analytics-tracking（分析跟踪）

Analytics Tracking - Implementation & Auditing

Category: Marketing Tags: GA4, Google Tag Manager, event tracking, conversion tracking, UTM, analytics audit, consent mode

Overview

Analytics Tracking is the implementation layer for marketing measurement. Bad tracking is worse than no tracking -- duplicate events, missing parameters, unconsented data, and broken conversions lead to decisions based on bad data. This skill covers building tracking right the first time and finding what is broken when it is not.

This skill handles implementation only. For analyzing campaign performance data, use campaign-analytics. For product analytics and in-app behavior, use the product-team skills.

Operating Modes

Mode 1: Build From Scratch

No analytics in place. Build the tracking plan, implement GA4 + GTM, define event taxonomy, configure conversions.

Mode 2: Audit Existing Tracking

Tracking exists but data cannot be trusted. Audit coverage, identify gaps, clean up duplicates, fix consent issues.

Mode 3: Debug Specific Issues

Events are missing, conversions do not match, GTM preview shows fires but GA4 does not record. Structured debugging workflow.

Event Taxonomy Design

Get this right before touching GA4 or GTM. Retrofitting taxonomy is painful and expensive.

Naming Convention

Format: object_action (snake_case, past tense verb)

Rules:

1. Always noun_verb order, never verb_noun
2. Snake_case only -- no camelCase, no hyphens, no PascalCase
3. Past tense verbs: _started, _completed, _failed, _viewed
4. Specific enough to be unambiguous, not so verbose it is a sentence
5. Prefix with domain when needed: onboarding_step_completed, billing_plan_selected

Standard Event Parameters

Every custom event should include applicable parameters from this table:

SaaS Event Taxonomy (Reference)

Core Funnel:

visitor_arrived              (automatic page_view in GA4)
signup_started               (user clicked "Sign up")
signup_completed             (account created)
trial_started                (free trial began)
onboarding_step_completed    (params: step_name, step_number)
feature_activated            (params: feature_name)
plan_selected                (params: plan_name, billing_period)
checkout_started             (params: value, currency, plan_name)
checkout_completed           (params: value, currency, transaction_id)
subscription_renewed         (params: value, plan_name)
subscription_cancelled       (params: cancel_reason, plan_name)

Micro-Conversions:

pricing_viewed
demo_requested               (params: source)
form_submitted               (params: form_name, form_location)
content_downloaded           (params: content_name, content_type)
video_started                (params: video_title)
video_completed              (params: video_title, percent_watched)
chat_opened
help_article_viewed          (params: article_name)
invite_sent                  (params: recipient_role)
integration_connected        (params: integration_name)

GA4 Configuration

Data Stream Setup

1. Create property: GA4 Admin > Properties > Create
2. Add web data stream with your domain
3. Enhanced Measurement -- review each:

- Page views: Keep enabled - Scrolls: Keep enabled - Outbound clicks: Keep enabled - Site search: Enable if you have search - Video engagement: Disable if tracking videos manually (avoids duplicates) - File downloads: Disable if tracking via GTM (for better parameters)

1. Configure domains: add all subdomains in your funnel
2. Data retention: Set to 14 months (maximum for free GA4)

Conversion Events

Mark as conversions in GA4 Admin > Conversions:

• signup_completed
• checkout_completed
• demo_requested
• trial_started

Rules:

• Maximum 30 conversion events per property -- curate carefully
• GA4 conversions are retroactive for 6 months when enabled
• Do not mark micro-conversions as conversions unless optimizing ad campaigns for them
• Conversion counting: set to "once per session" for lead events, "every" for purchase events

Custom Dimensions

Register custom dimensions for any event parameter you want to filter/segment by:

Register in GA4 Admin > Custom definitions > Create custom dimension.

