothing. Plan on the average, and treat an ad that barely spent as untested rather than beaten. If you still believe in the idea, bring it back in a later test next to different ads.
The creative test calculator runs this sum for you from your own target CPA and budget.
Test concepts first, variations second
Not every new ad is a new idea. It helps to separate two kinds:
- A creative concept is a different idea: a new angle, format, person or setting. "Problem–solution UGC-style photo", "price per cup against café coffee" and "founder story video" are three concepts.
- A variation is the same idea with one thing changed: a new hook line, headline, opening frame, colour, crop or button.
Test concepts first. The gap between two ideas is usually much bigger than the gap between two versions of the same idea, so concept tests find the big lever quickly. Once a concept sells, shift towards variations of it — they're how you get more life and more reach out of an idea that has already proved itself.
| Where you are | What most new ads should be | Example for a coffee brand |
|---|---|---|
| No winner yet | 3–6 distinct concepts per round | Morning-routine UGC photo, price per cup against a café, ingredient flat lay, founder story |
| First winner found | Variations of the winner, plus one or two new concepts | The price-per-cup ad with three new hooks, a 9:16 video version, a new persona |
| Winner wearing out | New concepts again, plus fresh takes on the winner | A new angle such as gifting or office life, the winner rebuilt in a new setting |
Watch the combinations. Four hooks × three visuals × two captions × two sizes is 48 ads — at $13 a verdict (2.5× a $5.15 target), that's $619 before you've learned anything. The ad variation calculator shows how fast combinations multiply and what testing them all would cost. You rarely need every combination: pick the handful that answer a real question.
How many new ads to launch each week
Here's the same maths turned into a starting plan. It assumes about a quarter of spend goes to testing and each ad needs about 2.5× target CPA for a verdict. That gives a handy shortcut: new ads a week ≈ monthly budget ÷ (40 × target CPA).
| Budget | Monthly spend at a $5.15 target CPA | New ads a week | How to run them |
|---|---|---|---|
| Small | Up to $1,031 | 2–5 | One test at a time: one ad set with 3–5 distinct concepts, refreshed weekly |
| Medium | $1,031–$4,124 | 5–20 | One or two tests a week, mostly new concepts plus variations of winners |
| Large | Over $4,124 | 20+ | Several tests in parallel. The limit becomes how fast you can make good ads |
If your target CPA is different, scale the budget column: what matters is how many target-CPA purchases your budget buys in a month. On very small budgets, run one test of three ads at a time and let it finish before you start the next.
Whatever your budget, keep at least three ads in every test — with fewer there's nothing to compare — and few enough that the average ad reaches 2–3× CPA of spend within about a week. Tests that drag on get muddied by fatigue, seasonality and changing offers.
Signs you've got the number wrong
- Too many: most ads end the week having spent less than your target CPA, verdicts never arrive, and you catch yourself choosing winners on clicks.
- Too few: the same two or three ads take all the spend for weeks, frequency keeps climbing, and when a winner fades there's nothing ready to replace it.
What changes with Meta's current delivery
Meta's delivery system has changed in a way that affects how you count. In December 2024, Meta's engineering team described Andromeda, a next-generation retrieval system that picks which ads are even considered for each person, using much larger models than before.
The practical upshot — echoed in Meta's own creative guidance and widely discussed by advertisers — is that with broad targeting your creative does more of the targeting, and variety matters. Genuinely different concepts give the system more to match to different people. Near-duplicates, like the same ad in five colours, seem to add little. Practitioners report that Meta treats very similar ads as largely the same.
Meta hasn't published a recommended number of creatives, so treat this as a direction rather than a rule. For the maths above, it means:
- Count concepts, not files. Ten colour swaps are closer to one ad than to ten.
- Spend the testing budget on distinct ideas first. Variations earn their place once a concept has sold.
- Prefer fewer, broader ad sets with more genuinely different ads over many small ad sets with one or two ads each.
Our guide to Meta Andromeda and creative diversity goes deeper.
