Skip to content

How many ad creatives you actually need to test each week

RIVYL~8 min read
Illustrative shape. Winners land at roughly one in twenty, so the number of tests you can run each week sets the ceiling on how many winners you can find.

About 5% of ad creatives turn into winners, so the planning number is twenty creatives per winner. Divide twenty by the number of winners you want a quarter and you have your weekly output. The harder finding underneath that arithmetic is that large accounts don't find winners at a better rate. They just put far more spend behind the ones they find.

What share of creatives actually win?

Roughly one in twenty. Motion's 2026 creative benchmarks, built from $1.29B of Meta spend across 578,750 creatives and 6,015 accounts between September 2025 and January 2026, classify about 5% of creatives as winners. A winner there means a creative that took at least ten times its account's median creative spend and at least $500, and Motion reports that threshold as sitting at the 92.3rd percentile.

~5%

Share of creatives that reach winner status

Motion, Creative Benchmarks 2026, April 2026

92.3rd

Percentile at which the winner threshold sits

Motion, Creative Benchmarks 2026, April 2026

18.8

Net-new creatives shipped per week by enterprise advertisers

Motion, Creative Benchmarks 2026, April 2026

Read the definition carefully before you use the number, because it's doing something specific. A winner is defined by how much spend the platform pushed through it, not by what it earned. Those correlate, since Meta concentrates budget on what its model expects to convert, and they aren't the same thing. More on why that matters further down.

How do you work backwards from a hit rate to a weekly number?

Two divisions. The first converts a hit rate into creatives per winner, the second converts creatives per winner into a schedule.

creatives per winner = 1 ÷ winner rate

At the measured 5%, that is 20 creatives for every winner.

weeks to first winner = 20 ÷ new creatives shipped per week

An expected value, not a promise. Winners arrive at random, not on a rota.

Expected value is the phrase doing the work in that note. At a 5% rate each creative is an independent coin flip, so shipping twenty doesn't guarantee a winner. It gives you about a 64% chance of at least one. To be 80% confident of finding one you need 32 creatives, because at twenty the odds of a complete miss are still better than one in three.

New creatives per weekExpected weeks to first winnerWeeks to be 80% likelyWinners per quarter
120320.7
210161.3
4582.6
82.545.2
121.72.77.8
18.81.11.712.2
Expected weeks to a first winner at a 5% winner rate, and winners produced per quarter. Derived from Motion's measured winner rate, April 2026.

The table is the whole planning exercise. An account shipping one creative a week is running a twenty-week discovery cycle, which is longer than most seasonal windows and longer than most patience. An account shipping four is finding roughly two and a half winners a quarter, which is enough to keep a rotation alive as older creative fatigues.

Four a week is the number worth defending for a small account. It falls out of the arithmetic above rather than from any published benchmark: at a 5% winner rate, four a week is roughly one winner a quarter, which is about the slowest discovery rate an account can run and still replace creative faster than it fatigues. Both halves of that sentence are estimates — nobody publishes a fatigue-cycle length — so treat four as a floor to argue with, not a finding.

Is volume what separates large accounts from small ones?

Less than you'd think. The same Motion dataset shows 55% of all Meta spend landing on winning creatives overall, and then splits that by advertiser size: 23% for micro advertisers against 64% for enterprise. Both groups are finding winners. Only one group is backing them.

Share of spend that reaches a winning creative

The gap isn't in how many winners they find. It's in how much spend reaches the winners they have.

A bar chart showing micro advertisers put 23 percent of their spend behind winning creatives, the all-advertiser average is 55 percent, and enterprise advertisers reach 64 percent.

Motion, Creative Benchmarks 2026, from $1.29B of Meta spend, April 2026.

A micro advertiser with 23% of spend on winners is leaving 77% of budget on creative the platform did not choose to scale. Some of that is unavoidable, since you can't know which creative will win before you fund it. Most of it is structural. Small accounts spread budget across too many ad sets, cap creatives before they accumulate enough conversions to be read, and keep underperformers alive because turning them off feels like waste.

Discovery gets all the attention, and concentration is where the money is. Finding a winner you then give 4% of budget to is not finding a winner.

There's a hard floor underneath this that most testing advice skips. Meta needs roughly 50 optimisation events per ad set per rolling seven days to leave the learning phase, so an ad set funded below that can't produce a readable result for any creative inside it. Splitting a small budget across five ad sets to test more creatives at once produces five unreadable tests rather than one clean one. The launch arithmetic works through what that floor costs in money.

