How Many Ad Creatives Should You Test? A Framework

Test several genuinely different concepts at a time — enough that at least one has a real chance of hitting, but few enough that each gets meaningful spend. For most small accounts that means a handful of distinct angles per testing round, each with a couple of hook variations; larger accounts run continuous pipelines with new creatives entering every week. The exact number depends on your budget and margins, but the principle is fixed: most creatives lose, so the account that tests more distinct ideas finds winners faster.

Why does creative volume matter so much?

Because creative performance follows a hit-driven distribution — a small share of ads drives most of the results, and nobody can reliably predict which ad that will be. Media buying levers have largely been automated away by the platforms; the creative is the targeting now. Testing one ad at a time means betting the account on a single guess. Testing a steady stream of concepts turns finding winners from luck into a process.

What counts as a "different" creative?

This is where most testing goes wrong. Ten versions of the same video with different background music is one test, not ten. Structure your rounds in two layers:

• Concepts (angles): genuinely different arguments — problem/solution, us-vs-them, founder story, demo, objection-handling. Test a few of these per round.
• Variations: within a promising concept, vary the hook, the opening visual, the actor, or the length. A hook generator is useful for producing these quickly.

Find the winning angle first; then multiply it with variations. Iterating on a loser wastes budget no matter how many versions you cut.

How do budget and margin change the number?

Each creative needs enough spend to produce a readable signal before you judge it — so your testing count is capped by budget divided by that per-creative minimum, which itself depends on your price point and conversion path. Higher-margin products can afford wider tests; thin-margin products should test fewer creatives more decisively. There's no universal number: work backward from what your economics allow, and sanity-check the math with a ROAS calculator so you know what a "winner" even means at your margin.

How do you judge the test?

Kill fast on leading indicators, confirm on money metrics. Hook rate tells you within a small amount of spend whether the opening works; watch-through and click-through tell you if the body holds; ROAS over a full round decides who survives. Losers get cut without sentiment. Winners get a fresh set of variations in the next round.

What's the bottleneck — and how do you remove it?

Almost always production. Teams know they should test more; they just can't make ads fast enough. That's the case for generating instead of shooting: Polaris connects to Claude via MCP, and you generate ads directly inside Claude by attaching a product photo and describing the angles. Batches render in parallel, so a full testing round of UGC-style ads is a chat session, not a production sprint — and the first batch of 12 is free, which conveniently is a testing round on its own.