Creative testing budget on Meta: how many ads to test, what each needs, and when to call a winner

A sensible creative testing budget on Meta is about 15 to 25 percent of your monthly spend, enough for every new idea to reach at least three times your target cost per purchase or lead. Launch only as many new ads as that budget can read properly, expect Meta to give most of the money to two or three of them, and do not call a winner until it has around 10 results. Anything under that is a hint, not a verdict.

Most of the guides ranking for this search give you the same formula: pick a target cost per purchase, multiply it by two or three, multiply that by the number of ads, and that is your test budget. The formula is fine. What it quietly assumes is that Meta will spend the money evenly across your new ads. It does not, and on the accounts we run that one assumption is the difference between a test that teaches you something and one that just burns a week.

How much of your budget should go to testing

The common split is 70 to 80 percent of spend on proven ads and 10 to 30 percent on new ones. We sit in the middle of that, around 15 to 25 percent, and move it with the state of the account:

Account state	Share of spend on new ads	Why
Strong winners, stable cost per purchase	10 to 15%	Protect what works, keep a steady flow behind it
Winners starting to tire	20 to 25%	You need a replacement before the old ad falls off
No clear winner yet, or a new brand	30% or more	Everything is a test until something proves itself

The share matters less than the maths underneath it. Take your test budget for the week, divide it by three times your target cost per purchase, and that is roughly how many separate ideas you can afford to read. If you spend 2,000 a week on testing and your target is 40 per purchase, that is about 16 ideas, not 60.

How many ads to test at once

This is where the even split assumption breaks. When you put several new ads into one ad set, Meta does not hand each one an equal share. It picks early favourites within a day or two and pours money into them.

We looked at single ad tests across the four accounts we ran the most tests on over the last 45 days, 69 tests in total. On every one of those accounts, the top three ads took between 45 and 80 percent of all the test spend. The single biggest ad took between a quarter and a half on its own. At the other end, 40 of the 69 tests got less than 2 percent of their account's test spend, and 41 of them finished with no attributed result at all.

That is not a failure of those ads. Most of them were never really shown to enough people to fail. So two practical rules follow:

• Count the ideas Meta will actually fund, not the ads you upload. In a single ad set of eight new ads, expect a real read on two or three. If you need a read on all eight, split them across more ad sets or run them in smaller groups week by week.
• Treat an unspent ad as untested, not as a loser. Relaunch the ideas you still believe in next to a different set of ads, rather than deleting them because they "didn't work".

Volume still matters. Meta's system now leans on the ad itself to find the buyer, so the brands that grow are the ones feeding it a steady stream of new ideas. We test 20 to 50 ads a week on our larger accounts. But volume only helps if the budget behind it can read what you launched. We cover the wider point in our piece on what a good Meta ads agency does each week.

How much each ad needs before you judge it

The usual rule of thumb, as in this guide to spend per creative, is to spend two to three times your target cost per purchase on an ad before you decide, and others suggest five times for a cleaner signal. That is a sensible floor. We find it more useful to count results than spend, because results are what you are trying to predict. This is the scale we hold ourselves to:

Results behind the ad or idea	What you can say
1 ad, any number of results	A story about one ad, not a pattern
Under 10 results	Thin. Worth testing again, not worth scaling on
10 to 30 results	Directional. Lean into it, keep testing alongside
30 or more results	Readable. Make the next batch around it

Holding to that scale is humbling. Of the 350 lessons we logged across our accounts in the last 90 days, 77 percent were stories about a single ad, and only 2 percent rested on 30 or more results. Most of what any account "learns" in a week is anecdote. The skill is in not treating it as more than that.

A real example from a UK luxury home brand we run. One single ad returned more than four times the account's target on the back of one purchase. A different idea aimed at sceptical first time buyers, tested across three ads with close to 40 purchases between them, came in at about two thirds of the target. The first looks like the winner on a dashboard. The second is the only one of the two we actually know something about: that idea, as made, misses.

When to call a winner

A winner is rare, and it is relative to your own numbers. We call an ad a winner when it beats your target cost per purchase or return on spend with enough results behind it to count as directional or better, and keeps doing it for at least a week. Three checks before you move budget:

1. Has it got at least 10 results? If not, give it more time or a dedicated budget. Do not scale on one or two sales.
2. Is it beating your target, not the account average? An ad that beats a weak account can still lose you money.
3. Did it hold for seven days? Plenty of ads look brilliant for two days while Meta finds the easiest buyers, then settle.

The same applies to formats and angles, not just single ads. In our first August Weekly we wrote that statics were beating video on a premium product. On our own scale, that pattern is still thin: the statics behind it had only a couple of purchases, while the videos had dozens and sat just under target. We are still making statics for that account because the early read is promising, but we have not moved the budget on it, and we would not tell you to either.

When to call a loser

Losing is easier to read than winning, because a clear miss shows up faster than a clear win. An ad has earned its exit when it has spent three times your target cost per purchase with nothing to show, or when it is directional or better and still well short of target. Cut it, write down what it was trying to say, and do not repeat the same idea in a new coat of paint next week.

What you should not cut on a loss is the idea behind an ad that never got spend. As the numbers above show, that is most of them.

A simple weekly testing routine

Day	What happens
Monday	Read last week: which ideas cleared target, with how many results behind them
Monday	Decide this week's batch from that read, sized to what the test budget can fund
Tuesday to Wednesday	Make and launch the new ads
Following Monday	Judge only what has 10 or more results, carry the rest

The brands that do this well are boring about it. Same day, same rules, every week, and a written record of what each test was trying to prove. It is what lets you scale fast when something does work, because you already know which ideas have earned it. On one of the accounts on our case studies page, Liquidation Store, we "scaled Meta spend nearly 4x in a month while ROAS rose from 4.2x to 5.7x".

If you want to see what this looks like on your own brand, we will make you 10 free ads built from your account and your site.

FAQ

How much should I spend to test one Meta ad? At least two to three times your target cost per purchase or lead, and ideally enough to reach around 10 results before you make a scaling decision. Below that you are reading noise.

How many ads should I put in one ad set? As many as you like, but expect Meta to fund only two or three properly. If you need a read on more, spread them across ad sets or across weeks.

Should I use a separate testing campaign? It helps when you need every new ad to get a fair share of spend, and it keeps tests from disturbing the ads that already work. Once an ad proves itself, move it into your main campaign.

How long should a creative test run? Seven days is a good default. It covers the weekday and weekend swing and gives most ads time to settle after Meta's early burst of easy buyers.