alexe-ev Analyze Funnel, Retention & Cohorts
Purpose
Turn funnel, retention, and cohort data into product insights that identify where users drop off, churn, or get stuck.
Skill type
Conceptual skill with calculation-aware components
Use this skill when
- A funnel has unexplained drop-off at specific steps
- Retention is declining or unexpectedly flat
- Cohort analysis is needed to understand behavioral differences across user groups
- A product change may have affected retention and needs investigation
Do not use this skill when
- No data is available (collect data first)
- The goal is experiment result analysis (use analyze-experiment-results)
Required inputs
- Funnel data, retention data, or cohort data (at least one)
- Product context (what flow or user group is being analyzed)
Optional inputs
- Baseline benchmarks
- Cohort definitions
- Time range
- Segment breakdowns
Upstream context
Works best when:
- Metrics framework is defined
- Data instrumentation is in place
If upstream context is missing
If data is absent, produce a data collection plan and flag that analysis cannot proceed without it.
Downstream handoff
Output can feed:
- detect-performance-signals
- formulate-experiment-hypothesis (data insights become experiment inputs)
- identify-problem-opportunity
Instructions
- Map the funnel or retention structure being analyzed.
- Identify drop-off points or retention cliff events.
- Compare cohorts if relevant (by acquisition source, segment, time period).
- Identify patterns and anomalies.
- Generate hypotheses for the observed behavior.
- Recommend next steps: further analysis, qualitative research, or experiment.
Output
Provide:
- Funnel analysis with drop-off rates per step
- Retention curve interpretation
- Cohort comparison summary (if applicable)
- Key patterns and anomalies
- Hypotheses for observed behavior
- Recommended next steps
- Data quality caveats
Risks / caveats
- Correlation in cohort data is not causation — use it for hypothesis generation, not conclusions
- Segment data before drawing conclusions — average retention often hides meaningful variation
- Small cohorts produce noisy results — flag when sample sizes are thin