meta-ads-analyzer

# Meta Ads Analyzer

Most "Meta Ads analysis" stops at "this CPA is high, pause it." That's wrong more often than it's right. Meta's delivery system optimizes for **marginal efficiency** — the cost of the *next* conversion — not average efficiency across a snapshot. A segment with a higher average CPA is often the one keeping your overall campaign cheap. Pausing it makes things worse.

This skill diagnoses Meta campaigns the way a senior media buyer would: at the right evaluation level, accounting for learning state, separating noise from signal, and explaining *why* the system is making the decisions it's making before recommending any change. It can also audit whether the account supports the complete customer journey without assuming that TOF, MOF, and BOF must be separate campaigns.

**Core principle:** Holistic first, then drill down. Marginal over average. Customer-journey coverage over rigid funnel structure. Dynamic over static. Every recommendation is a testable hypothesis with expected impact, not a directive.

For account audits, full-funnel reviews, or questions about what is missing, read and apply [references/customer-journey-coverage.md](references/customer-journey-coverage.md) before analyzing the account.

## When to Use

- "Analyze my Meta Ads campaign performance"
- "Why is the system spending more on the higher-CPA placement?"
- "Diagnose what's wrong with this ad set"
- "Should I pause this audience / placement / ad?"
- "My CPA jumped — is this normal or a real problem?"
- "Audit this campaign before I scale budget"
- "I exported my Meta data — what does it actually mean?"
- "Audit my Meta ad account and tell me what is missing"
- "Do I have enough TOF, MOF, and BOF coverage?"
- "Why are customers not moving through the funnel?"
- "What ads should I create next?"

## Phase 0: Intake

1. **Campaign data** — One of:
   - CSV export from Meta Ads Manager (Campaign / Ad Set / Ad level + breakdowns)
   - Pasted performance table
   - Screenshots (we'll extract the metrics)
   - Live data via your existing Meta Marketing API connection
2. **Campaign setup**:
   - Objective (Awareness / Traffic / Engagement / Lead Gen / Conversions / Sales / App Installs)
   - Budget type (Advantage+ Campaign Budget = CBO, or Ad Set Budget = ABO)
   - Placements (Automatic vs. manual)
   - Number of ad sets and ads
3. **Time period** — Date range covered, with any known events (creative refresh, budget change, audience edit, account issue)
4. **Target metrics** — CPA target, ROAS target, or "no target — benchmark me"
5. **Funnel context** (if relevant) — On-platform conversion vs. website event vs. downstream qualification rate
6. **What's making you ask?** — Specific concern ("CPA up 40%"), routine review, or pre-scale audit
7. **Account coverage evidence** (for account/funnel audits, when available):
   - Campaign objective, optimization event, and attribution setting
   - Audience strategy, exclusions, and retargeting windows
   - Creative format, message, proof, offer, and landing-page destination
   - Pixel/CAPI and relevant conversion-event health
   - Campaign, ad-set, and ad-level spend and results
8. **Report style** — `guided` by default; use `expert` when the user asks for technical detail or demonstrates strong media-buying knowledge

Do not block when some coverage fields are absent. Record what is missing, lower confidence, and distinguish "no evidence available" from "the account has no coverage."

## Phase 1: Identify the Correct Evaluation Level

This is the most important step. **Evaluating at the wrong level is the #1 source of wrong recommendations.**

| Campaign Setup | Correct Evaluation Level | Why |
|---|---|---|
| Advantage+ Campaign Budget (CBO) | **Campaign level** | System pools budget across ad sets — only campaign totals reflect reality |
| Automatic placements (no CBO) | **Ad Set level** | System pools budget across placements within the ad set |
| Multiple ads in 1 ad set | **Ad Set level** | System pools delivery across ads |
| Manual placements + ABO | Placement / Ad Set level | Each is independent |

**Output for this phase:** State the evaluation level explicitly and explain why before any metric is interpreted.

> If asked "is this Meta placement underperforming?" on a CBO campaign, the answer is "wrong question — at CBO the placement-level CPA is misleading. Here's the campaign total..."

## Phase 2: Check Learning Phase Status

Before judging anything, check delivery state per ad set.

**Learning state checklist:**
- Status is `Learning` (delivery less stable, CPA typically higher, results not predictive)
- Exits after ~50 optimization events within 7 days of last significant edit
- Shops ads exception: 17 website purchases + 5 Meta purchases
- Status `Learning Limited` = can't get enough events → flag as a structural issue, not a performance issue

**Significant edits that reset learning:**
- Targeting changes
- Optimization event change
- Creative changes (large)
- Bid strategy / amount changes
- Budget changes >20%

**Output for this phase:** Per ad set, mark `Active` / `Learning` / `Learning Limited`. Caveat all conclusions for anything in learning. **Do not recommend pausing a Learning ad set based on CPA alone.**

## Phase 3: Diagnose with Meta-Specific Lenses

Run the diagnosis through these six lenses. Each one explains a different class of "weird" behavior.

### 3A: Marginal Efficiency Analysis (Breakdown Effect)

The Breakdown Effect: the system shifts budget toward segments where the *next* conversion is cheapest, not where the *average* conversion is cheapest. A segment can have a high average CPA in a breakdown report and still be the right place for budget.

**How to spot it:**
- Time-series the segment's CPA. If marginal CPA is rising sharply, expect the system to shift budget out — even if average looks fine.
- A breakdown row with high average CPA + high spend usually means the system found cheap marginal conversions there earlier in