TODO.md

TODO

• It may be overly expensive to scan ALL events looking for first/last occurances of user's event parameters. We can move data from 1st & last session into a new table and scan that table instead.
• mechanism to take in an array variable listing custom events and output 1 model per event (is this possible?)
• Add event timing (avg time to next page) metrics
• Session + conversion metrics
• Anything else to do with privacy_info field? Right now removing 'null' client ids from user dim tables.
• Create staging tables for the following events:
  • view_promotion
  • add_to_cart
  • Audience trigger events. See https://support.google.com/analytics/answer/9934109?hl=en
  • Special treatment for conversion events?
  • Full event reference:
    • https://developers.google.com/analytics/devguides/collection/ga4/reference/events
    • https://support.google.com/analytics/answer/9216061?hl=en&ref_topic=9756175
• Review these issues for ideas for our repo: https://github.com/coding-is-for-losers/ga4-bigquery-starter/issues
• Any special considerations for handling >1 data stream?
• Implement dev profile considerations to limit processing: https://docs.getdbt.com/docs/guides/best-practices#limit-the-data-processed-when-in-development
• Example of a funnel model https://github.com/teej/sf-funnels
• Review LookML examples for inspiration: https://github.com/llooker/ga_four_block_dev/tree/master/views/event_data_dimensions
  • Add landing page / exit page, session start/end time, session duration, is bounce, campaign source to dim_sessions model
• Configuration and dynamic templates to create custom event tables and dimensions
• Configuration to create custom dimensions (session, user, event_*) from event parameters
• Refactor 'user properties' functionality to pull from the user_properties field
• Support for large intraday tables (100+ shards). Currently they are unioned in as a view on top of partitioned base table. We could load in data up until yesterday into the partitioned table and then union in today's data.
• Allow users to configure certain event names as conversions. provide additional metrics around conversion events (conversion count per session, per user).
• Update dim_sessions to pull based on session key rather than session_start event. Looks like session_start no longer correclty identifies all sessions. See https://measure.slack.com/archives/C03AE85U5/p1658244315268039
• Merge and clean up dim_sessions & fct_sessions. Just consider it ga4__sessions and ga4__users.
• Use Fivetran's union_data method (or something similar) to handle multiple, unioned GA4 exports. https://github.com/fivetran/dbt_xero_source/blob/main/models/tmp/stg_xero__account_tmp.sql

Misc

• DBT guide to package creation: https://docs.getdbt.com/docs/guides/building-packages
• DBT project structure notes: https://discourse.getdbt.com/t/how-we-structure-our-dbt-projects/355

Discussion: Configuration to create custom dimensions

Product-scope (or item-scope in GA4) custom dimensions are a much missed feature.

We can implement them, with some difficulty, mapping event properties to the custom dimension. However, it is possible that Google, presuming they add item-scoped CDs, will just add the dimension to the items array which could result in stg_ga4__items.sql automatically picking up item-scoped CDs the way that it is currently written.