Overview
Extracted from Meta Ads Manager .xlsx Upload a report to receive diagnostic results reflecting product involvement and operational philosophy, along with action options by intensity level.
5 Key Differentiators:
- Interpretation reflecting product involvement — Different ROAS benchmarks and interpretation methods for low-involvement (food/beauty) and high-involvement (education/B2B)
- Recommendations by operational philosophy — Branches into automatic trust type (algorithm-first) and active control type (prefers manual adjustment)
- Action options by intensity level (🟢🟡🔴) — No definitive commands; provides 3 options (conservative/moderate/aggressive) + two-sided interpretation
- Readability-first output — 7-Level hierarchical structure with progressive disclosure from summary to simulation
- Integrated demographic dimension analysis — Supports 4-dimensional cross-analysis of placement × age × gender
Attribution: Methodology adapted from agricidaniel/claude-ads v1.5.1 (MIT). 50-check audit matrix, weighted scoring, Quick Wins logic → Korean market adaptation. Details:
NOTICE.md §agricidaniel/claude-ads.
Triggers
This skill is always invoked for the following utterances:
- "Analyze my Meta ad report", "Diagnose my ad performance", "Analyze why ROAS is low"
- "Check performance by placement", "Should I cut Audience Network", "AN placement diagnosis"
- "Break down by age group and gender", "Do demographic analysis", "I want to see efficiency by persona"
- "Analyze my brand's ad report", "Check if my ad creatives are stale", "Creative fatigue diagnosis"
- "Check if my budget is appropriate", "Find ad spend waste", "Find leaking ads"
- "Compare two campaigns", "Check this month vs last month trends", "Multi-month comparison"
- "I'll upload my ad report", "Analyze xlsx", "Meta Ads Manager data analysis"
Workflow
[Step 1] 사용자 그룹 명시 선택 (인하우스 / 대행사 / 소규모)
↓
[Step 2] .xlsx 보고서 1~6개 업로드 + 사용자 입력 6종 확인
↓
[Step 3] 전처리 — 컬럼 자동 매핑 · 통합/분리 보고서 판단 · 정합성 검증 6종
↓
[Step 4] 5 자동 모드 판단 — 단일/통합/분리/다중월/다중캠페인 분기
↓
[Step 5] 9 분석 모듈 자동 실행 (인구통계 컬럼 유무에 따라 모듈 3·4 일부 분기)
↓
[Step 6] 7-Level 계층 출력 생성 (Level 1~6 필수, Level 7 선택)
↓
[Step 7] 강도별 액션 옵션 (🟢🟡🔴) 표준 템플릿으로 권장안 정리
↓
[Step 8] 선택 출력 형식 (HTML / DOCX / PPTX / MD) 생성
↓
[Step 9] DOCX/PPTX/MD → moai-core:ai-slop-reviewer → moai-content:humanize-korean 자동 체이닝
↓
[Step 10] 면책 문구 포함하여 최종 산출물 제공
User group input (explicit)
On first invocation, must be explicitly inputto select 1 of 3 groups (no auto-estimation).
| Group | Definition | Core Needs |
|---|---|---|
| 🏢 In-house | Advertiser's in-house marketer | Accuracy, consistency, internal reporting |
| 🤝 Agency | Ad agency AE | Efficiency, visualization, client reporting |
| 👤 Small-scale | Solo operator, CEO | Friendliness, actionability, easy explanations |
Group-specific output differentiation matrix (including analysis depth, tone, action emphasis, terminology explanation level) → references/A.md
6 types of user input
After uploading the report, enter the following 6 items:
| Item | Type | Options |
|---|---|---|
| Product category | Required | 8 presets (Food/Beverage · Fashion/Beauty · Health Functional Foods · IT/Digital · Household Goods · Education/Knowledge · Service/B2B · Other) |
| Product price range | Required | Low (approx. 30,000 KRW) · Mid (30,000-100,000 KRW) · High (100,000-300,000 KRW) · Premium (300,000 KRW+) · Direct input |
| Involvement | Auto-estimated + manually changeable | Auto-estimated based on category, price, and goal combination → user can manually change to low/mid/high |
| Campaign goal | Required | Sales · Lead · Traffic · Awareness · App Install |
| Operation period | Required | New (less than 1 month) · Early (1-3 months) · Stable (3-12 months) · Long-term (12 months or more) |
| Operation philosophy | Required | Auto-trust type · Active control type · Present both (default) |
Details of 6 inputs (default involvement for 8 category presets · involvement auto-estimation logic · operation philosophy output branching) → references/C.md
