Skip to main content

google_analytics.funnels

Step-by-step counts for each defined GA4 funnel.

Step-by-step counts for each defined GA4 funnel (currently sign_in and sign_up), for a given reporting window.

Each row represents one step of one funnel run, for one relying party (or the funnel’s total across all relying parties).

Keywords: Google Analytics, GA4, funnel, funnel step, task success, drop-off, abandonment, relying party

Provenance

A funnel is defined as YAML, not SQL: one file per funnel in funnel_src/funnels/ in the canadalogin-reporting-data-pipeline repo. A dedicated Lambda turns each funnel’s YAML into a GA4 Data API runFunnelReport request (via Workload Identity Federation, same auth pattern as property_traffic) and writes the result as JSONL. The YAML on GitHub is the canonical definition of what each funnel measures - which pages or events count as which step, in what order, and over what population. This page describes the table shape; it does not enumerate step names, because they are versioned in that YAML and can change between funnel versions.

  • Updated: Daily, at 06:30 ET (a nightly sweep; see Notes)
  • Retention: 7 years
  • Steward: EDCP Data and Research Team
  • Contact: Colin Douglas
  • Location: s3://<transformed-bucket>/google_analytics/funnels/

Fields

One row per funnel_id per funnel_version per window_days per window_end per funnelstepname per breakdown_value.

Field Name Type Description
funnel_id string Partition key. Which funnel this row belongs to (for example, “sign_in”, “sign_up”) - matches a filename under funnel_src/funnels/ on GitHub
funnel_version string Partition key. The funnel definition’s version at the time this row was run. A plain integer, but typed string by the crawler - see Notes
window_days string Partition key. Length of the reporting window this row summarizes: “1” for a nightly monitoring run, “7” for a rolling weekly rate, “28” for a rolling four-week rate
window_end string Partition key. Last date (inclusive) of the reporting window, as text (for example, “2026-07-23”)
funnelstepnumber bigint The step’s 1-indexed position in the funnel
funnelstepname string The step’s name, as declared in the funnel’s YAML on GitHub. Not a closed set - see Provenance
breakdown_dimension string The bare GA4 dimension a funnel’s breakdown: block splits by (currently always “rp_name”, the relying party)
breakdown_value string The relying party this row covers, or the sentinel RESERVED_TOTAL for the funnel’s grand total across all relying parties (including any past GA4’s 15-value breakdown cap) - see Notes
activeusers bigint Distinct active users who reached this step, for this relying party, within the window
funnelstepcompletionrate double GA4’s own step-over-step completion rate (this step’s users over the previous step’s)
funnelstepabandonments bigint Users who reached this step but did not reach the next one
funnelstepabandonmentrate double funnelstepabandonments as a proportion of this step’s activeusers
window_start string First date (inclusive) of the reporting window, as text. Derivable from window_end and window_days, but stored for convenience
window_incomplete boolean true if this window was trimmed to the funnel’s valid_from/valid_to bounds (see Notes) - exclude these rows from any trend read as a full window

Notes

A funnel’s step names, order and population can change between versions - always check GitHub, not a memory of a prior version. Every CanadaLogin funnel has already had at least one step renamed or re-scoped since it was first defined (see the task success rate metric for a worked example). funnel_version exists so old and new definitions coexist rather than silently overwriting each other: always pin funnel_version in a query, or use the funnels_current view, which resolves the newest version per date automatically. Summing or averaging across versions blends two different definitions of the funnel into one meaningless number.

RESERVED_TOTAL is a real GA4 sentinel, not a relying party. A funnel breakdown caps at 15 distinct values; RESERVED_TOTAL is GA4’s grand total across all relying parties, including any beyond that cap, and is the correct number to read as the funnel’s overall rate.

window_incomplete rows are written, not dropped, and must be excluded from a trend. A window that reaches earlier than a funnel’s valid_from (or later than valid_to) is trimmed and flagged rather than skipped, so a scheduled job keeps running as a tagging boundary recedes.

activeusers is de-duplicated within its own window and is not additive across windows. Summing seven daily (window_days = "1") rows for a week, for example, is not the same query as one window_days = "7" row for that week - repeat users within the week are counted once by the 7-day row and once per day they appeared by the daily rows. Never sum activeusers across rows with different window_end or window_days values.