modeling-product-usage-metrics

Group each person's activity by week, take their first active week as the cohort, then count distinct users active in each subsequent week. The Skill provides ready HogQL and dbt recipes computing retained counts and retention rates per cohort week and weeks-later offset.

Retention measures whether users return after their first interval as a cohort matrix. Stickiness measures how many distinct intervals users were active, surfacing power users. Lifecycle classifies users per interval as new, returning, resurrecting, or dormant to judge growth quality.

Yes, the Skill includes dbt recipes for fct_retention, fct_stickiness, and fct_lifecycle marts built on a stg_events staging model, plus a schema.yml with uniqueness and not-null data tests. The retention model uses dbt's cross-database datediff macro.

Count distinct active days per person over a trailing window such as 30 days, then bucket users by that count. The stickiness recipes produce this distribution directly, and the DAU/WAU or DAU/MAU ratio is a scalar summary of the same idea.

The lifecycle query computes first activity within the analysis window only, so users active just before it appear as new at the left boundary. Extend the window one interval earlier if that boundary classification matters for your analysis.

Use the native retention, stickiness, or lifecycle insights for quick interactive analysis. Build HogQL views or dbt marts when the metric must be reused, joined with other models, or consumed downstream, such as by activation-metric validation.