5 Auto mode determination
Analyzes the number of uploaded files, period, and dimensions to automatically select one of 5 modes:
파일 1개 + 노출지면만 → 단일 캠페인 진단 모드 (기본)
파일 1개 + 노출지면 + 인구통계 → 통합 분석 모드
파일 2개 + 같은 기간 + 다른 차원 → 분리 보고서 통합 모드
(노출지면 보고서 + 인구통계 보고서)
파일 2~6개 + 다른 기간 → 다중 월 비교 모드
(추세 분석 + 소재 라이프사이클 활성화)
파일 2~6개 + 같은 기간 + 다른 캠페인 → 다중 캠페인 일괄 모드
(대행사용)
그 외 → 사용자 확인 요청
Detailed branching logic for 5 modes + analysis items activated per mode → references/E.md
9 Analysis modules
When a report is uploaded, the following 9 modules automatically run:
| # | Module | Key output |
|---|---|---|
| 1 | Funnel decomposition | Impression→click→purchase funnel, stage-by-stage conversion rates, benchmark comparison by involvement |
| 2 | KPI decomposition | Absolute values (spend, impressions, clicks, purchases, revenue) + efficiency metrics (CTR, CPC, CVR, ROAS, AOV, CPA) + multi-month trends and change rates |
| 3 | Decomposition by dimension (v1.1 expansion) | Single decomposition by platform, placement, creative, concept, age, gender + 4-dimension cross (ad × placement × age × gender) |
| 4 | General-purpose matrix (v1.1 expansion) | Grade matrix based on user-selected axes (S/A/B/C/D/F) |
| 5 | Leakage and Pareto | Absolute value and share of spend with 0 purchases, Pareto analysis of combinations contributing 80% of revenue |
| 6 | Creative lifecycle | Ad name pattern-based aging diagnosis (activated when multi-month data is available) |
| 7 | Learning phase entry | Evaluation of the possibility of achieving the 50 conversions per week standard, recommended direction for ad set structure |
| 8 | Budget adequacy | Adequacy compared to recommended daily budget of product price × 3 (activated when monthly ad budget is entered) |
| 9 | Scenario simulation | Expected revenue and ROAS change by action + optimistic/realistic/pessimistic confidence intervals (optional) |
If demographic columns (age·gender) are absent, only the demographic dimensions of Modules 3 and 4 are disabled — the remaining analysis proceeds normally.
9 Module Detailed Specifications (directly citing D.1-D.9, including grading criteria and obsolescence determination formulas) → references/D.md
7-Level Output Hierarchy
All output formats follow the following 7-level hierarchy:
Level 1: 한 줄 요약
Level 2: 핵심 발견 3가지 (카드형)
Level 3: KPI 대시보드 (차트)
Level 4: 영역별 진단 (지면·소재·연령·성별 등)
Level 5: 누수·기회 분석
Level 6: 강도별 액션 옵션 (🟢🟡🔴)
Level 7: 시뮬레이션 (선택)
Level 1-6: Required for all output formats / Level 7: User selection
Action options by intensity (🟢🟡🔴)
All recommendations have the following 5 elements of the standard template without prescriptive commands:
### [발견] {제목}
**사실**: {데이터로 확인된 내용}
**해석 (양면)**:
- A 입장 (자동 신뢰): {해석}
- B 입장 (적극 통제): {해석}
**권장 옵션**:
🟢 보수안: {액션} | {효과} | {리스크}
🟡 중도안: {액션} | {효과} | {리스크}
🔴 적극안: {액션} | {효과} | {리스크}
**의사결정 가이드**: 예산 규모·운영 기간·운영 철학별 권장
Full action recommendation standard template + tone adjustment examples by group → references/F.md
4 Output Formats
| Format | Priority | Primary Use | Features |
|---|---|---|---|
| HTML | 1st Priority | Immediate check·presentation·demo | React+Recharts single file, interactive charts |
| DOCX | 2nd Priority | Client·internal reporting | 8-section structure, including tables·charts |
| PPTX | 3rd Priority | Presentation·lecture | 10-15 slides |
| MD | Supplementary | Practical checklist | Checkboxes+priority, optimized for Notion·Slack |
HTML design tokens (moai-content:html-report Common palette — SSOT):
| Token | Hex | Purpose |
|---|---|---|
--ivory |
#FAF9F5 |
Page background |
--paper |
#FFFFFF |
Card·panel background |
--slate |
#141413 |
Body text |
--clay |
#D97757 |
Emphasis·links |
--clay-d |
#B85C3E |
High-risk indicator (🔴 aggressive option) |
--oat |
#E3DACC |
Secondary background·chart default bars |
--olive |
#788C5D |
Positive signal (🟢 conservative option) |
--g300 |
#D1CFC5 |
Borders·dividers |
4 Output Format Detailed Specifications (full HTML CSS variables · DOCX 8 sections · PPTX slide composition · MD format) → references/G.md
