On this page
- How to calculate creative win rate
- What counts as a win on each platform
- What counts as normal: how spend concentrates
- Creative win rate vs related metrics
- Is a low win rate a sign of weak creative?
- How to raise creative win rate
- Example: A quarterly win rate review for a fitness app
- Creative win rate with Starti AI Studio
- Frequently asked questions
Creative win rate is the share of new ad creatives that beat the current best ad in a test. It tells a creative team how often its testing produces a new winner, the ad that then receives scaled spend, and so how many variants each round needs.
The working formula:
Creative win rate = new creatives that beat the control ÷ new creatives judged × 100
A team that judges 50 new creatives in a quarter and finds 4 that beat the control has an 8% win rate.
How to calculate creative win rate
Each part of the formula needs a rule written down before the test, the way Unity's creative testing guide advises: "Define the deciding KPI up front. Pick the metric that settles the test before you see any data."
- Control: the ad a new creative has to beat. Unity's scaling guide says to "Compare new creatives with the best performing creative from the main campaign."
- Deciding metric: IPM for an early read on install campaigns, then the cost of the outcome that matters, such as cost per funded account for a finance app or cost per trial start for a subscription app, or ROAS.
- Win threshold: the confidence level at which the platform, or the team, declares a winner.
- Denominator: creatives that received enough delivery to be judged. In an October 26, 2026 analysis of creative volume by Singular, between 4% and 10% of new creatives on networks including Facebook and TikTok got zero impressions in their first week. Count those separately instead of as losses.
- Inconclusive tests: results below the platform's threshold count as non-wins and are reported next to the rate.
- Holding at scale: a win counts once the creative keeps beating the control at scaled spend for a set period, such as two weeks.
- Period: a fixed window, such as a month or a quarter, so the rate can be compared over time.
The rate then sets the size of the next round: variants needed = target winners ÷ win rate. A team that wants 3 new winners a month at a 12% win rate needs 25 judged variants.
What counts as a win on each platform
Platforms set different bars for declaring a winner:
- Meta: its help center states that "For A/B tests, a 65 percent or higher confidence percentage represents a winning result," where confidence is the likelihood that an outcome "would be consistent if repeated again under the same conditions," and suggests an estimated power of 80% or higher before a test starts.
- TikTok: split testing runs "with a 90% confidence rate," and the model picks a winner when the results "are statistically significant." TikTok's best practices recommend at least 7 days and a budget whose estimated power reaches 80% or more.
- Google Ads: experiments run two-tailed significance testing "using the 95% confidence interval." Video experiments report directional results at 70% or 80% confidence and conclusive results at 95%.
- Unity: the scaling guide states that "The creative pack with the highest conversion rate is your winner."
The same set of creatives yields more winners at a 65% bar than at 95%. Report the platform and the threshold next to every win rate, and compare win rates only across tests run under the same rule.
What counts as normal: how spend concentrates
Win rate has no industry standard: none of the platforms or measurement companies above defines one. Measurement companies report how concentrated spend is instead, and those figures show how few creatives carry a campaign:
- AppsFlyer: its State of Creative Optimization 2026, built on 1.2 million creative variations from about 1,400 apps, found that the top 2% of gaming video creatives took 53.5% of spend in Q1 2025 and 60.7% in Q2 2026, and that across verticals the share held by the top 10% of video creatives "climbs past 75% everywhere."
- Singular: its Creative Benchmark Report 2026 found that "the top 10% of creatives captured 75% to 95% of total spend," and in Google Ads campaigns with 100 or more assets, "about 13 assets carry the campaign."
AppsFlyer counts each creative variation separately, so a change of language, color, wording, or logo makes a new one. Counted by variation, a win rate has a larger denominator than one counted by concept, and the report puts the consequence directly: "If only 2-10% of variations ever end up carrying the vast majority of the spend, most tests are misses by design."
Normal, then, is a team's own baseline. Track the win rate per quarter, by platform and threshold, and judge each round against the last.
