skills/Creative Strategist/references/evaluation-framework.md

# Evaluation Framework

Shared reference for evaluating creative assets. Used by qa-feedback, analyze-ad, and any skill that needs to assess whether a creative will perform.

---

## The 4-Question Evaluation Framework

Evaluate every creative by answering four questions in order. They're ranked by importance — failing the first one overrides strengths lower down.

### 1. Does this make sense fast?

Within the first 1-3 seconds, the viewer needs to understand what this is about. Confusion is the biggest performance killer. If someone has to work to understand it, they leave.

*Fails when:* the hook is vague or abstract, too much is happening visually, the message could mean multiple things, or there's no clear signal of what's being offered.

### 2. Will the right person feel like it's for them?

The creative should make the intended viewer feel recognized — like this was made for someone in their situation. If it could be for anyone, it's effectively for no one.

*Fails when:* the language is generic, the tone doesn't match how the target audience actually talks or thinks, the casting or setting doesn't reflect who it's trying to reach, or the problem being named isn't one the target audience actually has.

### 3. Will they believe it?

Trust is where most performance ads fail. Claims need to feel credible, not just be true. A claim that's accurate but unsupported still reads as hype. Anything that sounds like a salesperson pushing too hard triggers skepticism before the viewer has even processed the message.

*Fails when:* benefits are stated without proof, the tone is overly promotional or corporate, social proof is absent or feels staged, or the promise sounds too good to be true.

### 4. Will they take the intended action?

The creative needs to build enough desire and make the next step feel obvious. A viewer who's interested but unclear on what to do next doesn't convert.

*Fails when:* the CTA is missing, buried, or unclear; there's no reason to act now vs. later; or the ad promises something different from what the landing page delivers.

---

## Metric Pattern Correlations

When performance data exists, use it to diagnose which question is failing. Don't guess when the numbers can tell you.

| Metric Pattern | Which Question Is Failing |
|---|---|
| Low thumbstop | Q1 — opening isn't stopping the scroll. Everything downstream is moot until this works. |
| High thumbstop, low hold rate | Q1 passes, but middle loses people — something after the opening creates friction or loses the thread. |
| High thumbstop + hold rate, low CTR | Q2 or Q3 — viewers stay but don't act. CTA unclear or no compelling reason to click. |
| High CTR, low conversion | Q4 — ad promises something the landing page doesn't deliver. |
| Strong engagement, low conversion | Q3 — creative entertains but doesn't persuade. |

If no performance data exists, say so. Don't present judgment as if it's data-backed.

---

## The Ready / Iterate / Rethink Call

Every evaluation ends with one of three calls. Make it clearly. Don't bury it.

**Ready.** The creative can run as-is. Any notes are polish, not blockers. Lead with what's strong.

**Iterate.** There's a workable foundation but specific problems need fixing before it runs. Name what to fix in priority order. Be specific enough that the team can act without a follow-up conversation.

**Rethink.** Something fundamental isn't working — the audience signal, the core message, the proof structure, or the format itself. Surface fixes won't save it. Name exactly what's broken and why it can't be patched.

**The difference between Iterate and Rethink:** Iterate means you can fix what's there. Rethink means the angle, the brief, or the approach needs to change — not just the execution.

Don't soften Rethink to avoid disagreement. Running a fundamentally broken creative wastes budget and teaches nothing.

---

## Multi-Persona Feedback

Multi-persona reviews help teams see blind spots by evaluating the same creative through different lenses. Useful when stakeholders have conflicting intuitions about an ad.

**How it works:**
1. The user defines 2-4 personas and what each one cares about.
2. Each persona evaluates the creative against their stated priorities — not the same observations reworded in a different tone.
3. Where personas agree, the signal is strong. Where they disagree, there's a tradeoff the team needs to decide on.

**Quality bar for persona feedback:**
- Each persona must surface at least one observation the others don't
- Personas should disagree in useful ways that reveal real tradeoffs
- Feedback must be specific to the creative being reviewed, not generic role-based commentary
- If a persona would have no meaningful reaction, say so rather than forcing feedback

If the user asks for multi-persona feedback without defining personas, ask them to describe 2-4 with their priorities and concerns.

---

## Performance Context Awareness

Performance data can mislead if you don't account for context. Always consider these before drawing conclusions.

**Funnel stage matters.** A prospecting ad that drives strong conversion is a real signal. A retargeting ad that converts is expected — the audience was already warm. The more important question for retargeting is whether the creative is moving people who stalled, not just converting people who were already close. When comparing creatives across funnel stages, be explicit about which stage each belongs to. Don't flatten them into a single ranking without accounting for structural differences.

**Budget starvation.** A creative that looks weak may have just been starved of budget. A new ad competing against strong established winners may never receive enough spend to gather signal. Low performance could be structural, not creative.

**Account stage.** Early-stage accounts are still in the learning phase — results are often unstable and not representative. Mature accounts have established baselines. Calibrate the weight you put on results accordingly.

**Test fairness.** Ask: did this ad have a fair chance? A test only teaches you something useful if the ads had equal opportunity.