google_analytics.funnel_metrics
Derived metrics computed from landed funnels rows - currently task success rates, one per funnel.
Each row represents one metric’s value, for one funnel run, for one relying party (or the funnel’s total across all relying parties).
Keywords: Google Analytics, GA4, funnel, task success, task success rate, derived metric
Provenance
Like a funnel’s steps, a funnel’s derived metrics are declared in that funnel’s YAML on GitHub, in a metrics: block - see funnel_src/funnels/ in the canadalogin-reporting-data-pipeline repo. The YAML is the canonical definition of what each metric measures, including which two steps a ratio divides. A separate Lambda reads the already-landed rows from funnels (no new GA4 call) and computes every metric each funnel declares. See the task success rate cookbook entry for how the one metric type in use today (step_ratio) is calculated and interpreted.
- Updated: Daily, at 06:45 ET, after the 06:30 ET funnel export
- Retention: 7 years
- Steward: EDCP Data and Research Team
- Contact: Colin Douglas
- Location:
s3://<transformed-bucket>/google_analytics/funnel_metrics/
Fields
One row per funnel_id per funnel_version per window_days per window_end per metric_id per breakdown_value.
| Field Name | Type | Description |
|---|---|---|
| funnel_id | string | Partition key. Which funnel this metric was computed from (for example, “sign_in”, “sign_up”) |
| funnel_version | string | Partition key. The funnel definition’s version the metric was computed under. Typed string by the crawler - see the funnels Notes |
| window_days | string | Partition key. Length of the reporting window: “1” (nightly monitor), “7” (rolling weekly rate), or “28” (rolling four-week rate) |
| window_end | string | Partition key. Last date (inclusive) of the reporting window, as text |
| metric_id | string | Which metric this row is. Declared per-funnel on GitHub; currently always “task_success” for both funnels, but not a closed set - see Provenance |
| metric_type | string | The metric’s calculation shape, from the same YAML declaration. Currently always “step_ratio” (one step’s activeusers over another’s) |
| value | double | The computed metric value. null, not 0.0, when the denominator is zero (nobody entered, so the rate is undefined - not a false zero) |
| breakdown_dimension | string | The bare dimension the source funnel’s breakdown: block splits by (currently always “rp_name”) |
| breakdown_value | string | The relying party this row covers, or RESERVED_TOTAL for the funnel’s overall value - see funnels Notes |
| window_start | string | First date (inclusive) of the reporting window, as text |
| window_incomplete | boolean | true if the underlying funnel window was trimmed to valid_from/valid_to - exclude these rows from a trend, same rule as funnels |
Notes
This table deliberately does not carry the raw step values a ratio was computed from - only the result. The inputs are recoverable by joining back to funnels on the same (funnel_id, funnel_version, window_days, window_end, breakdown_value), if they’re needed.
Same versioning caveat as funnels, and the same fix applies: always pin funnel_version, or query funnel_metrics_current rather than this table directly.
A metric is recomputed automatically when its funnel’s funnel_version bumps (the same nightly job backfills every past window under the new version), but adding a metrics: block to a funnel for the first time also triggers a one-time backfill across that funnel’s entire already-landed history - so a metric’s earliest window_end is not necessarily the date the metric was defined.