How much to spend on creative testing.

A creative testing budget is the part of monthly ad spend reserved for creatives that have not yet earned a verdict. Size it from the bottom up, not as a flat percentage: price one verdict (your conversion floor times your CPA), count the winners you need each month, divide by your hit rate to get the number of tests, and multiply. In the worked example below, an account spending $100,000 a month at a $60 CPA needs $15,000, and the share of spend falls out of its own numbers rather than someone else's rule of thumb.

The reason to do the math is that creative is now the scaling lever. Targeting, bidding and placement have largely been absorbed by platform automation (Advantage+ on Meta, Performance Max on Google), so the input that still decides whether an account grows is the supply of creatives that can absorb spend efficiently. That makes the testing budget a supply plan. This guide is the math, three worked examples, how to split the budget, and the four creative economics numbers that move it.

How much should you spend on creative testing?

Enough to find the winners your account will need next month, and not a dollar of verdicts more. The whole budget is five lines of arithmetic:

verdict price            = conversion floor x CPA
winners needed per month = winners lost to fatigue + winners for new spend
tests needed per month   = winners needed / hit rate
testing budget           = tests needed x verdict price
share of spend           = testing budget / monthly spend

Every input is yours. The conversion floor is the number of conversions you need before a creative's result is readable; the hit rate is the share of launches that become winners; winners lost to fatigue comes from how long your winners last. Borrowed rules ("give every ad $100", "keep testing to a fixed slice of spend") fail because they buy different amounts of information depending on your CPA. Per-creative test spend is an assumption you set, never a benchmark.

How do you price one verdict?

Multiply your conversion floor by your CPA. At a $40 CPA, a 10-conversion floor prices a verdict at $400; at $120, the same confidence costs $1,200. The creative testing framework explains how to set the floor and why attention signals can end a test sooner.

Two platform facts set the ad set around it. Meta says an ad set usually leaves the learning phase after about 50 results in the week after its last significant edit, and that performance is less stable and CPA usually higher until then. TikTok says volatility typically starts to ease after about 25 results or 7 days. So a testing ad set that should deliver stably on Meta needs roughly 50 results a week of budget, and on TikTok no verdict should come before the first week ends. If your testing budget cannot buy that, run fewer testing ad sets rather than thinner ones, which is also Meta's own advice on ad volume.

In Peachblue, the composite score tiers every creative from CTR, ROAS, CPA and spend percentiles, so the verdict rule is applied the same way every week, and the Low-data preset in the Creative Library keeps creatives that have not reached a readable volume apart from the losers.

How many winners do you need each month?

As many as fatigue takes away, plus the ones new spend requires. An account carrying nine winners that each last about three months loses three a month to fatigue; if it plans to grow, it needs one or two more on top. When supply runs below that line, nothing announces it. Budget concentrates onto aging winners, frequency climbs, efficiency drifts down a few percent a month, and someone says the account hit a ceiling. The ceiling is usually a production shortfall.

In Peachblue, the Today and Creative Economics dashboards show bench depth and a fatigue board that flags CTR decay against each creative's own best week, so "winners lost to fatigue" is a count you read rather than a guess. The creative fatigue guide covers the diagnosis.

Creative testing budget: three worked examples

The same five lines at three account sizes. Every input is an assumption for illustration; plug in your own.

The shares are outputs, not targets. Account A shows the small-account squeeze: 50 results a week at a $40 CPA is about $2,000 a week, roughly $8,700 a month and more than twice its testing line, so it should test in one ad set and treat each verdict as precious. Account C's constraint is the opposite: 40 verdicts a month are easy to fund and hard to learn from unless every one is logged.

Kill speed changes the total. Account B's budget assumes every loser runs to the full floor. If a third of its 21 losers are cut early on a weak hook rate at a quarter of the verdict price, the month costs $11,850 instead of $15,000, the same verdicts for $3,150 less.

Run the arithmetic on your own account with the creative testing budget calculator: it takes monthly spend, launches, hit rate and winner ROAS, and returns the testing budget, what losing creatives cost, your cost per winning creative and what a higher hit rate is worth.

How should you split a creative testing budget?

