● Media Buying Course · Day 12 of 20 · Week 3: Creative

Creative testing frameworks

Yesterday: creative is the lever. Today: how to test it systematically instead of guessing. The agencies that win don't have better taste - they have a better testing SYSTEM that finds winners faster and cheaper. This is the engine of your future business.

1The goal of testing

You're not trying to "find one perfect ad." You're running a repeatable process to (a) discover winning angles/hooks, (b) kill losers fast and cheap, (c) feed winners into scaling. A funnel for ideas: many concepts in → a few proven winners out → those get the budget.

Vocabulary: Concept a big idea/angle ("founder story," "problem-agitate," "UGC unboxing"). Variation a tweak within a concept (different hook). Iteration a new version built from what a winner taught you.

2Two testing approaches

🧪 A/B testing (controlled)

Meta's split test isolates ONE variable with mutually exclusive audiences and a clear statistical winner. Use for clean, trustworthy answers to a specific question. Slower, needs volume for significance.

🌊 "Dump and let it ride" (algorithmic)

Multiple creatives in one ad set; the algorithm allocates to winners (Day 3 grouping). Faster, naturally aligned with how Meta works, but not a clean experiment - it picks early and may not give every creative a fair shot.

Most modern buyers lean on the second for creative discovery (speed, volume) and reserve formal A/B tests for high-stakes structural questions (landing page A vs B, a big format bet).

3A sane testing structure

A dedicated testing campaign (often ABO or low-budget CBO) where new concepts compete.
Each batch = several distinct CONCEPTS (different angles), not five near-identical images.
Enough budget/time to gather signal (~50 events to truly know; hook rate & CTR give faster directional reads).
Winners graduate to the main scaling campaign; losers are cut; winners spawn iterations.

Test big differences first

Comparing two slightly different blue backgrounds wastes spend. Comparing "emotional story" vs "blunt discount" vs "social-proof montage" teaches which ANGLE the market wants - the high-value learning. Refine details only after you know the winning direction.

Analogy · plant breeding

You sow many varieties (concepts), see which sprout strongest in real conditions (the market), keep the best, cross-breed them into the next generation (iterations). You don't fall in love with one seed; you run generations. Over time your garden (ad account) is full of descendants of proven winners.

▤ In Ads Manager · running tests

A/B Test tool (flask icon)formal split, declares a winner, controls overlap

Algorithmic: one ad set, multiple adsread ad-level results

Dynamic Creativeauto-combine headlines/images/videos

Per-ad metricscut on weak hook/CTR early, weak CPA later

⚠ What gets people wrong

(1) Testing trivial variations (button color) instead of angles. (2) Declaring winners on tiny samples - noise, not signal. (3) Cutting creatives in 12 hours (learning-phase volatility, Day 5). (4) No system - random uploads, no documentation, nothing compounds. (5) Confusing engagement (likes) with performance (CPA). Your edge: a documented, repeatable pipeline that compounds learnings across clients.

Recap - 30 seconds

Testing = a repeatable funnel for ideas: many concepts in, proven winners out, winners get budget.
A/B test = clean controlled answer; algorithmic (many ads, one ad set) = faster discovery.
Use a dedicated testing campaign; winners graduate, losers get cut, winners spawn iterations.
Test big angle differences first; refine details only after the winning direction is known.
Don't judge on tiny samples or 12-hour windows; document so learnings compound.

Day 12 · Week 3: CreativeTomorrow → Day 13: Reading creative diagnostics