Build a Creative Testing System

    You are a performance creative strategist building a testing system, not running a single test.

What I'm advertising: {{offering}}
Current creative and how it's performing: {{current_creative}}
Monthly ad spend and conversion volume: {{spend_and_volume}}
Production capacity — people, budget, turnaround: {{production_capacity}}
What I've learned so far about what works: {{existing_learnings}}
Platform: {{platform}}

Produce:

**The testing hierarchy.** Test in order of expected effect size, because most teams invert this:
1. **Concept** — the underlying idea and format. Largest effect by far.
2. **Hook** — the first three seconds, holding the concept constant. Cheap and high-return.
3. **Offer and message framing.**
4. **Execution details** — copy, thumbnail, captions, colors. Smallest effect, most commonly tested.

Say where {{current_creative}} sits and what level to work at next.

**Volume target.** Given {{production_capacity}} and {{spend_and_volume}}: how many new creatives per month, and how many concurrent tests the account can actually resolve. Set this against the reality that testing volume correlates strongly with performance — advertisers producing many new creatives monthly substantially outperform those producing few — but also against the constraint that a test needs enough conversions to distinguish a real difference from noise. If {{spend_and_volume}} is low, say plainly that most creative tests won't reach significance and recommend judging on directional signal plus consistency across weeks rather than waiting for statistical certainty.

**Test design.** How many variants at once, how budget is split, how long to run before judging, and what to hold constant. Include the rule against changing anything mid-test and the rule against calling a winner on day one.

**Decision rules, written in advance.** What result means scale it, kill it, or iterate on it. Include the threshold below which a difference is noise. Pre-committing to these prevents the usual outcome, where whichever creative the team liked gets kept.

**Fatigue management.** How to detect it — rising cost per thousand impressions, falling click-through, frequency climbing past the point where costs start rising noticeably — and what to do. Note that costs can rise substantially once fatigue sets in and that introducing fresh creative often restores them within days, so this is one of the fastest available wins. Set a frequency ceiling and a refresh cadence appropriate to how quickly creative burns out in this category.

**Iterate on winners.** When something works, the next move is variations of it — new hooks on the same concept, a different length, a different presenter — not abandoning it for a new idea. Most of the return comes from mining a winning concept properly.

**The learning library.** How to record what each test taught in a form that survives staff changes: the concept, the variable, the result, and the conclusion. Against {{existing_learnings}}, structure what I already know so it stops being folklore.

**Kill the sacred cows.** The rule that lets data override the creative the founder likes, agreed before the test rather than argued after it.

**Monthly rhythm.** What happens each week — production, launch, review, retire — so this runs as a process rather than a project.