Google Tag Manager Implementation

Container Architecture

GTM Container
├── Tags
│   ├── GA4 Configuration (All Pages trigger)
│   ├── GA4 Event Tags (one per custom event)
│   ├── Google Ads Conversion Tags (per conversion action)
│   └── Meta Pixel / LinkedIn Insight (if running ads)
├── Triggers
│   ├── All Pages (Page View)
│   ├── DOM Ready
│   ├── Custom Event triggers (one per dataLayer event)
│   └── Element Click triggers (CSS selector based)
└── Variables
    ├── Data Layer Variables (one per dataLayer key)
    ├── Constants (GA4 Measurement ID, etc.)
    └── Lookup Tables (if needed for mapping)

Implementation Pattern: Data Layer Push

Your application pushes events to the data layer. GTM picks them up and sends to GA4.

Application code:

// Push event when user completes signup
window.dataLayer = window.dataLayer || [];
window.dataLayer.push({
  event: 'signup_completed',
  method: 'email',
  user_id: userId,
  plan_name: 'trial'
});

GTM configuration:

Trigger:
  Type: Custom Event
  Event name: signup_completed

Tag:
  Type: GA4 Event
  Event name: signup_completed
  Parameters:
    method:    {{DLV - method}}
    user_id:   {{DLV - user_id}}
    plan_name: {{DLV - plan_name}}

SPA Handling

Single Page Applications need special attention because page views do not fire automatically on route changes.

Option A: History change trigger (GTM built-in)

• Enable "History Change" trigger in GTM
• Fires GA4 page_view on every pushState/popState

Option B: DataLayer push on route change (more control)

// In your router (React Router, Next.js, etc.)
router.events.on('routeChangeComplete', (url) => {
  window.dataLayer.push({
    event: 'page_view',
    page_location: url,
    page_title: document.title
  });
});

Conversion Tracking: Ad Platforms

Google Ads

Recommended approach: Import GA4 conversions into Google Ads (single source of truth).

1. Link GA4 and Google Ads accounts
2. In Google Ads > Goals > Conversions > Import > Google Analytics
3. Select GA4 conversion events to import
4. Set attribution model: Data-driven (if 50+ conversions/month), otherwise Last-click
5. Conversion window: 30 days for lead gen, 90 days for high-consideration B2B

Enhanced Conversions: Enable for 15-30% better conversion measurement. Sends hashed first-party data (email, phone) to match conversions that cookies miss.

Meta (Facebook/Instagram)

1. Install Meta Pixel base code via GTM
2. Configure standard events: PageView, Lead, CompleteRegistration, Purchase
3. Conversions API (CAPI): strongly recommended -- client-side pixel loses approximately 30% of conversions due to ad blockers and iOS App Tracking Transparency
4. Deduplication: when using both pixel and CAPI, send the same event_id to prevent double-counting

LinkedIn Insight Tag

1. Install via GTM (Tag type: LinkedIn Insight)
2. Configure conversion events in LinkedIn Campaign Manager
3. Match events to your taxonomy: signup_completed -> LinkedIn "Sign-up" conversion

UTM Strategy

Convention Enforcement

Critical rules:

• Never tag organic traffic with UTMs (overrides GA4 automatic attribution)
• Never tag direct/internal links with UTMs
• Use a UTM builder spreadsheet or tool -- manual entry causes inconsistency
• Lowercase everything -- Google and google are different sources in GA4

Attribution Windows

Cross-Domain Tracking

For funnels crossing domains (e.g., acme.com to app.acme.com):

1. GA4 Admin > Data Streams > Configure tag settings > Configure your domains > Add both domains
2. GTM: GA4 Configuration tag > Fields to Set > linker > Add domains
3. Admin > Data Streams > List unwanted referrals > Add both domains

Verification: Visit domain A, click link to domain B, check GA4 DebugView. The session should NOT restart. If a new session starts, cross-domain tracking is broken.

Consent Management

Consent Mode v2

Required for EU compliance and for maintaining data quality in consent-heavy markets.

Recommendation: Implement Advanced Consent Mode v2 via GTM with a CMP (Cookiebot, OneTrust, Usercentrics).

