There is no universal minimum budget, there is a calculation worked backwards: how many conversions each ad set needs before the platform can optimise, what you expect each one to cost, and how many ad sets run at once. The daily figure falls out of that. Below the floor the money still goes, and whatever you read at the end means nothing.
The arithmetic runs backwards
Almost everybody decides a budget from the wrong end: look at what is free this month, divide it by the days and start. That answers how much can I spend, which is a cash flow question. The campaign's question is a different one: how much does it have to spend before there is anything worth reading at the end.
A campaign optimising for conversions does not hand out impressions at random. It looks for patterns among the people who convert, and to find them it needs cases. Meta publishes 50 conversions per ad set in a seven day window as its own guidance for leaving the learning phase, and that threshold is the only one written down here, with the platform's name attached to it. On another network, or with a different conversion event, the number is different and you supply it.
Given a threshold and a cost estimate, the calculation is three lines long. The example below uses made up figures, purely so the order of operations is visible.
| Step | Where it comes from | In the example |
|---|---|---|
| Learning threshold | Platform guidance, or your own history | 50 conversions per ad set per week |
| Expected cost per conversion | Your estimate, not a measurement | 12 euros |
| Spend per ad set per week | Threshold times cost | 600 euros |
| Ad sets running at once | Your structure decision | 1 |
| Daily figure to type in | Weekly spend divided by seven | About 86 euros |
| The whole test | Weekly spend times ad sets times weeks | 2,400 euros over four weeks |
Change one cell and everything moves. With two ad sets instead of one, either the daily figure doubles or each of them gets half the threshold, and there is no third option. Double the cost per conversion and the floor doubles. That is the whole mechanism, and it is why two agencies quote different minimums for the same client without either of them lying: they assumed different costs per conversion and opened a different number of boxes.
Keep the last two figures apart, because they answer different questions and get confused every month. The daily one is what you type into the platform and what governs while the campaign runs. The whole test figure is what goes in the proposal, and it is not a recurring monthly cost: it is what it costs to arrive at an answer.
What actually happens below the floor
Nothing dramatic happens, and that is the problem. The campaign starts, spends, shows ads and hands back a report full of figures. What it does not do is learn, and that does not appear in any column.
With few conversions a week, the ad set stays in the learning phase. Delivery keeps exploring instead of sharpening, the cost per result swings hard from one day to the next, and the differences you see between one ad and another fit entirely inside the margin of error. Three conversions against one is not a winner against a loser, it is noise with names on it.
Then comes the second trap, which is behavioural. When a small campaign gives no signal, the natural reaction is to touch it: change the audience, move the bid, swap the creative. Editing those things sends the ad set back to the start of the learning phase, so every attempt to speed things up erases the little that had accumulated. A short, heavily edited campaign can spend a whole month without ever leaving the starting square.
The spend is real from day one and the reading is not real until there are enough cases behind it. A budget under the floor does not buy a smaller version of the test, it buys an invoice with a conclusion that will not hold.
The three ways out when you cannot reach it
There are three, and none of them is raising the bid. Raising the bid buys the same conversions at a higher price, which is the opposite of what you need when the problem is that there is no volume.
The first is cutting ad sets and concentrating the same money where it does reach the threshold. One ad set that learns is worth more than five that do not, even though the report shows less apparent work. How many boxes are worth opening, and when duplicating beats editing, is covered in the Meta campaign structure guide.
The second is measuring a conversion further up the funnel. A sign up or a form costs less than a purchase and therefore accumulates faster, and the ad set leaves learning on real volume. The price of that route is that the conversion is worth less: optimising towards sign ups gets you sign ups, and somebody has to check afterwards how many of them buy.
The third is running the test for longer, and it comes with a warning nobody gives. The threshold is counted per window, not in total, so more weeks do not take the ad set out of the learning phase. What they give you is accumulated volume to read something at the end, which is not the same thing, though it is not nothing when the budget cannot move.
Everything in this article happens in one place in GoFeed.
Try it freeThe weakest number in the calculation
The whole thing hangs off the expected cost per conversion, and that number is an assumption until the campaign gives you its own. It is where optimism gets in: somebody uses last year's best month, or the previous client who sold something else, and the floor comes out 40 % lower than it is going to be.
The discipline that fixes this is boring and it works. Write the estimate down before starting, let the first week run without touching anything, and redo the calculation with the real cost. If the real cost is double the estimate, the floor was double, and deciding that on day seven costs far less than discovering it on day thirty with the budget spent.
It also helps to separate what each acronym measures before arguing about them, because a rising cost per conversion does not always mean the campaign is doing worse: the space may simply have got more expensive for everyone. Which figure divides what is set out in the CPM, CPC, CPA and ROAS guide, and it is worth having straight before the first results meeting.
What to tell whoever is paying
The hard conversation is not asking for more budget, it is explaining what the budget buys. What a test buys is information, with an honest chance that the information turns out to be that this does not work here.
Three sentences carry that conversation. First: this is what the test costs, not what it will return. Second: below this figure the campaign spends the same and we will not be able to conclude anything, so a smaller budget is not a smaller test, it is a test that does not exist. Third: in four weeks we will know the real cost per conversion, and with that number we decide whether to continue, change what we measure or stop. When that last decision is due is covered in the guide on when to pause a campaign.
What you never do is promise a result in order to justify the budget. The floor is defended with the arithmetic of learning, which is verifiable, and not with a sales forecast nobody can guarantee.
Watching it while it runs
The full calculation, with the threshold, the cost, the ad sets and the weeks, is done for you in the campaign budget calculator: it returns the spend per ad set per week, the daily figure you type in, what the whole test costs and, if your budget falls short, how many ad sets it genuinely funds.
Then somebody has to watch the spend while it runs, which is where small campaigns get lost: one ad set eats the other, the platform does not spend evenly across days, and the month ends before the test does. The ads dashboard reads Meta, Google, LinkedIn, TikTok and Pinterest and leaves them in one table, with their spend and their results beside each other, and every account hangs off its own brand. Meta's hierarchy can also be built from there; campaigns on Google, LinkedIn, TikTok and Pinterest are created on their own platform, and a Google campaign is never created or renamed outside Google Ads. Any profile can have it from the second tier of its plan upwards, with the detail on the ads page.