Split it three ways, and write each split down so it survives a busy month.

1. By stage. A cheap attention screen first (hook rate and early CTR after a few thousand impressions), then the full conversion verdict only for creatives that pass. This is where kill speed comes from.
2. Iterations versus new concepts. Iterations of proven winners carry a better hit rate than cold concepts, so most slots go to iteration, with a protected minority, for example a third, for new concepts so the well does not run dry. Under Meta's Andromeda delivery, new concepts must be materially different, or they share the original's audience and fatigue with it.
3. By platform. Fund tests where the spend needs winners, and port proven concepts between platforms as iterations: a Meta winner tested on TikTok usually starts with better odds than a cold idea.

In Peachblue, Intelligence shows the patterns and archetypes your winners share across 31 tagged dimensions, so the iteration slate starts from evidence, and the Next Creative Brief turns those patterns into concepts, proven hooks, do and don't rules and a prompt block for the generator you use.

What are the four numbers that set the budget?

Hit rate, kill speed, winner longevity and bench depth. Together they are creative economics: whether winner supply keeps up with spend.

Creative hit rate is the master number. A team launching 50 creatives a month at a 10 percent hit rate and a team launching 20 at 25 percent find the same five winners, at very different testing cost. The method is in the creative hit rate guide.

Kill speed fails in two directions. Too slow burns budget after the data already answered, and at volume it is usually the largest recoverable line in the account. Too fast manufactures false negatives and quietly lowers your measured hit rate.

Winner longevity sets the replacement cadence. Winners that last 90 days need a fraction of the testing of winners that burn out in three weeks at the same spend, which is why it has to be measured per account.

Bench depth is the difference between fatigue as maintenance and fatigue as an account-level event. One hero with an empty bench forces panic production, the mode with the lowest hit rate.

Why does a low hit rate have the best return?

Because winners are scarcest when each one matters most. Going from 10 to 20 percent doubles winner supply from the same testing budget; going from 40 to 50 adds a quarter. Teams with weak hit rates hold the largest upside, not the smallest. Winner quality compounds it: a better winner ROAS pays on the whole scaled budget, so better creative moves both numbers at once.

Is money spent on losing creatives wasted?

Not by itself. Losers are the price of finding winners, and a team that never funds them is not testing. The waste is the spend above the minimum: losers running past the point their verdict was clear, and tests on creatives that ignore what your winners already prove. Those zombies appear on no standard report, because no single day of them looks like a decision.

In Peachblue, the Creative Economics dashboard shows that wasted spend as dollars flowing to confirmed losers, with every flag carrying its consequence, while awareness and DSP delivery spend sits in its own bucket and is never counted as waste.

Why not run the testing budget in a spreadsheet?

You can, and should once. It breaks where the numbers start to matter: knowing which ads are the same creative. One video running in six ads looks like six tests in an export and in Ads Manager, so hit rate and kill speed are counted per ad, and every number above drifts.

Peachblue groups ads into creatives by perceptual fingerprint across ads, campaigns and platforms, so hit rate is computed rather than hand-counted, and runs the rest on top: composite scoring, the economics dashboards, Intelligence and the brief. Agent Peach, and Claude through the MCP server, can rank last month's launches and compare the winners with the losers from your own data. Plans carry monthly analysis credits (10 per video, 5 per image or text creative), so size the plan to your launch volume; they run from Starter at $79 a month to Agency, priced to your client roster, on the pricing page. Intelligence and the brief start on Pro.

A monthly testing budget review

Once a month, put five numbers on one page: creatives launched, hit rate, average spend per killed creative, the age of your top three spend-carrying winners, and the bench count. Then re-run the five lines of arithmetic with them. Write the verdict criteria down so hit rate means the same thing in March and in September, and baseline against your own trailing 90 days, since published benchmarks for creative metrics contradict each other.

One scope note: this is paid social math. Amazon DSP and streaming TV run on delivery economics (flights, budgets, pacing, CPM goals), a different discipline covered in the Amazon DSP pacing guide.

Run the review for a quarter and the conversation changes from "why was last month soft" to "what do we build next." That is creative economics doing its job.