金大哥 - Meta Ads Analysis Skill 详情

Required Credentials

| Credential | Where to Get | Used For | OAuth Scope | |-----------|-------------|---------|-------------| | META_ACCESS_TOKEN | Meta Developer Console → Graph API Explorer → Generate Token | All Meta Marketing API calls | ads_read (read-only, least-privilege) | | META_AD_ACCOUNT_ID | Ads Manager URL: adsmanager.facebook.com/act_XXXXXXXXX | Identifying which account to query | — |

When to Use

Loaded when user wants to understand how their Meta Ads campaign is performing — either a specific aspect (audience, creative, overall) or a full diagnostic. Can run standalone or as prerequisite for meta-ads-recommendation.

Step 0: Understand the Campaign

Before pulling any data, get context on what this campaign is trying to do. Ask 1–2 quick questions if not already clear from conversation:

"Is this an always-on campaign, or tied to a specific promotion?"
"Are you testing new things, or trying to scale what's already working?"

Keep it conversational. If META_ACCESS_TOKEN or META_AD_ACCOUNT_ID is not available, tell the user exactly what is needed and do not proceed with API calls.

Step 1: Gather Campaign Structure

Ask the user for their Meta Ads account access. Use the Meta Marketing API to fetch:

Campaign level

Campaign name, ID, status (is it ACTIVE?)
Objective (what counts as a "result" — leads, purchases, clicks?)
Budget type (CBO = campaign-level budget optimization, or ABO = adset-level)
Bid strategy, daily budget, start date

Adset level

Adset name, ID, status
Optimization goal (this determines what "results" means for this adset)
Targeting countries, budget, bid amount

Ad level

Ad name, ID, status
Which adset it belongs to

Store this as campaign_context. Reuse across the session — don't re-fetch.

Step 2: Confirm Time Range and Attribution Window

Always confirm the time period and attribution window before pulling performance data:

Time range: "What time period would you like to analyze? (e.g., last 7 days, last 30 days, or a specific date range)"
Attribution window: "What attribution window are you using? Default in Meta is 7-day click. For e-commerce, 1-day click often gives a clearer signal. For lead gen, 7-day click is usually better."

| Campaign type | Recommended attribution window | |--------------|--------------------------------| | E-commerce / purchase | 1-day click | | Lead generation | 7-day click | | App installs | 7-day click | | Brand awareness | 28-day view |

Convert to YYYY-MM-DD,YYYY-MM-DD format. Default granularity:

≤14 days → daily
15–60 days → weekly

Step 3: Route the Analysis

| User is asking about | What to analyze | |----------------------|-----------------| | Overall / general health | Campaign-level metrics, no breakdown | | Audience / country / platform / age / gender | Adset or ad level with breakdowns | | Creative / which ad is better | Ad-level metrics | | Everything / full diagnostic | All three in sequence | | Follow-up on existing data | Use data already in session — don't re-fetch |

Step 4: Pull and Interpret Performance Data

Use the Meta Marketing API to fetch metrics for the relevant level and time range.

Key Metrics to Collect

At campaign or adset level:

Spend, impressions, frequency
CPM (cost per 1,000 impressions)
CPC (cost per link click)
CTR (link click rate)
LPV rate (landing page view rate = landing page views / link clicks)
Results, cost per result
Conversion rate (results / landing page views)

At ad level, add:

Individual ad performance
Creative elements if identifiable

What Each Metric Tells You

| Metric | Declining = | Growing = | |--------|-------------|-----------| | CPM | Cheaper reach | Competition up | | CPC / cost_per_result | More efficient | Less efficient | | CTR / LPV rate / conversion_rate | Problem | Healthy | | Results volume | Dropping | Growing | | Frequency | — | Fatigue risk if > 3 |

Understanding "Results"

The meaning of "results" depends on the optimization goal:

LEAD_GENERATION → results = leads
LINK_CLICKS → results = link clicks
LANDING_PAGE_VIEWS → results = landing page views
PURCHASE / CONVERSIONS → results = purchases

Always clarify what "results" means when discussing cost_per_result.

