cohort-analysis

cohort-analysis performs cohort analysis on user-engagement data to reveal retention curves, feature-adoption trends, and segment-level insights. It solves the problem of understanding how groups of users behave over time — why some cohorts retain and others churn — by combining quantitative analysis with qualitative research recommendations. It triggers when analyzing retention by cohort, studying feature adoption over time, investigating churn patterns, or identifying engagement trends.

The skill follows a five-step workflow. It first reads and validates data from CSV, Excel, or JSON files, checking the structure (cohort identifier, time periods, engagement metrics), flagging missing values, and summarizing key statistics like cohort sizes and date ranges. It then generates quantitative analysis — retention rates, drop-off patterns, feature-adoption rates, and period-over-period changes — and can produce reusable Python scripts using pandas and numpy. Next it creates visualizations such as retention heatmaps (cohorts vs. time periods), progression line charts, and adoption comparison charts. It then identifies significant patterns (early churn, late-stage engagement shifts, adoption clusters, seasonal trends) against established baselines, and finally suggests follow-up research: targeted interviews with churning users, surveys of engaged cohorts, session replays, win/loss analysis, and A/B tests or feature experiments. Its key capabilities span data reading, retention analysis, cohort comparison, anomaly detection, Python script generation, visualization, research design, and statistical summary.

The skill targets product managers, analysts, and growth teams who want to move from raw engagement data to prioritized decisions. It works best with a clear time dimension and cohort definition, meaningful context about product changes, multiple metrics, and at least three to four cohorts, and it can output data summaries, quantitative findings, visualizations, 2–3 key insights, research recommendations, and optional analysis scripts.