| name | analyze-metrics |
| description | Analyze product metrics and identify trends when the user asks to review metrics, analyze KPIs, or assess product health |
| author | chalk |
| version | 1.0.0 |
| metadata-version | 3 |
| allowed-tools | Read, Glob, Grep, Write |
| argument-hint | [metric name, period, or metrics file] |
| read-only | false |
| destructive | false |
| idempotent | false |
| open-world | false |
| user-invocable | true |
| tags | analysis, metrics, data |
Analyze Metrics
Overview
Review product metrics against targets, identify trends across cohorts, distinguish leading from lagging indicators, and generate hypotheses for unexpected changes. Turns raw numbers into actionable insight.
Workflow
-
Read metrics context — Scan .chalk/docs/product/ for any metrics framework, KPI definitions, or previous metrics reviews. Identify which metrics have defined targets and baselines.
-
Gather metrics data — Parse $ARGUMENTS for the specific metrics or period to analyze. If the user provides data inline or references a file, read it. If no data is provided, ask the user to supply current metric values.
-
Classify each metric — For each metric, determine:
- Type: leading (predictive) vs. lagging (outcome)
- Category: acquisition, activation, engagement, retention, revenue, referral
- Comparison basis: target value, previous period, cohort benchmark
-
Assess current vs. target — Compare each metric's current value against its target. Classify as: on-track (within 10%), at-risk (10-25% off), or off-track (>25% off). If no target exists, note the gap.
-
Identify trends — For each metric with historical data, classify the trend: improving, stable, or declining. Note acceleration or deceleration (is improvement slowing down?). Flag inflection points.
-
Cohort comparison — Where cohort data is available, compare across user segments (new vs. returning, plan tiers, acquisition channels). Identify cohorts that outperform or underperform the average.
-
Generate hypotheses — For any metric that is off-track or shows unexpected changes, propose 2-3 hypotheses for the cause. Each hypothesis should be testable. Connect to recent product changes, market events, or seasonal patterns.
-
Identify metric relationships — Flag leading indicators that predict lagging indicator changes. Note correlations and potential causal chains.
-
Determine the next file number — Read filenames in .chalk/docs/product/ to find the highest numbered file. Use highest + 1.
-
Write the review — Save to .chalk/docs/product/<n>_metrics_review_<period>.md.
Output
- File:
.chalk/docs/product/<n>_metrics_review_<period>.md
- Format: Markdown with a health dashboard table and detailed metric sections
- Key sections: Health Summary Table (metric / current / target / status / trend), Detailed Analysis per metric, Cohort Insights, Hypotheses for Off-Track Metrics, Recommended Actions
Anti-patterns
- Vanity metrics without context — Reporting "10K signups" without conversion rate, activation rate, or retention is misleading. Always pair volume metrics with quality metrics.
- Confusing correlation with causation — "We launched feature X and signups went up" is a correlation, not a causal claim. Always note confounders and suggest experiments to validate.
- Ignoring leading indicators — Only reviewing lagging indicators (revenue, churn) means you are looking in the rearview mirror. Prioritize leading indicators that let you act before outcomes materialize.
- Reporting without hypotheses — Stating "retention dropped 5%" without proposing why is not analysis. Every unexpected change needs at least one testable hypothesis.
- Missing cohort breakdowns — Aggregate metrics hide important variation. A stable overall retention rate can mask declining retention in new cohorts offset by strong retention in old cohorts.