How much to spend testing Facebook ad creative

A testing budget is three numbers multiplied together: how many variations you test, your target cost per purchase or lead, and how many times that cost you let each variation spend. Four variations, a $40 target CPA and a multiple of two gives a $320 test. Five variations at the same CPA and multiple gives $400.

Use your target CPA, not your current one. If you do not have a target yet, work it back from your margin. The most you can pay for a sale is the profit on that sale before ad costs. The guide at /guides/good-roas-for-meta-ads shows the same idea in ROAS terms.

Media buyers use a handful of rules of thumb for creative tests. None of them comes from Meta, and each is a starting point you adjust once you have your own data.

• Let each variation spend one to three times your target CPA before you judge it.

• Stop a variation that has spent about twice your target CPA with no purchase.

• Keep 10 to 20 percent of total spend for testing, more while you are looking for a first winner and less once you have several.

• Test three to five variations at a time.

• Run a test for around a week so every weekday is covered.

Budget sets how many results you can buy. At $50 a day and a $30 target CPA, the whole ad set buys fewer than two purchases a day. Split that across five variations and each one might get a purchase every few days, which is too slow to learn from. Test two or three variations instead, and make the differences large, such as different concepts or different hooks rather than different button colours, because big differences show up in fewer results.

Meta's guidance has long been that an ad set needs around 50 optimisation events in a week to leave its learning phase, so check the current figure in Meta's help pages. Small accounts rarely reach it, which is one more reason to keep tests simple and judge them on the purchases you do get.

Put four ads in one ad set and Meta will not give each a quarter of the budget. It shifts spend towards the ads it predicts will do best, often within the first day, and some ads get very little. Treat an ad Meta barely spent on as a weak signal, not a verdict. If you want an even split, give each variation its own ad set and budget, or use the A/B test feature in Ads Manager, which divides the audience between versions. The A/B test calculator at /tools/ab-test-calculator tells you whether a difference in results is likely to be real.

Test concepts and hooks before images and button text, because they change results the most and each answer shapes the next batch. The guide at /guides/ad-creative-testing sets out the order.

Budget is only half of testing. The other half is having distinct ads ready. Mockingbird's Make me ads option produces 1 to 10 concepts at once, each on its own angle, which gives you variations for a concept test. For a hook test, type a note such as 'Keep the layout, change the headline to a question' and get a new version. Each revision uses one credit.

A test is done when each variation has enough purchases or leads to compare and the ranking has stopped changing from day to day, or when every loser has hit your kill line. Early leaders often fade, so do not call a winner on day one. Once the winner is running at scale, watch it for the signs in /guides/ad-creative-fatigue and have the next test ready before it slows.