tdx Parent-Segment-Analysis - CDP Parent Segment Data Analysis

Workflow

1. Get output database schema
2. Read ALL columns from the schema file to understand available data
3. Identify relevant columns for the analysis (use sub-agent if schema is large)
4. Query customers or behavior tables

Get Output Database Schema

Parent segments generate an output database (cdp_audience_{id}) with:

• customers table - master + all attributes joined
• behavior_{name} tables - one per behavior

# Save schema to file (recommended for large schema data)
tdx ps desc "Customer 360" -o customer_360_schema.json

# Output summary:
# Schema saved to customer_360_schema.json
#   Database: cdp_audience_12345
#   Tables: 1 customers + 5 behaviors
#   Columns: 42 total

Schema JSON Structure

One column per line for grep and progressive disclosure:

{
  "database": "cdp_audience_12345",
  "parent_segment": "Customer 360",
  "parent_id": "12345",
  "customers": {
    "table": "customers",
    "columns": [
      { "name": "cdp_customer_id", "type": "varchar" },
      { "name": "email", "type": "varchar" }
    ]
  },
  "behaviors": [
    {
      "table": "behavior_purchases",
      "columns": [
        { "name": "cdp_customer_id", "type": "varchar" },
        { "name": "amount", "type": "double" }
      ]
    }
  ]
}

Discovering Relevant Columns

IMPORTANT: Always read the full schema file first to understand what data is available. After reviewing the schema, you can use grep/search to find specific columns.

For large schemas: use a sub-agent to discover relevant columns for your analysis goal

Analysis Guidelines

• Check null ratios for identified customer columns and exclude null values from aggregations to ensure accurate analysis.
• Build queries incrementally rather than attempting everything at once.
• Provide an extremely brief summary of key findings in one sentence.
• Always be very explicit and detailed about the data sources you are using, including table names, column names.

Array Columns

• For SQL queries needing array analysis/aggregation: Use CROSS JOIN UNNEST.
  sqlCopy

  SELECT
    tag,
    count(0) AS cnt
  FROM customers
  CROSS JOIN UNNEST(tags) AS t(tag)
  GROUP BY 1 ORDER BY 2 DESC LIMIT 10
  

• Don't use the same name for alias as column name (e.g., use tag not tags for the alias)
• Never use ARRAY_CONTAINS — use contains(array_column, value) or filter via CROSS JOIN UNNEST

Time Function Usage

Always first check the format of the time column before applying time functions.

Time functions:

• td_interval(time, '-1d') — filter records within a time interval
• td_time_parse('2024-01-01', 'UTC') — parse a time string into Unix timestamp (seconds)

Related Skills

• parent-segment - Manage parent segment configuration
• segment - Create child segments from parent segment data
• trino - Trino SQL syntax and TD-specific functions