How much to spend testing Facebook ads is a formula anchored to your target CPA: what a real test costs per creative, how to pace it, and when a winner is safe to scale.
Five new creatives are ready to launch, and someone who owns the money wants to know what the test will cost. So you search, and the internet hands you everything from $10 a day to $1,000 per variant, sometimes in the same article. None of those numbers mean anything until they are anchored to the one figure that is yours alone: what a conversion costs you.
I ran paid acquisition at up to €150k a month, and this question never disappeared. It just changed shape as the budgets grew. What follows is the math I ended up using, the cheap shortcut for when the math gives an ugly answer, and the two moments where test budgets quietly die: mid-flight and at scale-up.
The takeaways
The unit of an ad test is conversions, not dollars. A creative needs roughly 30 to 50 conversions before its CPA is worth believing, so the minimum test budget per creative is your target CPA multiplied by that count. With a €30 CPA, each creative needs €900 to €1,500. Ten conversions (€300 in this example) buys a directional read you should hold loosely.
This is also why the same threshold keeps appearing in every guide: Meta's documentation puts learning-phase exit at about 50 optimization events within a week. An ad set that never gets there delivers erratically and reports noisy numbers, so a thin test is worse than a small one. Five creatives sharing €50 a day will all starve. Two creatives sharing the same €50 might each reach a verdict. Fund fewer creatives properly instead of giving ten of them pocket money.
Then run a cheaper proxy test first and stay honest about what it measures. Hook rate and CTR stabilize at a fraction of conversion-level spend; in my experience €50 to €100 per creative is enough to rank a batch on attention. The catch: an ad that stops thumbs can still sell nothing. A proxy test only tells you which creatives deserve a conversion test. Picking the final winner stays a conversion-level job.
That makes the affordable version of testing a staged one. Stage one, give every creative €50 to €100 and cut the bottom half on hook rate and CTR. Stage two, give the survivors the full CPA-times-30 budget. You spend the expensive conversions only on creatives that already cleared the cheap bar. On a €1,000 total budget with five creatives and a €30 CPA, that is the only structure I know that produces a defensible winner.
Test budgets fail mid-flight in two quiet ways: they either eat the week's budget in a day or stall and deliver nothing. In Adscalr I built pacing checks that run every 5 minutes against the day's cap, with three thresholds: an overspend flag at 110% of cap, a runaway alert at 150%, and an underspend flag when an ad set has spent less than 70% by after midday. You can copy the thresholds by hand with a midday and an evening check.
The underspend case is the one buyers skip. A test that spends only 40% of plan stretches a 7-day test into a 14-day one, and creative fatigue does not pause while you wait. If delivery is starved, fix the bid or the audience the same day rather than letting the calendar absorb it.
Scale after the winner exits the learning phase and holds a stable CPA for about three more days, then raise the budget in 20% steps with 48 hours between raises. That cadence is my own practice, for one reason: large budget jumps push an ad set back into learning, and you pay for the relearning with a CPA spike right when you committed more money.
Two warnings from spending real budgets on this. First, distrust the day-three star: a 4x ROAS on a handful of conversions is usually luck, and I wrote a whole piece on telling a winner from a lucky ad. Second, protect the test from your own trigger finger. The most common testing mistake I have watched is killing an ad on day two. Adscalr's automation refuses to kill anything under 5 days or €200 of spend for exactly this reason, and every kill stays reversible for 24 hours.
A test budget is one line in a bigger allocation question: which campaigns deserve money at all, and on which platform. That is the part of Adscalr I use most myself. It builds 3 to 5 prioritized campaign plans from your real 12-week performance, each with a conversion goal and an AI-estimated CPI, and runs the pacing checks above so a test week ends with a verdict instead of a shrug. If working backward from CPA matches how you want to plan spend, the budget-intelligence page shows how those plans get built.
This is the thinking behind Adscalr.
See the product →