A simple rule for test budgets by price point, and when to stop a test early.
Every new ad we launch gets the same test, so we can compare them fairly. The question clients ask most is how big that test should be. Here's the rule we use.
The rule: enough for 10 purchases
Give each new ad roughly ten times your current cost per purchase, spread over four days. If your account buys a customer for $40, that's about $400 per ad. If it's $100, it's $1,000. Ten purchases is the point where cost per purchase stops swinging wildly from one order to the next.
Your cost per purchase | Test budget per ad | Over |
|---|---|---|
$20–40 | $300–500 | 4 days |
$40–80 | $500–1,000 | 4 days |
$80+ | $1,000, then judge on add-to-carts | 4–7 days |
For expensive products, ten purchases can take too long. There we judge the test on add-to-carts or checkouts started, then confirm winners with purchases over the following week.
Stop early when the attention numbers say so
We don't wait for the full budget if an ad is clearly failing. After $150 or so, if the hook rate is under 20% and the click rate is under half the account's average, nobody's stopping to watch and nobody's clicking. More money won't change that. We switch it off and write down why.
We never stop a test early because it's winning. Early winners regress more often than they hold.
Keep tests separate from winners
Tests run in their own campaign with their own fixed budget, usually 10 to 15% of spend. If you put new ads into the same campaign as proven winners, the platform sends almost all the money to the winners and your new ad never gets a fair test.
The short version
Ten purchases' worth of budget, four days, its own campaign, and the same rules for every ad, including the ones you love.
Nadia Okafor
Co-founder and head of buying. Ran $2M a month on Meta before Winrate.
