GA4 BigQuery Skills

A collection of Agent Skills that give AI agents the ability to write correct, performant, and cost-efficient BigQuery SQL for Google Analytics 4 data — from everyday reporting queries to ML-ready dataset creation.

Skills

ga4-bigquery-query — Reporting & Analysis

Covers everything needed to query GA4 event data in BigQuery:

• Schema & table structure — GA4 export format, all top-level fields, nested/repeated record patterns
• Event parameter extraction — 3 UNNEST patterns, value type reference, user property propagation
• Users & sessions — Session key construction, all core user/session metric calculations
• Traffic sources (all 4 scopes) — traffic_source, session_traffic_source_last_click, collected_traffic_source, event params — plus full default channel grouping CASE logic
• Ecommerce — Transaction & item queries, funnel analysis, market basket analysis, revenue by source
• Page, event & date/time dimensions — Landing/exit page, page path levels, date formatting, device/geo
• Attribution models — Last-touch, first-touch, linear, position-based (40-20-40), time decay (7-day half-life)
• Advanced patterns — Cohort revenue, path analysis, checkout abandonment recovery, retention analysis
• Cost optimization — _table_suffix patterns, column selection, materialization strategy
• Privacy & consent — Consent mode impact on data quality, consent-aware query templates
• BigQuery SQL tips — SAFE functions, MAX_BY, QUALIFY, PIVOT, ARRAY_AGG, named windows, REGEXP

ga4-bigquery-ml-query — ML Dataset Creation

Covers building ML-ready datasets from GA4 data in BigQuery:

• Dataset architecture — Lookback/lookahead temporal windows, entity pools, snapshot dates, leakage prevention
• Feature engineering — Cumulative metrics, recency signals, site content features, traffic source and geo encoding
• Training vs inference modes — Stored procedure pattern with MODE parameter, deterministic train/val/test splits
• GA4-specific examples — Full propensity and predictive LTV query templates with 5-layer CTE architecture

Skill Structure

ga4-bigquery-query/
├── SKILL.md                              # Main skill file (loaded by the agent)
└── references/                           # Detailed reference files (loaded on demand)
    ├── schema-and-tables.md
    ├── unnesting-patterns.md
    ├── users-and-sessions.md
    ├── traffic-sources.md
    ├── ecommerce.md
    ├── page-event-dimensions.md
    ├── attribution-models.md
    ├── advanced-patterns.md
    ├── cost-optimization.md
    ├── privacy-and-consent.md
    └── sql-tips.md

ga4-bigquery-ml-query/
├── SKILL.md                              # Main skill file (loaded by the agent)
└── references/                           # Detailed reference files (loaded on demand)
    ├── dataset-architecture.md
    ├── feature-engineering.md
    ├── training-and-inference.md
    └── ga4-examples.md

Installation

Copy one or both skill folders into a supported skill location for your agent host. Install only the skills you need.

Project-level (shared with your team via source control)

Copy into any of these directories at your repository root:

.github/skills/ga4-bigquery-query/
.github/skills/ga4-bigquery-ml-query/
.agents/skills/
.claude/skills/

Example:

cp -r ga4-bigquery-query/ /path/to/your-project/.github/skills/ga4-bigquery-query/
cp -r ga4-bigquery-ml-query/ /path/to/your-project/.github/skills/ga4-bigquery-ml-query/

Personal (available across all your workspaces)

Copy into any of these directories in your home folder:

~/.copilot/skills/
~/.agents/skills/
~/.claude/skills/

Example:

cp -r ga4-bigquery-query/ ~/.copilot/skills/ga4-bigquery-query/
cp -r ga4-bigquery-ml-query/ ~/.copilot/skills/ga4-bigquery-ml-query/

Usage

Reporting queries (ga4-bigquery-query)

Once installed, ask your agent GA4 BigQuery questions such as:

Write a query to get sessions by source/medium for the last 30 days

For runnable SQL, the agent needs your BigQuery project ID and either:

• your GA4 dataset name, or
• your GA4 property ID

If only the property ID is known, the dataset is usually analytics_<property_id>. Documentation examples use {project} and {dataset} as placeholders, but the agent should replace them automatically when enough context is available and ask for the missing identifier if it is not.

ML dataset creation (ga4-bigquery-ml-query)

Ask your agent to build ML-ready datasets, for example:

Build a propensity-to-purchase dataset using GA4 data with a 30-day lookback and 14-day label window

Before generating a query, the agent will confirm your ML objective, observation grain, label definition, and dataset identifiers. The output is a BigQuery stored procedure that supports both TRAINING and INFERENCE modes.

How It Works

Agent Skills use progressive loading:

1. Discovery — The agent reads the skill name and description from SKILL.md frontmatter
2. Instructions — When relevant, the agent loads the SKILL.md body (~200 lines of core patterns)
3. References — As the agent works, it loads specific reference files on demand (e.g., only the ecommerce reference when you ask about revenue queries)

This means the skills stay efficient — they don't flood the context window with all reference files at once.

Compatibility

This skill follows the open Agent Skills standard.