Creative testing budget calculator

How many conversions should each variant get?

Each variant needs enough conversions to show the CPA gap you care about. At 40 conversions per variant, a test reliably separates two variants only when one’s cost per purchase is about 47 per cent lower (method).

The count is your call, because Meta’s figure belongs to an ad set. Ad sets exit the learning phase “as soon as they can deliver stably. This usually occurs after about 50 results in the week after the ad set’s last significant edit” (About the learning phase, read 26 September 2026).

That page gives no count for each ad. It asks advertisers to “Avoid high ad volumes”, since with many ads and ad sets “the delivery system learns less about each ad and ad set”. Ten AI variants in one ad set share what it learns.

Meta’s creative test identifies “top performing test ads” and says “A confidence level is not included”, so the count you set is what guards against a lucky leader.

Below 100 conversions per variant, only a CPA gap of a third or more shows up reliably. Each row compares one pair of variants on equal spend (method).

To show a 20 per cent gap, each variant needs about 315 conversions (method). At a cost per purchase of $30, that is $9,450 a variant and $47,250 for five, or 237 days at $200 a day.

Five variants make ten pairs. A Bonferroni split keeps the chance of any false winner across all ten at 5 per cent or less. It moves the 40-conversion row from 47 to about 56 per cent (method).

Your eCom Team’s calculator maps 80, 90 and 95 per cent confidence to 20, 30 and 50 conversions per concept. It names no source for the mapping (read 26 September 2026). On the table above, its 95 per cent setting (50 conversions each) reliably catches only CPA gaps of about 43 per cent or more.

Judging a test on conversion rate from clicks? AI UGC vs human UGC sizes that sample with a two-proportion formula. It needs a baseline rate, which this calculator does not ask for.

How the detectable gap is worked out

ratio = exp(2.80 × sqrt(2 / n))

What the calculator leaves out

The calculator leaves out what the variants cost to make. At the example numbers each added variant costs $1,200 of spend and 6 more days to read. The cost per AI ad calculator prices the making, plan by plan, to set beside it.

It also leaves out:

• How Meta splits the money. Its creative test will “aim to spend that amount per day on test ads” but does not say how the amount divides between them. The spend formula assumes equal shares, as Meta’s A/B test page recommends: “the same budget for both versions”.
• The rest of your budget. Meta suggests “no more than 20%” of the existing budget for test ads. Following it, a $200 daily test needs a campaign or ad set budget of at least $1,000 a day.
• The seven-variant cap. The creative test compares “up to 7 creative variants”, so the calculator’s row for 10 variants means a second test or another setup.
• Late conversions. Meta’s default attribution counts “7-day click, 1-day view and 1-day engagement” (Results), so a purchase can be credited to an ad up to 7 days after the click. Counts for the last week keep rising after the test stops.
• Seasonality. A sale or holiday inside the run moves every variant’s cost per purchase, and with it the spend and days above.

What to do when the test runs too long

Set the days against how fast ads tire

Meta’s fatigue page gives no fixed window. Its “Creative fatigue” status shows when cost per result is “twice as much as ads you ran in the past” or more.

The page calls the feature “only available for ad sets with one creative except those with Advantage+ catalog ads (previously dynamic ads), dynamic creative, or Meta Advantage+ app campaigns”. Yet it says the status shows “for your ad set or ad”. Whether each ad in a test gets its own flag is unclear there.

A test that runs 60 days may be judging tired ads by the end, and Meta may not flag it.

Cut variants before you cut conversions

Halving the count from 40 to 20 halves the days. It also widens the smallest gap you can see from 47 to 59 per cent (table). Halving the variants from 10 to 5 halves the days too, from 60 to 30, and keeps the 47.

Test fewer ideas that differ more, since small counts only see big gaps, and keep the rest for the next round.

Meta’s creative test and A/B test sit beside paid testing tools in AI creative testing tools, mapped by job. The steps for planning the test itself are in a creative testing framework for Meta ads.