Expression principles ✓/✗
| ✓ Recommended | ✗ Prohibited |
|---|---|
| Present facts → two-sided interpretation → options by intensity (directionality level) | Prescriptive commands ("Turn OFF") |
| Traffic light system (🟢🟡🔴) | Meta Ads Manager operation step guide ("Edit ad set → Placements → ...") |
| Present improvement directions by placement·creative·demographics | Meta UI captures·screenshots |
| Specific action recommendations (e.g., "Review turning OFF AN placement") | Biased generalization |
| Tone adjustment by user group | Unverified benchmarks (source not specified) |
| Cite sources when quoting benchmark figures (public benchmark sources) | Forcing operational philosophy |
Expression level examples:
- ✓ "Since AN placement ROAS has been 0 for two consecutive months, recommend reviewing turning OFF placement"
- ✗ "In Meta Ads Manager, edit ad set → Placements → Manual selection → Uncheck Audience Network"
Disclaimer Text
The following text is included in all outputs:
This analysis result is a data-driven diagnosis. Final operational decisions may vary depending on the user's experience, context, and strategy, and the Skill provides decision-making options, not correct answers.
Users should verify the actual method of application in Meta Ads Manager themselves.
Full disclaimer text + explicit list of limitations + exceptions prohibiting time predictions → references/I.md
Automatic Chaining
meta-ads-analyzer (보고서 분석)
↓ DOCX/PPTX/MD 텍스트 산출물
moai-core:ai-slop-reviewer (AI 슬롭 검수)
↓
moai-content:humanize-korean (한국어 AI 티 제거)
↓ 인프라 문제 발견 시
moai-marketing:pixel-audit (픽셀·1st Party 데이터 인프라 진단)
↓ 랜딩 문제 발견 시
moai-marketing:landing-page-conversion-audit (랜딩 6섹션 전환율 진단)
Because the HTML dashboard's main deliverables are React components + charts,
moai-core:ai-slop-reviewer·moai-content:humanize-koreanit is excluded from chaining targets (only applies to body copy and interpretive text).
Pair Skill Boundaries
| Pair Skill | Responsibility | Boundary with this Skill |
|---|---|---|
moai-marketing:pixel-audit |
Meta Pixel and Infrastructure Diagnosis | Pixel installation, CAPI, Lookalike seed → moai-marketing:pixel-audit, report result analysis → this Skill |
moai-marketing:landing-page-conversion-audit |
Landing Page 6-Section Conversion Rate Diagnosis | Landing UX after ad click → Pair, Ad system KPIs (impressions/clicks/purchases/revenue/ROAS) → this Skill |
moai-marketing:campaign-planner |
Campaign Pre-Planning | Pre-planning (goals, targets, creative concepts) → moai-marketing:campaign-planner, post-execution diagnosis and actions → this Skill |
(Pair Separation Principle)
Reference Appendix
| Appendix | Content | Link |
|---|---|---|
| A | User group differentiation matrix (A.4 output differentiation 6 items) + explicit input UX guide | references/A.md |
| B | 12 required columns auto-mapping + 4 rules for handling missing columns | references/B.md |
| C | Details of 6 user inputs (8 category presets, unit price, involvement level, goal, operating period, operating philosophy) | references/C.md |
| D | Detailed specifications for 9 analysis modules (direct citations from D.1-D.9) | references/D.md |
| E | 5 automatic mode determination logic (E.1-E.5 branch details) | references/E.md |
| F | Standard template for action recommendations (F.1 five elements + F.2 ✓/✗ examples + F.3 tone by group) | references/F.md |
| G | 4 output format specifications (HTML design tokens, DOCX 8 sections, PPTX slides, MD checkboxes) | references/G.md |
| H | 6 consistency checks + Layer 2 MCP call guide | references/H.md |
| I | Full disclaimer text + explicit limitations + exceptions prohibiting time predictions | references/I.md |
| J | Sister skill candidates (creative-generator, budget-planner, curriculum-builder) | references/J.md |
| K | Sample report fixture guide + test examples | references/K.md |
Source
Methodology adapted from agricidaniel/claude-ads v1.5.1 (MIT).