engagement-analytics-skill（参与度分析技能）

Engagement Analytics Tracker Skill

A comprehensive skill for designing, implementing, and interpreting behavioral analytics across four touchpoint layers: website, email, social, and mobile app.

Four Tracking Modules

Load strategy: Load only the relevant module(s) based on the user's question. For full analytics stack questions ("build me a complete analytics system"), load all four.

Universal Data Principles

These apply across ALL four modules:

Event Naming Convention (Use Everywhere)

object_action
# Examples:
page_viewed          button_clicked       form_abandoned
video_played         product_viewed       email_opened
session_started      feature_used         purchase_completed

• Always lowercase with underscores
• Object first, then action
• Be specific: checkout_form_abandoned not form_event
• Keep consistent across all platforms — the same action has the same name everywhere

Data Layer Structure (Web)

window.dataLayer = window.dataLayer || [];
dataLayer.push({
  event: 'event_name',           // string — always required
  user_id: 'u_abc123',           // hashed or anonymized
  session_id: 'ses_xyz',
  timestamp: new Date().toISOString(),
  page_path: window.location.pathname,
  // event-specific properties below:
  element_id: 'hero_cta',
  element_text: 'Start Free Trial',
});

Engagement Scoring Formula

A composite score usable across web, email, and app:

Engagement Score = 
  (Sessions × 1) +
  (Pages per session × 2) +
  (Scroll 75%+ events × 3) +
  (CTA clicks × 5) +
  (Email opens × 2) +
  (Email clicks × 5) +
  (App sessions × 3) +
  (Feature completions × 8) +
  (Conversions × 20)

Score tiers:
  0–20:   Cold (re-engagement candidate)
  21–50:  Warming (nurture sequence)
  51–100: Engaged (sales-ready consideration)
  100+:   High Value (priority outreach)

Adjust weights based on business model. Recalculate weekly per user.

Privacy & Compliance Baseline

• Never collect raw PII in event properties — hash emails/IDs before sending to any platform
• Implement consent gating: fire tracking tags only after user consents (GDPR)
• Use server-side tagging (GTM Server-Side) for sensitive data flows
• Respect Do Not Track headers and browser privacy modes
• Apple ATT opt-in required for IDFA on iOS — design attribution without assuming access
• CCPA: provide opt-out mechanism; do not sell behavioral data without consent

Quick Implementation Checklist

New Analytics Setup

• [ ] Define tracking plan: events, properties, naming convention — before touching any tool
• [ ] Set up GTM container (web) or SDK (mobile)
• [ ] Implement dataLayer or SDK event calls
• [ ] Configure GA4 or destination analytics platform
• [ ] Validate all events in debug/preview mode before going live
• [ ] Set up consent management (CMP) gating
• [ ] Create dashboards for key metrics
• [ ] Schedule regular data quality audits

Existing Analytics Audit

• [ ] Are events named consistently? (check for duplicates with different names)
• [ ] Is user_id passed and consistent across sessions and platforms?
• [ ] Are conversion events firing correctly? (test end-to-end)
• [ ] Is there data loss from consent mode, ad blockers, or iOS ATT?
• [ ] Are email UTM parameters correctly attributed in GA4?
• [ ] Are mobile sessions merging correctly with web sessions (cross-device)?

Cross-Channel Attribution Model

When a user touches multiple channels before converting:

Journey: Paid Ad → Email Click → Direct Visit → Converted

Attribution options:
  Last-click:     Direct gets 100% credit (most common, least accurate)
  First-click:    Paid Ad gets 100% credit
  Linear:         All 3 channels get 33% each
  Time-decay:     Direct > Email > Paid Ad (recency-weighted)
  Data-driven:    ML model (GA4 DDA) — most accurate, needs volume

Recommended: Use GA4 Data-Driven Attribution (DDA) when you have 500+ conversions/month. Below that volume, use Linear to avoid bias toward any single channel.

Track cross-channel with UTM parameters on all non-direct traffic:

?utm_source=klaviyo&utm_medium=email&utm_campaign=may_reengagement&utm_content=cta_button

Output Templates

Event Schema Definition

Event Name: [object_action]
Trigger: [when exactly does this fire?]
Properties:
  - property_name (type): description, example value
  - property_name (type): ...
Platform: [GTM / Firebase / Klaviyo / etc.]
Destination: [GA4 / BigQuery / Amplitude / etc.]
Privacy: [PII risk? How handled?]

Analytics Health Report

DATE: [date]
COVERAGE: [% of key user actions being tracked]
DATA QUALITY: [issues found — missing events, duplicates, naming inconsistencies]
TOP INSIGHTS THIS PERIOD: [what the data shows]
ACTION ITEMS: [what to fix or investigate]