How many ad creatives to test: the short answer
How many ad creatives to test comes down to one sum. Take your monthly ad budget and divide it by three times your average order value. That number is the fewest ads you should make. Then divide the same budget by three times what you are willing to pay for one sale. That is your ceiling.
On R450 a day, selling a R497 product, wanting sales at R150 each, that comes out at nine ads minimum and thirty maximum. We use the same sum on client accounts at V8 Media, where we have now put more than R200 million through ad platforms testing exactly this.
Most South African founders running a small budget make two or three ads and wonder why nothing moves. Two or three is not a test. You are guessing and paying for it.
This is a rule of thumb, not a law. It comes from watching accounts, not from anything Meta publishes, and the numbers move if your margins are unusual. It is still a great deal closer to the truth than picking a number because it felt about right.
Here is where those numbers come from, and where the maths stops working.
Worked example
How many ads a R450 a day budget pays for
Sunglasses brand, one product, R497 a pair, target cost per sale R150
Fewer than nine and one loser eats the month. More than thirty and none of them get enough money to tell you anything.
Switch an ad off once it has spent three times your order value
An ad needs enough money to prove itself, and it also needs a point past which you stop feeding it money that is clearly not coming back.
Three times your average order value is a practical stopping point. If an ad has spent triple what a customer is worth to you and brought back nothing, there is no momentum in it. Meta is struggling to find people who respond to that particular ad, and another R500 will not change that.
Notice this is not really a profit calculation. An ad can make sales that lose money on paper and still be worth keeping, because the algorithm has found traction and the cost usually falls as it gathers data. The thing you are actually watching for is whether anything is happening at all.
When nothing is happening, switch it off and load the next one. That is the job.
You cannot judge one ad on its own
This is the correction that matters most. Get it wrong and you will cut your best work yourself, then spend two weeks wondering why the account got worse the week after you tidied it up.
Meta gives credit for a sale to the last ad a person clicked, but it decides who to show your ads to based on everything that person has already seen, which means the ad that introduced someone to your brand two weeks ago gets nothing while the ad they clicked on the day they were finally ready takes the whole sale.
Run the three times rule on individual ads and sooner or later you will switch off the one that was doing the introducing, your numbers will get worse, and nothing in the account will tell you what you did.
Read the group, not the individual. Put five to twenty ads in one ad set, judge that set on its combined seven day return, and use each ad's spend only to see which ones Meta is choosing rather than as a reason to switch off the rest.
Where to look
Judge the group of ads, not one ad on its own
The same numbers, read at two different levels
Meta gives the sale to the last ad someone clicked, but it decides who to show you based on everything you saw. So the ad with the worst return is often the one doing the introducing.
One more setting worth changing while you are in there. Meta's default attribution window includes people who only saw your ad and never clicked it. Switch that off and count seven day click plus one day engaged, or you will keep crediting ads for sales they did not cause.
What actually happens
Meta will not split your money evenly across your ads
A typical month once the auction has picked its favourites
Two of these nine ads take eight rand in every ten. Build for that instead of fighting it, and never plan on nine ads each getting a fair ninth of the money.
Your ads will never get an equal share of the money
The formula makes it sound like nine ads each get a ninth of your budget. They will not.
Meta concentrates spend, and harder than most founders expect. Motion studied 578 750 ads across 6 015 advertiser accounts and $1.29 billion of Meta spend in 2026, and found that about six ads in every hundred take the majority of an account's money. We have not found a setting that changes it, and we see it in accounts of every size.
This changes what you are actually buying with nine ads. Most of them will barely spend. You are buying nine chances for the auction to find the one or two that work, and then most of your money follows those.
It is also why thirty is a ceiling rather than a target. Meta will never spread thirty ads evenly enough for most of them to clear the R450 each one needs to prove anything.
It also means you want a fixed budget per ad set while you are testing, because that forces every ad to get some exposure before the auction puts most of the money on one ad. Once you have a proven winner, hand the decision back to Meta and let it push. That is also the point where scaling an ecommerce account becomes a different job to testing one.
Nine ads is not nine separate ideas
Nine separate concepts on R450 a day is a month of filming and no signal at the end of it.
Break it into layers instead. A concept is three things joined together: who you are talking to, what you are saying to them, and what they get. Single men who care how they dress, told that these glasses make them look sharper on a first date, at twenty percent off. That is one concept.
Under the concept sits the body, which is the main piece of footage or the main image: a street interview, a before and after, someone talking straight to camera about how the date actually went.