When to stop testing an ad and scale it
Every ad needs a verdict, not an opinion. These rules work for most accounts:
| What you see | Verdict | What to do |
|---|---|---|
| About 2× target CPA spent, no purchase | Loser | Stop it and let the budget flow to the others |
| 3+ purchases at or under target after about 48 hours | Winner | Scale it in steps and make variations of it |
| Some sales but too few to judge, or CPA just above target | Unclear | Give it another 24–48 hours, then decide |
| A former winner's cost per purchase creeping up as frequency rises | Wearing out | Launch variations and new concepts to replace it |
Scale in steps — say 25–50% at a time — rather than doubling a budget overnight. Meta says significant edits, which can include big budget changes, can send an ad set back into the learning phase.
If you'd rather not watch all this by hand, Autopilot can run the tests and apply these rules for you. On the tests it launches itself, it stops an ad that spends 2× your target CPA with no sale (you can set the line anywhere from 1.5× to 3×), calls a winner after 48 hours with three or more sales at or under target, raises its budget one step without going past your daily limit — asking you first, by default — and drafts four fresh variants of every winner.
Testing never really stops, though. Even the best ad wears out as more of your audience sees it. Our creative refresh schedule guide covers how often to replace winners.
A simple weekly testing rhythm
- Monday: launch. Put this week's new ads — the number from your maths — into one test ad set, paused until you've checked them. BulkAds' bulk ad launcher builds the campaign, the ad set and one ad per creative in one go.
- Midweek: cut losers. Stop anything that has spent about 2× target CPA without a sale.
- End of the week: call winners. Ads with three or more sales at or under target move up, unclear ones get another day or two.
- Next Monday: iterate. Make variations of this week's winners, add new concepts and go again.
Keep a simple test log — what you tested, what won, what you learned — so each week builds on the last instead of quietly re-testing old ideas.
Questions people ask
How many ad creatives should I test per ad set?
Three to six genuinely different ads per ad set suits most budgets: enough to compare, few enough that each gets meaningful spend. Go higher only if the ad set's budget can give every ad roughly 2–3× your target cost per purchase within about a week.
How much should I spend testing each creative?
A useful rule of thumb is 2–3× your target cost per purchase per ad. That's enough to stop a clear loser (no sale after about 2× target) or spot a likely winner (around three sales at or under target).
How many new ads should I launch each week?
Divide your weekly testing budget by 2–3× your target cost per purchase. For many small D2C accounts that's 2–5 new ads a week, larger accounts can judge 20 or more.
Should I test new concepts or variations of a winner?
Start with genuinely different concepts to find what works. Once one sells, spend more of each round on variations of it — new hooks, openings and formats — while keeping one or two new concepts in every round.
Does Meta Andromeda mean I need more creatives?
It means you need more different creatives. Meta's retrieval system matches ads to people, and near-duplicates give it little new to work with, so a handful of distinct concepts is worth more than dozens of small tweaks.
How long should a creative test run?
Until each ad has a verdict, which usually takes a few days to a week. Clear losers can be stopped as soon as they spend about 2× target with no sale. Winners are best confirmed after around 48 hours of results.
Keep going
- Features
Autopilot
Stops losers, scales winners, inside your caps
- Features
Bulk ad launcher
A whole batch live on Meta in one go
- Free tools
Creative test calculator
How many new ads you can test, and how long until every one has a verdict
- Free tools
Ad variation calculator
How many ads your hooks, visuals and copy make, and what testing costs
- Guides
Meta Andromeda and creative diversity: what changed and what to do
What Meta's Andromeda retrieval system is, what it changes for your ads, a practical diversity checklist and how to tell whether your ads are different enough.
- Guides
How often to refresh Facebook ads: a schedule that fits your budget
Ads wear out at different speeds. A starting refresh schedule by monthly budget, what makes ads tire sooner, three kinds of refresh and a short weekly routine.
Written by the BulkAds team. Published , last checked . Meta changes its tools often — if something here no longer matches Ads Manager, tell us at contact@bulkadsai.com.
BulkAds AI is an independent product and is not affiliated with, endorsed or sponsored by Meta Platforms, Inc. Facebook and Instagram are trademarks of Meta Platforms, Inc.