The Learning Phase Budget Formula: How Much You Actually Need Per Ad Set

Picture two ad accounts running the same product, the same creative, the same offer. One has ten ad sets, each on A$40 a day, all of them stuck blinking "Learning" three weeks in. The other has two ad sets on A$200 a day, both out of the learning phase by day five and quietly compounding.

Same monthly spend. Wildly different result. The only thing that changed was how the budget was split.

That gap is what this post is about. There's a real formula for how much budget an ad set needs to learn properly, almost nobody runs it before they build their campaigns, and the brands that underfund the maths pay for it with months of mediocre delivery.

Here's the formula, then the honest part that most people skip: what to do when it asks for more money than you've got.

What the learning phase actually needs

Meta's optimisation needs data before it can deliver predictably. The number it's built around is 50 conversion events per ad set per week. Get there and the ad set exits the learning phase, delivery stabilises, and the algorithm starts making informed decisions instead of guessing. Fall short and you sit in "Learning Limited" indefinitely, which is a polite way of saying the system never got enough signal to do its job.

So the budget question isn't "what can I afford to spend". It's "what does this ad set need per day to clear 50 conversions in seven days". Those are different questions, and the second one has an actual answer.

The formula, worked end to end

Here it is in plain terms:

50 conversions x your CPA, divided by 7 = your starting daily budget per ad set.

There's a shorthand for the same sum, which is CPA x 7.14. Both give you the same number. Use whichever one sticks in your head.

Let's run it with real numbers. Say your cost per acquisition is sitting at A$30.

• 50 conversions x A$30 = A$1,500 a week needed to learn
• A$1,500 divided by 7 = roughly A$214 a day, per ad set

So that single ad set wants about A$214 a day to have a clean shot at exiting the learning phase inside a week. Not A$40. Not "whatever's left in the budget". A$214.

Now change the CPA and watch what happens. A higher-ticket product with a A$60 CPA needs A$60 x 50, which is A$3,000 a week, or about A$429 a day per ad set. A cheaper impulse buy at a A$15 CPA needs A$15 x 50, A$750 a week, around A$107 a day. The formula scales straight off your CPA, which is exactly why two brands can need completely different budgets to learn the same way.

Per ad set, not per campaign (this is where most people trip)

This is the bit that quietly wrecks accounts, so I'll say it plainly. That number is per ad set, not per campaign.

If your campaign holds two ad sets, you need to double it. Three ad sets, triple it. Going back to the A$30 CPA example, one ad set needs about A$214 a day. Build a campaign with four ad sets inside it and you've just committed to roughly A$856 a day if you want all four to learn properly.

Here's the thing - this is exactly where the trouble starts. A founder reads "A$214 a day" once, mentally files it as the campaign budget, then splits it across five ad sets because they want to test five audiences. Now every single one is getting about A$43 a day, none of them will ever sniff 50 conversions in a week, and the whole campaign sits in permanent learning. Five half-starved ad sets, zero of them with enough food to grow.

That's not a testing strategy. That's five ways to learn nothing at once.

What to do when the formula asks for more than you have

Now the part competitors skip, because it's the part that actually matters for most brands.

You run the formula, it tells you that you need A$214 a day per ad set, you want to test five audiences and a couple of creative angles, and you do the multiplication and realise the maths wants A$1,500 a day. You're working with A$400. So what now?

The wrong answer is to spread your A$400 across all ten boxes anyway and hope. That's the underfunded account from the top of this post. Here's what I'd do instead.

Consolidate ruthlessly. This is the lever almost nobody pulls hard enough. Instead of ten ad sets fighting over scraps, run two or three that each clear the threshold. Fewer audiences, broader targeting, more budget concentrated where it can actually exit the learning phase. Meta is far better at finding your buyer inside one big well-fed ad set than you are at hand-carving ten tiny ones. Give the algorithm room and budget in one place and let it do the sorting.

Use the full attribution window to your advantage. Your CPA on any single day will bounce above and below your target. That's normal. The system averages out over the attribution window, so don't panic-edit an ad set on day two because it had one expensive day. Every time you reset it with a change, you can kick it back into learning and waste the budget you'd already spent teaching it. Make fewer, bigger decisions, not lots of nervous little ones.

Let ROAS-goal bidding do the consolidating for you. If you sell a real spread of price points, this is worth a hard look. With a single product you can usually run a clean CPA target. But when one ad set is exposed to a A$45 product and a A$190 product at the same time, lowest-cost bidding will naturally chase the cheaper conversion every time, because that's the cheapest result to buy. Set a ROAS goal instead and you're telling Meta to optimise for the margin outcome, not just the cheapest sale. It's a tidy way to put several products into one consolidated, properly funded campaign without the algorithm defaulting to your lowest-value item.

Lower the CPA you're solving for, honestly. If the formula is impossible at your current CPA, that's information. Either your CPA needs to come down before you scale the structure, which is a creative and offer problem, or your ambitions for how many audiences you can test need to shrink to match the budget you actually have. Both are fine answers. Pretending you can fund ten ad sets on a two-ad-set budget is not.

The one-line version

If you take nothing else from this: count your ad sets, multiply 50 by your CPA, divide by seven, and that's the daily budget each one needs to learn. If the total is more than you can spend, you don't underfund them all. You cut the number of ad sets until the maths works.

A handful of well-fed ad sets will beat a swarm of starving ones every time. I've watched it play out across accounts of every size, and the pattern barely changes. The brands that win the budget game aren't the ones spending more. They're the ones who refused to spread what they had so thin that none of it could learn.

So before your next build, run the formula against your real numbers and count how many ad sets it can genuinely support. If that count comes back lower than what you've currently got live, you've probably found the reason your delivery feels stuck. If you'd like a fresh pair of eyes on whether your structure is funded to learn or quietly starving itself, that's exactly the kind of thing a Signal/Noise Audit is built to surface - it maps your spend against what each ad set actually needs before you commit another dollar to a campaign that was never going to clear the threshold.