Ad Lead Quality Analyzer

Meta optimizes for whatever conversion event you fire. For lead-gen and participant-recruitment campaigns that's almost always "signup" — but a signup is worthless if the lead never qualifies, never completes the requested action, or never gets paid out. The lowest-CPA campaign is often the one bringing in the worst leads.

This skill joins what the ad platform knows (spend, signups) with what your own product knows (downstream funnel) and replaces vanity CPA with true CAC per qualified lead. It then classifies every creative into actionable buckets so you stop scaling the wrong winners.

Core principle: The ad platform's CPA is a half-truth. Real optimization needs both halves of the funnel — pre-signup (the platform has it) and post-signup (you have it). Until they're joined, you're flying blind.

When to Use

• "Which ads are bringing in real leads vs. junk?"
• "True CAC per qualified contributor / customer / participant"
• "Why is my lowest-CPA campaign performing worst downstream?"
• "Audit lead quality across creatives / audiences / placements"
• "Should I trust Meta's CPA when scaling?"
• "Find the creatives that look like winners but aren't"

Pipeline Pattern Assumptions (Read First)

This skill is opinionated about what to measure (true CAC per qualified lead, with cohort maturation, with vanity scoring) and agnostic about how the data is sourced.

It assumes one of three standard attribution patterns:

If none of these patterns is wired up, the skill switches to tracking-gap mode — it produces a fix-the-tracking report instead of an analysis.

Phase 0: Discovery Interview

6 short questions. Don't proceed until each is answered (default = "I don't know — let's find out").

1. Where do downstream events live? (Postgres / MySQL / Airtable / custom internal admin / spreadsheet / "no idea")
2. Can the agent query that source directly? (DB credentials / API endpoint / CSV export / "needs a person to pull it")
3. Does the signup form capture utm_* params or fbclid? ("I don't know" → inspect the signup form's HTML / network requests)
4. Is the app sending CAPI events back to Meta for any downstream stage? (None / signup-only / signup + qualification / full funnel)
5. What is a "qualified lead"? (Default: ≥1 unit of value-producing action completed within 14 days of signup. Examples: first purchase; demo attended; subscription activated; trial converted; first task completed and paid out)
6. Cost basis per qualified lead? (Flat payout, variable, tiered by quality, or N/A — needed to compute margin)

Output of Phase 0: a one-paragraph Pipeline Brief stating the assumed pattern (A/B/C), the join key, the qualification definition, and any unknowns.

Phase 1: Tracking Validation (Gating Step)

Pull a sample of 10–20 recent signups from the downstream source. For each, check:

• Is utm_source / utm_campaign / utm_content present? (Or fbclid? Or lead_id?)
• Does the join key resolve back to a specific Meta ad?
• Are there orphan signups (in your DB but no Meta join key)?
• Are there orphan Meta signups (in Meta but no matching DB record)?

Coverage thresholds:

Output of Phase 1: a Data Quality Report with coverage %, sample of orphan records, and exact field-level findings.

Phase 2: Build the Per-Creative Funnel

For every ad / ad set / campaign with statistical volume (default ≥30 signups in the window), construct:

The skill should pull Meta-side data via the existing Meta Marketing API connection (MCP, native API, or pasted CSV) and downstream-side data via whichever source Phase 0 identified.

Phase 3: Compute True CAC

Per creative / ad set / campaign:

• Platform CPA = spend ÷ signups (what Meta reports)
• True CAC = spend ÷ qualified leads (what actually matters)
• Quality Multiplier = True CAC ÷ Platform CPA (how badly the platform is misleading you per ad — higher = worse vanity problem)
• Margin per qualified lead = (cost-basis or LTV-equivalent value) − True CAC

Phase 4: Score and Classify Each Creative

Compute three quality scores per creative with sufficient volume:

• Vanity score = 1 − (Started ÷ Signups). High = clicks but no work
• Audience-fit score = Approved ÷ Started. Low = wrong people getting through
• Retention score = Repeat ÷ Approved. Low = one-and-done

Then classify into action buckets:

Every classification cites the data and gets a confidence flag (sample size + CI on True CAC).

Phase 5: Cohort Maturation Handling

The biggest analysis trap: judging signups before they've had time to complete the funnel.

• Exclude signups newer than the qualification window (default 14 days) from "Cut" decisions
• Show two parallel views in the report:
  • Mature cohort (≥14 days old) — the basis for action
  • Recent cohort (<14 days) — leading indicator only
• If recent-cohort True CAC is diverging sharply from mature, flag a creative-fatigue or audience-shift hypothesis for investigation in meta-ads-analyzer

Phase 6: Generate Report

Use this exact structure.

1. PIPELINE BRIEF
   - Pattern (A/B/C), join key, qualification definition, unknowns

2. DATA QUALITY
   - Coverage %, orphan counts, confidence level

3. HEADLINE
   - Overall True CAC vs. Platform CPA
   - Overall Quality Multiplier
   - Period-over-period delta

4. PER-CREATIVE TABLE
   - Ad ID | Spend | Signups | Qualified | Platform CPA | True CAC | Quality Mult. | Vanity | Class

5. ACTION LIST (prioritized)
   - Cut (dangerous winners) → Scale (proven quality) → Investigate (low-vol promising) → Keep
   - Each action: hypothesis + expected impact + rollback plan

6. AUDIENCE / PLACEMENT PATTERNS
   - Which interests / lookalikes / geos / placements correlate with qualified leads
   - Which correlate with vanity signups

7. TRACKING GAPS (if any from Phase 1)
   - Specific fields, code locations, or events to wire up

Tracking-Gap Mode (Output if Phase 1 Fails)

If <50% o