Under the body sits the hook. The first three seconds. This is where nearly all your nine ads should come from, because it is by far the cheapest layer to change and it is the layer that decides whether anyone watches at all.
Two bodies, five hooks each. That is ten ads off two shoots, which clears your nine, and you should test three to five hooks on a body before you conclude the body itself is what failed.
Where the nine ads come from
Nine ads is not nine separate ideas
Three layers, and only the bottom one needs to be filmed nine times
Two bodies with five hooks each is ten ads for the price of two shoots. Nine separate concepts on R450 a day is a month of filming and no signal at the end of it.
Most of your ads should not mention your product at all
Here is a pattern in almost every small account that has stalled. Every single ad talks about the product, the price and the discount.
Meta reads your ad and goes looking for the audience that matches it. So an ad about your product is an ad for people who already know your product, and Meta serves it to the few thousand people who have heard of you. You run out of them. Costs climb. It looks like the platform capped you.
It did not cap you. Your ad did.
Aim for roughly eight in ten of your ads speaking to someone who does not know they have a problem yet, or knows the problem but not the solution. Split the rest between the people comparing their options and the people who already know your product and just need a reason to buy today.
| Who the ad is talking to | Share of your ads | What that ad actually says |
|---|---|---|
| Does not know they have a problem, or knows the problem but not the fix | About 8 in 10 | Looking sharp on a first date |
| Comparing the different ways to solve it | About 1 in 10 | Why rimless beats the chunky pair they own |
| Already knows your product and needs a reason to buy today | About 1 in 10 | Twenty percent off, free delivery on the second pair |
For the sunglasses brand, that means most of the ads are about looking sharp on a first date, and only one or two mention that the glasses are rimless and light.
Buy one get one free only works when the buyer already knows the product
A two for one deal at R1 000 sounds generous. What it actually does is ask a stranger to spend double your own average order value on their very first purchase from you.
Now stack what else is missing. A brand nobody has heard of, a product nobody has held, no reviews to lean on, and no way to pay it off over a few weeks. At that point the deal stops being the thing they are weighing up. Now they are weighing the risk.
Two for one works when the buyer already knows exactly what they are getting. We ran exactly that offer for Scentimental, an inspired-by fragrance brand, where the customer reads "inspired by Gucci" and already knows the smell, the feeling and roughly what the real thing costs. Two for the price of one lands as an obvious win. Take that same mechanic to a product nobody has tried and it lands as nothing, because two of something you do not trust is still something you do not trust.
The offer, not the ad
Two for one at R1 000 asks a stranger to double your usual order value
Asking a first time buyer to spend double your usual order value
This is not a price cut. The first pair keeps its price. What changes is that the risk of trying you drops, and the second pair becomes a decision they make after they have already said yes.
The instinct is to cut the price. Resist it. Discounting the thing you actually sell teaches people to wait for your next sale, and the cut comes straight out of your margin rather than out of anything you added.
Change the shape of the offer instead. Sell a single pair at R497. Offer the second pair at thirty or forty percent off once the first is in the cart. Then put free delivery behind that second pair, so there is a real reason to take it rather than a vague nudge.
The first pair keeps its price. The risk of trying you drops, and the second pair turns into a decision they make after they have already said yes, which is a far easier moment to sell into than the one before it.
Do not split a small budget across audiences
The instinct on a tight budget is to be careful. Run men separately from women. Split by age. Give each group its own campaign so you can see what works.
It feels careful. Each campaign gets too little. None of them learn.
What splitting does to a small budget
Splitting a small budget feels careful and leaves every campaign short
What founders believe, against what the account actually does
Under roughly R900 000 a month, run one testing campaign. Split by product margin or by a genuinely separate customer, never by gender or by the week you uploaded the file.
Meta needs about fifty sales in one ad set in a week before it leaves the learning phase and can find a pattern. Split an already small number across four campaigns and each one holds a quarter of a signal that was too thin to begin with. Meta never finds the pattern. The numbers look random.
Under roughly R800 000 a month, run one testing campaign. If you are unsure what that budget should be in the first place, we broke the Facebook ad budget maths down separately. Split only when there is a genuine commercial reason: products with very different margins, or a customer group whose economics are actually different. Never by gender, never by the week you uploaded the file.
The same goes for building a customer profile. A twenty four year old Afrikaans man who still lives with his mom and wants confidence for a first date is a description of one real person, not a group you can advertise to. There are ten of him, and reaching those ten costs a fortune. Go one level up: men, interested in fashion, age left open. Let the creative do the narrowing.