Expected consent rates by region:

• EU/EEA: 60-75%
• UK: 70-80%
• US: 85-95%
• Rest of world: 80-90%

Implementation via GTM

1. Install CMP tag (fires first, before any other tags)
2. Set default consent state:
   - analytics_storage: denied
   - ad_storage: denied
   - ad_user_data: denied
   - ad_personalization: denied
3. CMP updates consent state on user choice
4. GA4 and ad tags respect consent automatically

Data Quality Auditing

Audit Checklist

Event Quality:

• No duplicate events (check GTM Preview for double-fires)
• All custom events have required parameters
• Event names follow naming convention
• No PII in event parameters (names, emails, phone numbers)
• Enhanced Measurement not duplicating GTM custom events

Configuration Quality:

• Data retention set to 14 months
• Internal traffic filter enabled (office and developer IPs)
• Bot filtering enabled (default in GA4)
• Cross-domain tracking working (if applicable)
• Custom dimensions registered for filtered parameters
• Conversion events marked correctly

Consent Quality:

• Consent Mode v2 implemented (if serving EU users)
• CMP banner appearing on first visit
• Tags respect consent state (no firing before consent)
• Consent state persisting across pages

Common Data Quality Issues

Debugging Workflow

Step 1: Open GTM Preview mode
  - Is the tag firing? Check triggers and conditions
  - Is the data layer populated? Check dataLayer in console

Step 2: Check GA4 DebugView (Admin > DebugView)
  - Is the event appearing? If yes, GTM is working
  - Are parameters populated? Check parameter values

Step 3: Check GA4 Realtime report
  - Events appearing with 5-minute delay? Normal
  - Events not appearing at all? Check measurement ID

Step 4: Check Network tab (DevTools)
  - Filter by "collect" or "analytics"
  - Is the request being sent? Check status code
  - Is the request being blocked? Check ad blockers / consent

Proactive Triggers

Surface these findings without being asked:

• Events firing on every page load with identical parameters: misconfigured trigger causing data inflation
• No user_id parameter on authenticated events: cannot connect analytics to CRM or understand cohorts
• GA4 conversion count differs from Google Ads by more than 15%: attribution window or deduplication issue
• No consent mode in EU markets: legal exposure and 25-40% data underreporting
• All pages showing as /(not set): SPA routing not handled properly
• utm_source showing as direct for known paid campaigns: UTMs missing or being stripped by redirects

Related Skills

Troubleshooting

Success Criteria

• All custom events follow consistent noun_verb snake_case naming convention with zero violations in schema audit
• GA4 conversion counts match ad platform conversion counts within 10% variance
• Server-side tracking recovers 20%+ of previously lost attribution data within 90 days of deployment
• UTM parameter validation passes 100% on all active campaigns (no mixed case, no spaces, no missing required params)
• Consent Mode v2 limits EU data loss to under 15% via behavioral modeling
• Event parameters contain zero PII violations as verified by automated schema checker
• Data retention set to 14 months, internal traffic filtered, and cross-domain tracking verified

Scope & Limitations

In Scope: GA4 configuration, GTM implementation, event taxonomy design, conversion tracking setup, UTM strategy, consent management, data quality auditing, server-side tagging architecture, cross-domain tracking, ad platform conversion integration (Google Ads, Meta, LinkedIn).

Out of Scope: Product analytics platforms (Amplitude, Mixpanel), data warehouse configuration, custom ETL pipelines, mobile app tracking (Firebase), marketing attribution modeling (see marketing-analyst skill), A/B test statistical analysis (see ab-test-setup skill).

Limitations: Server-side tracking requires a cloud-hosted GTM container (GCP, AWS, or third-party) with associated infrastructure costs. Privacy-first analytics with Consent Mode v2 produces modeled data for non-consented users — modeled data has 5-15% variance from actual. This skill does not make LLM or API calls; all validation is deterministic.

Scripts