---

name: analytics-design description: Design data analysis from purpose clarification to visualization. Use when analyzing data, exploring BigQuery schemas, building queries, or creating Looker Studio reports.

Analytics Design

Workflow

Use references/analytics-design-template.md to document every analysis.

1. Clarify Purpose: What do you want to know? Why is this analysis needed? Who will use it? One-time or ongoing monitoring?

2. Discover Data: Explore available datasets and understand schema.

   • Ask user for project/dataset context and business background
   • bq ls, bq show --schema for BigQuery tables
   • db/schema.rb for Rails projects
   • API docs or sample data for external services
   • Understand table relationships (ER diagrams help)

3. Build Query: Write SQL based on discovered schema.

   • Use CTEs for readability
   • Execute using BigQuery CLI (see BigQuery Query Execution below)
   • Interpret results and document findings

4. Create Dashboard (if ongoing monitoring needed):

   • Use references/looker-studio-template.md to design
   • Define decisions: What actions will users take based on this dashboard?
   • Check existing resources: Similar dashboards or queries already exist?
   • Align time granularity with usage frequency (daily/weekly/monthly)
   • Design data sources, pages, and charts

BigQuery Query Execution

Prerequisites

Check gcloud configuration before running queries:

Copygcloud config get-value project

• If authentication error: prompt user to run gcloud auth login, then resume
• If project unset: prompt user to run gcloud config set project <PROJECT_ID>

Execution Process

1. Dry run: Validate syntax and estimate cost

   Copybq query --use_legacy_sql=false --dry_run "SELECT * FROM \`project.dataset.table\`"
   

   Cost: ~$5/TB. <1GB is light, 2GB+ needs optimization.

2. Execute: Run and confirm results

   Copybq query --use_legacy_sql=false --format=csv "SELECT * FROM \`project.dataset.table\`"
   

Always use fully-qualified table names: project.dataset.table

Query Design Tips

• Specify exact date ranges
• Filter partitioned tables by partition key
• Avoid correlated subqueries (use JOINs/CTEs)
• Filter early with CTEs before joining large tables

Looker Studio Best Practices

Data Source Design

• One data source per analytical purpose
• Pre-aggregate in SQL for performance
• Include bucket fields for distribution analysis
• Include sort-order fields for proper chart ordering
• Descriptive data source names

Report Structure

• Separate pages by time granularity (daily/monthly)
• Group related metrics per page
• Consistent filter scopes within pages

Chart Type Selection