Creative win rate vs related metrics
Related metrics, and what each one tells a creative team:
- Creative win rate: winners ÷ creatives judged. How often testing produces a new best ad.
- Spend share of top creatives: the share of spend held by the top 2% or 10%. How concentrated delivery is.
- Zero-delivery rate: new creatives with no impressions in their first week ÷ new creatives launched. How many tests never start.
- Test confidence: the platform's confidence or power for one test. Whether a single result can be trusted.
- IPM and hook rate: the IPM and hook rate of one creative. Why it won or lost.
Is a low win rate a sign of weak creative?
Not on its own. Spend concentrates on a small share of creatives in both the AppsFlyer and Singular data, so a low share of new winners fits the pattern. A falling win rate is a signal when the team's testing has not changed: the same number of variants, the same budget per variant, and the same threshold.
The cost of a low rate is the budget spent on losers. Unity's guide sizes a variant on an install campaign from CPI: "At a $2 CPI and a 50-install threshold, that's roughly $100 per variant." On those numbers, a 4% win rate spends about $2,500 of test budget per winner, and a 12.5% rate about $800.
How to raise creative win rate
Changes that improve the odds of each test, and the reasoning behind them:
- Bigger differences: Unity advises teams to "Test fewer variables with bigger differences," since large effects need less volume to detect.
- Enough delivery per variant: two to four variants per A/B test, the range Unity's guide says will "suit most budgets," each budgeted for the platform's power target of 80% on Meta and TikTok, so a real winner is not missed.
- Iterations on winners: new hooks, lengths, and calls to action on winning ads from a proven concept, tracked as their own win rate next to new concepts.
- Element tagging: tags for hook, format, and offer show which elements recur across winners, as described in creative analytics.
The full sequence of a test, from hypothesis to variants, is set out in creative testing starts before you generate the variants.
Example: A quarterly win rate review for a fitness app
The review:
- Tested: 24 new concepts and 36 iterations on Meta, each judged against the best current ad on cost per trial start at Meta's 65% threshold.
- Not judged: 2 concepts and 3 iterations with too little delivery, counted separately.
- Concepts: 2 winners out of 22 judged, a 9.1% win rate.
- Iterations: 6 winners out of 33 judged, an 18.2% win rate.
- Overall: 8 winners out of 55 judged, a 14.5% win rate.
- Next quarter: concepts keep a fixed share of the test budget, the first iterations go to the two new concept winners, and the overall rate is compared with this quarter's 14.5%.
Creative win rate with Starti AI Studio
In AI Studio, Video Agent turns a brief into concepts, scripts, and storyboards, and a revision updates a single scene while the rest of the cut stays as approved. Smart Insight reads results across connected ad accounts. Its Winner, Scaling, Watch, and Fatigue signals help teams prioritize which campaigns and creatives to review, and each finding lists its supporting creatives, counterexamples, and sample size. Starti's creative team and the client make the final calls, and the findings go into the next brief.
Book a demo to plan a creative testing round.
Frequently asked questions
What is creative win rate?
Creative win rate is the share of new ad creatives that beat the current best ad in a test and go on to receive scaled spend.
How do you calculate creative win rate?
Divide the number of new creatives that beat the control by the number of new creatives judged in the period, and multiply by 100. Leave out creatives without enough delivery to judge, count inconclusive tests as non-wins, and report both next to the rate.
What is a good creative win rate?
Win rate has no industry standard, so a good one holds or rises against the team's own baseline under the same platform, threshold, and budget per variant. Spend data sets the context: in AppsFlyer's 2026 report, the top 10% of video creatives held more than 75% of spend in every vertical and format.
What confidence level declares a creative winner?
Meta treats 65% or higher confidence as a winning result in A/B tests, TikTok split tests use 90%, and Google Ads experiments test at 95%, with directional video results at 70% or 80%.
Should concepts and iterations be counted together?
Track them separately. Iterations on a proven concept face a different bar from new concepts, and a combined rate hides which kind of testing is producing winning creatives.






