Concepts vs Variants: Sizing a Live Creative Slate

The question our 2025 math answered differently

Our 2025 guide, Portfolio Theory for Creative: How Many Concepts to Run, sized a creative portfolio by working backwards from statistical significance: set a minimum sample size per concept, divide the test budget by the cost of reaching it, and let the result cap how many concepts run in a cycle. The maths still holds. What has changed in a year is that a UA manager asking how many concepts to run now reaches for a single number borrowed from a conference talk or a network rep rather than sitting down with the calculation, especially since AppsFlyer's State of Gaming for Marketers 2026 put top advertisers at 2,400 to 2,600 creative variations produced per quarter, a figure these pages covered in January under the headline "2,500 creatives a quarter." That is a production number, not a portfolio number. Treating it as a target skips the decision that matters: how many genuinely distinct concepts, as opposed to variations on one concept, should run at the same time?

Concepts, not variants

A concept is a hook, a mechanic demonstration, an emotional angle. A variant wraps a different actor or colour palette or end card or length around that same idea, and the 2025 sizing formula was always meant to size the first thing rather than the second; run it against variant counts and it flatters a portfolio that isn't diversified at all. Networks count variants because variants are easy to count, and internal reporting follows the networks. A studio chasing the AppsFlyer number by producing forty cuts of the same three ideas hasn't diversified its risk, it has produced overhead. Portfolio theory only applies once you're counting concepts.

Treat a live creative slate the way a fund manager treats a book of positions. Concentration in one high-performing concept works beautifully until that concept fatigues or a competitor copies it, at which point the whole account's performance moves at once. Spread spend across too many unproven concepts instead and you dilute the signal on all of them, slowing the point at which any single one earns enough spend to read cleanly.

Neither extreme is a strategy so much as the absence of one.

A working range

Most mid-size UA teams do better running somewhere between four and eight live concepts per major segment at any one time, counting by segment rather than by network or by market. Two or three of those carry the bulk of spend as proven performers. Another two or three sit in active testing with enough budget behind them to reach statistical usefulness, while you hold back no more than one or two as genuine long-shot bets. The range isn't a formula. It's a starting discipline, wide enough to survive one concept fatiguing without the whole slate going with it, and narrow enough that every concept picks up enough impressions per week for the weekly read to mean something.

The mistake most teams make isn't the range, it's the churn rate. Kill a concept that hasn't proven itself within two to three weeks of adequate spend; leaving it running at low volume just in case is how a slate rots. Zombie concepts that never get enough spend to read clearly and never get killed waste more creative budget than any individual bad hook ever does.

Where the AppsFlyer number still matters

The 2,400 to 2,600 variation figure is still useful, just not as a target. It describes the volume of iteration the top of the market runs against each concept once one earns spend: five, ten, sometimes twenty variants per proven idea, tested for what keeps a hook fresh rather than to find a new hook altogether. Read that way, the number describes what happens after a concept has already won its place in the portfolio. It says nothing about how many concepts should compete for that place to begin with.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• Portfolio Theory for Creative: How Many Concepts to Run
• Creative diversity is a portfolio risk decision