Campaign Structure Matters

CBO (campaign-level budget): Evaluate at campaign level first
ABO (adset-level budget): Evaluate at adset level
Multiple ads in one adset: Evaluate at adset level, not individual ads
Active but less than 7 days old + less than 50 results since last significant edit: Treat as still in learning phase — data is unstable
Breakdown Effect: Meta optimizes for marginal CPA (cost of the next result), not average CPA. A segment showing higher average CPA may be protecting overall campaign efficiency. Do not judge system decisions by average CPA in breakdown reports alone.

Step 4b: Lead Generation Campaigns

Use meta-ads-lead-gen-analysis instead — it has dedicated LPV benchmarks for lead gen, form friction analysis, CAPI verification, and lead quality diagnosis.

Step 4c: Landing Page Diagnostic

This section has moved to meta-ads-recommendation → Step 4. Landing page diagnosis is now part of the recommendation workflow for better action alignment.

Step 5: Check Ad Relevance Diagnostics

When an ad's cost_per_result is elevated and basic metrics (CTR, LPV rate) don't fully explain it, check the Ad Relevance Diagnostics in Meta Ads Manager.

| Diagnostic | What it measures | Low ranking suggests | |-------------|-------------------|-----------------------| | Quality Ranking | Perceived ad quality vs. competitors | Improve creative | | Engagement Rate Ranking | Expected engagement vs. competitors | Test new angles, improve hook | | Conversion Rate Ranking | Expected conversion vs. competitors with same optimization goal | Check landing page or audience-offer fit |

Usage rules:

Requires 500+ impressions to be available — below that, diagnostics are not meaningful
These are diagnostic tools only, not auction inputs
When all three rankings are low simultaneously → strong audience-creative mismatch

Step 6: Assess for Problems

Flag these when you see them:

Frequency > 3 → Audience fatigue risk
CTR less than 1% → Weak creative hook or audience mismatch
LPV rate less than 70% → Ad-to-landing-page disconnect
cost_per_result rising → Efficiency problem
Status is not ACTIVE → Explain impact
Learning / Learning Limited → Warn: data not yet stable
Spend less than 50% of budget → Delivery problem — audience may be too narrow or bid too low
One segment vastly outperforming others → System is correctly finding winners

Step 7: Structure Your Output

Campaign Context: one sentence covering objective, budget type, and status.
Summary: Spend, impressions, results, cost per result, CTR, CPM.
Funnel: Impressions → Link Clicks → Landing Page Views → Results, with rates at each stage.
Trend Analysis: Key metrics with direction and what it means for the business.
Key Issues: Specific problems and why they matter.
Data Notes: Any caveats — learning phase, insufficient data, zero-spend entities excluded.

Tone and Language

Match the language the user speaks (English or Chinese)
Speak as a professional growth partner — translate numbers into business insights
Every judgment must cite specific data. Never say "performance is bad" without a number
When data is insufficient, say so: "2 days isn't enough for reliable trends — let's revisit in a week"

Restrictions

Analysis only — never output recommendations in this skill
Do not recalculate metrics already computed by Meta (CTR, CPM, CPC are already calculated)
Never conclude an ad or adset is "underperforming" based on CPA alone without comparing to its own historical baseline and the account average

Session Context — What This Skill Writes

After completing analysis, store the following in session context:

| Key | Description | Example | |-----|-------------|---------| | funnel_weak_points | Where the biggest funnel drop-off occurs | "LPV rate 58%, well below 70% benchmark" | | trend_signals | Direction of key metrics | "CPM up 18% WoW; CTR down 0.4pp" | | anomalies | Anything unusual or unexpected | "Frequency 4.2 with CPL still at target" | | data_quality | Whether there's enough data to act | "Only 2 days — too early to judge" | | lp_diagnosis_general | Is the problem ad-side or landing page-side? | "Ad side — CTR declining, frequency stable" |

Routing: If the campaign is Lead Gen, route to meta-ads-lead-gen-analysis and preserve these keys. meta-ads-recommendation reads these keys to produce the action plan.