Inside the creative testing engine

Most ad accounts don't have a creative testing engine. They have a creative production line and a hope that something will pop.

A real engine has rules, for how concepts become tests, how tests become winners, and how winners become scale.

The four-layer engine

1. Pillars: the 3–5 themes the brand is allowed to talk about. Comes from positioning, not the platform.
2. Angles: within each pillar, the specific lens a piece of creative takes. (E.g., pillar: "energy"; angles: "morning routine," "afternoon slump," "post-workout.")
3. Hooks: the first three seconds. The reason someone stops scrolling.
4. Iterations: once a hook is working, the same hook re-cut in three to five formats: voiceover, text overlay, founder POV, UGC.

Tests don't move up the stack until the layer below is doing its job. A new pillar gets new angles, new hooks, new iterations. A winning hook gets new iterations, not a new angle.

What gets tested

Each week, the test slate is short and decided in advance:

• 2 new hooks within an existing winning angle (iteration tests).
• 1 new angle within an existing pillar (angle test).
• 1 new pillar test, quarterly (not weekly).

Resist the urge to test everything every week. Discipline at this layer is what turns a creative team into a creative engine.

The decision rules

For each test, agree the rule before the data comes in:

• Spend threshold: minimum spend before a decision is allowed.
• CTR floor: under this, kill regardless of cost.
• CPA ceiling: over this, kill regardless of CTR.
• Iterate trigger: if it clears CPA but underperforms by less than 20%, iterate the hook, don't kill the angle.

The point of decision rules: take the emotion out of Monday morning.

Why most engines stall

They stall on production capacity. A winning angle needs 4–6 iterations to scale safely; teams produce 1 and call it a day. Then the algorithm gets bored, performance dips, and the team blames the platform.

Build the production capacity for iteration before you go looking for the next angle. Compounding sits on the iteration axis, not the novelty axis.

The one-line version

Pillars produce angles. Angles produce hooks. Hooks produce iterations. Iterations produce scale.

Run that engine, weekly, for a quarter. You'll know what works, why, and what to do next.