Ask around and you'll hear plenty of confident percentages for how much of your budget should go to testing. The problem with any fixed share is that it ignores the two things that matter most: what a conversion costs you, and how quickly your current creative is wearing out.
A better approach works backward. Figure out what a single test needs in order to produce a readable result, then decide how many of those tests you can afford right now.
Separate testing from scaling
Testing and scaling budgets do different jobs, and they shouldn't compete for the same dollars.
- Scaling budget goes to proven ads and pages. Its job is efficient revenue, and you judge it on CPA, ROAS and MER.
- Testing budget goes to new angles, new creative and new landing pages. Its job is learning, and you judge it on whether each test reaches a clear answer.
When new ads go into the same campaign as your proven winners, the platform will often keep spending on what already works. New creative never gets enough delivery to be judged, and you conclude it failed when it never really ran. A separate testing campaign or ad set gives each new idea a fair share of spend.
Give each test enough to mean something
A test that spends $40 and gets zero purchases tells you almost nothing. It might be a weak ad. It might just be bad luck.
One common rule of thumb is to let a new ad spend a set multiple of your target CPA before you judge it on purchases. The right multiple depends on how noisy your data is and how much risk you're comfortable with. Pick one, write it down and apply it to every test, so you're not making the rules up as you go.
Leading indicators can help you cut obvious losers sooner. A very low click-through rate, poor hold rate on video, or plenty of clicks with no add-to-carts are all reasons to stop early. Final calls on purchase performance still need purchase data.
Size it from your CPA: a worked example
Here's one way to put numbers on it. These are made-up round numbers for illustration, not benchmarks.
Say your target CPA is $50, and you decide each new ad needs at least 3x CPA in spend, or $150, before you'll judge it on purchases.
- You want to test 4 new angles a week, with 2 ads per angle. That's 8 ads.
- 8 ads × $150 = $1,200 a week in testing spend.
If your total ad budget is $6,000 a week, that plan uses 20% of it for testing. If your budget is $3,000 a week, the same plan would take 40%, which is a lot of money going to learning instead of revenue. At that budget you might test 2 angles a week instead: 4 ads × $150 = $600, or 20% of $3,000.
Notice the order. The test plan set the budget, and then you checked whether the share was sustainable. That's the right direction to reason in.
Also be honest about what this buys you. An ad that performs exactly at a $50 CPA would be expected to produce about three purchases on $150 of spend. That's a small sample. Treat this stage as screening, a way to find likely winners, and then confirm them with more spend before you trust them.
Landing page tests work differently. If you split existing traffic between your current page and a new variation, you aren't adding much spend. You're dividing traffic you already pay for. The constraint is time: how long until each side has enough purchases to compare. The guide on how long to run an A/B test covers how to think about that.
Graduate winners deliberately
A test winner isn't a scaling winner until it holds up with more spend behind it. Set a simple promotion path and use it every time:
- Screening. The ad hits your CPA target, or comes close, at test budget.
- Confirmation. It gets more budget inside the testing campaign and keeps performing with more purchases behind it.
- Graduation. It moves into the scaling campaign. On Meta, you can usually build the new ad from the existing post, so comments and reactions carry over.
Write down why each winner won: the angle, the hook, the format, the page it pointed to. That note is often worth as much as the ad itself, because it tells you what to test next.
When to test less
- Your budget is small. If you can only afford one or two meaningful tests a month, make them count. Test big differences in angle or offer rather than small tweaks, because small changes need far more data to detect.
- Performance is strong and stable. If your winners are holding steady and frequency isn't climbing, it's reasonable to lean toward scaling for a stretch.
- Your tracking is unreliable. If you can't trust purchase data, fix measurement first. Testing without it is spending money on guesses.
When to test more
- Creative fatigue is setting in. Rising frequency, falling click-through rate and a creeping CPA on your main ads mean you need replacements ready before the current ones fade. The post on spotting creative fatigue covers the warning signs.
- You're launching a new product or market. You don't yet know which angle works, so learning is worth more than short-term efficiency.
- A seasonal peak is coming. Find your winners before the expensive weeks, not during them.
- You plan to scale spend. Bigger budgets reach more people, and you'll need more angles to keep finding new buyers at a workable cost.
Keep the plan visible
Whatever split you land on, write it down as a weekly plan: how many angles, how many ads per angle, the spend threshold for each, and the rule for graduating a winner. Review it monthly against your CPA and your fatigue signals, and adjust.
The point isn't to hit a particular percentage. It's to make sure every test dollar can actually produce an answer.
The short version
- Don't start from a percentage. Start from what a readable test costs.
- Keep testing and scaling budgets separate so new creative actually gets spend.
- Size tests from your target CPA, then check whether the resulting share of budget is sustainable.
- Treat early wins as screening and confirm them with more spend before scaling.
- Test less on tight budgets or broken tracking. Test more when fatigue sets in or you're launching something new.