The standing test bench
Most studios don't have a creative problem; they have a creative throughput problem. Winning concepts exist, but nobody ships variants fast enough to find them before the old ones fatigue. The bench fixes throughput: a fixed monthly cadence of concepts, tests, and honest funerals.
What ships
20–40 concepts a month across video, static, and UGC-style formats. Each arrives as a hypothesis: the hook, the audience it should move, the number it should move.
What decides
Install rate, CPI against control, and the day-1 retention of the cohort each concept brings. A pretty ad that recruits tourists dies like an ugly one.
What you keep
Every asset, every test read, and the kill-log: a running document of what died, why, and what that teaches your next brief. Clients call it the cheapest strategy doc they own.
Why the kill rate is the product
About 83% of what we ship dies in testing, and that number staying high is the point. The market moved this way for a reason: 78% of top-quartile campaigns now refresh creative at least weekly, nearly double the share from a year before, because ad platforms exhaust an audience's patience faster every year. A bench that ships winners only is a bench that ships too slowly.
The economics work because a killed concept costs a test budget, while a fatigued account costs the difference between your CPI and your category's: gaming averaged around $3.90 per install in 2025, with iOS running roughly three times the price of Android. When the gap between your number and the median starts widening, the creative is nearly always the leak. The testing framework we run is published in full, and the Graveyard shows the discipline applied to our own ideas.
Where the bench plugs in
Arcade mode assumes your media team keeps the wheel. We take read access, define test cells that won't contaminate your learning phases, and hand your buyers a weekly winners memo. If the same roadmap should include us running media, that's campaign mode; if the data says your video winners deserve interactive siblings, the playable bench bolts on.
Bench questions
Twenty to forty shipped concepts, depending on how fast your account gives statistically honest reads. Top global advertisers push 2,400 to 2,600 creatives a quarter with in-house armies; a focused bench of 30 a month, killed ruthlessly, is what actually moves a mid-size account. Volume without a kill rule is just noise you paid for.
A concept dies when it underperforms the control on install rate and cost per install with enough impressions to trust the read, or when its cohort's day-1 retention comes in rotten. Median time to that decision on our bench is six days. Every kill is logged with the numbers, and the instructive ones get a public epitaph in the Graveyard.
Read access, yes; the wheel, no. Arcade mode is built to feed an in-house media team. We ship the concepts, define the test cells, and read the results in your accounts, while your team keeps bidding. If you want us driving media too, that's campaign mode, and it's priced separately.
Put your account on the bench
Bring your current top three ads to the working session. We'll tell you which one is already dying.
Start a run