Send the click to a page that says what the ad said
You can get all of that right and still lose the sale in the last five seconds.
If your ad promises to help someone look sharper on a first date, and the click lands on a homepage showing eight products and a newsletter popup, the thought they arrived with is gone. They have to start again, and most people do not.
Build a page for each concept, carrying the same promise in the headline, the same people in the photos and the same proof, then answer the questions they are quietly holding: what if it does not suit my face, how long is delivery, can I pay it off, and what happens if I want to send it back.
This is not a small effect. The greens gummy brand Grüns went from launch to more than $300 million in annualised sales inside two years, roughly R4.8 billion, off one product written up more than ten different ways with a page for each. Inc. reported that run in 2026, the year Unilever bought the company. One product, ten stories, ten pages.
And keep the path short. Every extra step between the click and the add-to-cart button loses people. Do not send them to a landing page that then links to a product page. Send them to the page where they can buy.
Your first spend buys data, not sales
Say you have R6 000 and two weeks. There is no clever structure that turns that into profit on the first pass, and anyone selling you one is guessing.
You have no data yet, so the first money buys data. Put in R6 000 and you might get half of it back, and what you actually bought was knowing which ads never to make again.
Then you improve. The next R6 000 does better, and the one after that better again, because by then you are no longer guessing about which ads and which pages actually move people.
So treat that first amount as money you are prepared to lose, and if losing it would genuinely hurt you then do not spend it yet. That is how you budget.
But do not carry that mindset past the testing phase. The moment an ad reliably returns more than it costs, the rule inverts completely. You are no longer risking money, you are buying it at a discount, and the only sensible question becomes how much more you can put in before the returns fade.
Make sure ads are your actual problem before you make nine of them
One last thing. Do it before any of the rest.
A business grows on four levers: how many people see you, how many of them buy, what you charge, and how many come back. Only one of them is your real constraint at any moment.
Everything in this article is a traffic and conversion answer. If your real constraint is price, or the fact that nobody ever buys twice, then nine perfect ads will only help you lose money faster and more efficiently than before.
So before you go and make nine ads, ask honestly which of the four is holding you back. Then fix that one.
Frequently asked questions
How many ad creatives should I test each month?
Divide your monthly budget by three times your average order value for the minimum, and by three times your target cost per sale for the maximum. On R13 500 a month with a R497 product and a R150 target, that is nine to thirty.
When should I turn off a Facebook ad?
Once it has spent about three times your average order value with no sale, and only after you have checked the ad set as a whole. An ad with a poor return can still be the one introducing people to your brand.
What is the minimum budget to run Meta ads in South Africa?
There is no official minimum, but the maths above sets a practical one. If your budget will not pay for at least nine ads to reach three times your order value each, you do not have enough to run a real test yet.
Should I run separate ad sets for men and women?
Not on a small budget. You halve an already small number of sales and neither group collects enough for Meta to learn from. Split by product margin or a genuinely different customer, not by gender.
Does buy one get one free work for a new brand?
Usually not. It works when the buyer already knows what they are getting. On a brand nobody has tried, sell a single unit and offer the second at a discount once the first is in the cart.
Key takeaways
- Fewest ads to make is your monthly budget divided by three times your average order value. Most is that budget divided by three times your target cost per sale.
- Judge the ad set, not the single ad, or you will switch off the ad that introduces people to you.
- A handful of your ads will take most of the money. Motion put it at about six in every hundred. Plan for that instead of expecting an even split.
- Get your nine ads from a few bodies with several hooks each, not from nine separate ideas.
- Around eight in ten ads should speak to someone who does not know your product yet, or Meta will only ever show you to people who already do.
- A two for one deal that doubles your average order value asks a first time buyer to take a risk most of them will not take. Change the shape of the offer rather than cutting the price.
- Under roughly R800 000 a month, run one testing campaign. Splitting a small budget starves the signal.
- Your first spend buys data, not sales. Once an ad reliably returns more than it costs, stop protecting the budget and spend more.
You should not have to learn all of this on your own money
If you are running a small budget and you would rather not spend three months learning this on your own money, that is what we do. We have spent over R200 million testing ads, so we have already made most of these mistakes.
We take on about one in ten of the businesses that ask, because we would rather turn you away than take money we cannot grow. So the call is as much us working out whether we can help as it is you sizing us up.
Book a free call and we will look at your account and your offer before you spend another rand.