Spend concentrates because delivery concentrates. The tail is not waste, it is the cost of finding the head.

Should you test new concepts or iterations of a winner?

Both, at a ratio that changes with what you currently have. A concept is a different argument: a new claim, a new format, a new kind of proof. An iteration keeps the argument and changes one variable: the first three seconds, the thumbnail, the caption, the aspect ratio, the call to action.

With no winner in the account, run concepts almost exclusively. Iterating a creative that never worked produces variations on a failure, and the reason it failed is usually the argument rather than the execution. The angle taxonomy is a way to be certain that your first ten concepts are ten different arguments rather than ten treatments of one.

With a winner in the account, iterate hard on it while continuing concept tests underneath. A winning argument usually has more than one execution in it, and iterations are cheaper and faster than concepts because the script, the offer and the proof already exist. The reason to keep running concepts anyway is that every winner eventually fatigues, and a pipeline you start after the fatigue begins arrives too late.

What does this volume cost to produce?

Less than the price lists suggest, if you stop assuming every creative is a shoot. The four-a-week account that films four videos a week has an expensive problem. The four-a-week account that films twice a month and cuts the rest from existing footage, customer reviews, product photography and static formats doesn't.

Creator video pricing does circulate in fairly consistent ranges, commonly quoted at roughly $150 to $300 for a beginner, $400 to $800 mid-tier and $500 to $1,200 or more for an established creator, with usage rights for paid distribution adding perhaps 30% to 50%. Treat all of those as unverified. They come from agency rate cards and marketplace listings, nobody publishes a method behind them, and they vary enormously by category and market.

Whatever the real number is, it has to fit inside contribution margin alongside the media itself. Production is a fixed cost against a variable return, which is why it's worth knowing your break-even before you commission anything. The break-even ROAS piece covers what belongs in that calculation.

Why is a 5% hit rate not a quality problem?

Because of how the winner is defined. The threshold is ten times the account's median creative spend, and a median is by construction the middle of the account's own distribution. Setting the bar at ten times the middle guarantees that only a small fraction clears it, whatever the standard of the work. If every creative in the account improved tomorrow, the median would rise with them and the share above ten times the median would barely move.

That isn't a criticism of the benchmark. It's a benchmark of concentration, and concentration is what it says on the label. It just means a low hit rate isn't a report card. Chasing a higher one by making fewer and more considered ads is the single most expensive mistake available here, because it lowers the number of coin flips while leaving the odds per flip about the same.

The second thing the definition does not capture is profit. A creative that Meta funded heavily is a creative Meta's model expected to convert, which isn't identical to a creative that made money after shipping, returns and discounting. Use the winner rate to plan your output, and use your own contribution margin to decide what to keep running.

Four new concepts and two iterations is a realistic week for a small account. The iterations are not filler — they are how a winner is exhausted properly.

What does a sustainable testing week look like?

A rate you can hold for a quarter beats a burst you can't repeat. Eight creatives in week one followed by nothing for a month is a worse programme than three a week for thirteen weeks, even though the second one ships fewer in the first fortnight.

Checklist

0 / 9

One more thing worth watching while you build the pipeline. Meta reported ad impressions up 12% and average price per ad up 9% across full-year 2025, so the auction is both bigger and more expensive than it was. Rising prices raise the cost of every week an account spends running creative that was never going to win, which is the real argument for testing volume — not that more ads are better, but that finding the winner sooner is worth more each year.

For how to judge the creatives once they're live, hook rate and hold rate covers which video metrics diagnose what and which ones quietly predict nothing. For where the concepts themselves come from, reading a competitor ad library is the cheapest source of tested arguments available, because an ad still running after three months is one somebody keeps paying for.

Sources

  1. Motion, Creative Benchmarks 2026, from $1.29B of Meta spend across 578,750 creatives and 6,015 accounts. April 2026
  2. Meta Business Help Centre, About the learning phase and the roughly 50 optimisation events an ad set needs. Accessed August 2026
  3. Meta Investor Relations, Fourth quarter and full year 2025 results, ad impressions and price per ad. January 2026

RIVYL

Every ad your competitors are running, in one place

Search competitor creative across Meta, TikTok and TikTok Shop, sorted by how long each ad has run.