Cohort Retention: Retention tracked by signup cohort over time. The canonical formula reports week 8 for the period's signup cohort.
Formula
Cohort Retention (Week N) = Active users in week N / Users in cohort × 100Computable definition
The canonical formula in the GASP formula grammar. The MCP server compiles it to SQL over the datasets listed below.
all_time(count_distinct(field.user_id where field.product_event_type = 'qualifying' and in_period(field.signup_date) and (field.event_date - field.signup_date) >= 49 and (field.event_date - field.signup_date) < 56)) / count_distinct(field.user_id where in_period(field.signup_date))Unit: percent (the formula yields an unscaled ratio)
Notes: Week 8 retention for the signup cohort of the period: users who signed up in the period and had qualifying activity 49 to 55 days after signup, over users who signed up in the period. The numerator reads all time because week 8 falls after period_end.
Data requirements
Grain: Monthly cohort · Frequency: Daily · Aggregation: cohort (month) · Staged: no · Main source: Product Analytics.
| Field | Dataset | Type |
|---|---|---|
event_date | product_event | date |
product_event_type | product_event | string |
signup_date | product_event | date |
signup_date | user | date |
snapshot_date | user | date |
user_id | product_event | string |
user_id | user | string |
Example SQL (Postgres)
SELECT (CAST((SELECT COUNT(DISTINCT user_id) FROM product_event WHERE (((product_event_type = 'qualifying' AND (signup_date >= :period_start AND signup_date <= :period_end)) AND (event_date - signup_date) >= 49) AND (event_date - signup_date) < 56)) AS DOUBLE PRECISION) / NULLIF((SELECT COUNT(DISTINCT user_id) FROM user WHERE snapshot_date = :period_end AND (signup_date >= :period_start AND signup_date <= :period_end)), 0)) AS cohort_retention From an agent: with the GASP MCP server, call
lookup_metric({ "name": "Cohort Retention" }) or generate_query({ "metric": "Cohort Retention", "adapter": "<your adapter JSON>" }).
What It Tells You
How retention evolves and whether it is improving for newer cohorts.