bosskuai-analytics-metrics git:20261012.17c6803A

Immutable. This exact content never changes and is served at /api/v1/blob/965a451a78bc1f1b.

---
name: bosskuai-analytics-metrics
description: "Use when working out what to measure: instrumentation strategy, event design, funnels, experimentation, north-star metrics, KPI definitions, attribution logic, and whether the numbers support product decisions."
---

# BosskuAI Analytics / Metrics

Use this skill when the main question is **what to measure, how to instrument it, and whether the numbers will support good decisions**.

## How this differs from nearby skills

- **`bosskuai-product-strategy`**: defines product bets and hypotheses; this skill defines how those bets will be measured.
- **`bosskuai-marketing-growth`**: optimizes acquisition and distribution; this skill designs measurement across product and growth surfaces.
- **`bosskuai-engineering-delivery`**: implements tracking; this skill decides the event model, funnel definitions, and metric semantics first.
- **`analytics`**: GA4/GTM/UTM implementation, tracking plans and "are my events firing" debugging; this skill decides what to measure and whether the numbers can be trusted. **`ab-testing`** / **`bosskuai-growth-experiment`**: experiment design; this skill owns exposure events and metric semantics.

## Mindset

- If a metric can’t drive a decision, it is probably vanity.
- Event definitions are contracts; sloppy naming creates years of bad dashboards.
- Funnel stages must map to real user behavior, not wishful narratives.
- Instrumentation should be privacy-conscious and operationally debuggable.

## Measurement lenses

**Outcome alignment**
- What decision or hypothesis will this metric support?
- Is there a north-star or guardrail metric?
- What would success or failure look like?

**Event model**
- Are events named for user actions or system facts clearly?
- Is event grain explicit?
- Are properties stable, typed, and necessary?

**Funnels and experiments**
- Are stages mutually understandable and time-bounded?
- Can drop-off be explained by available events?
- Are experiment variants and exposure events explicit?

**Retention and lifecycle**
- Are D1/D7/D30 retention curves defined and baselined?
- Are users segmented by acquisition cohort, so retention is compared on an apples-to-apples basis?
- Is the distinction between resurrection (returning after churn) and retention (never churned) explicit?
- Is churn defined precisely: no activity for N days? Explicit cancellation? Both?

**Attribution and identity**
- How are users, sessions, orgs, devices, and anonymous states linked?
- Where can attribution be lost or duplicated?
- Are cross-platform identities reconciled clearly?
- GA4 and Google Ads offer only data-driven and last-click models since Nov 2023; first-touch or multi-touch views need a warehouse or CDP.

**Data quality**
- Are dedupe, sampling, clock skew, and late events handled?
- Is there a way to verify the numbers against source actions?
- What monitoring catches tracking drift?

**Privacy and retention**
- Is sensitive data unnecessarily captured?
- Are consent and retention requirements reflected in instrumentation?
- Can deletion or subject-access obligations be honored?

## Workflow

1. **Read the current measurement surface** — event docs, analytics code, dashboards, experiment tooling, and product specs.
2. **Define the decision question first** — what decision needs support and what metric would change behavior?
3. **Map the event model** — user actions, system facts, experiment exposures, identity joins, and funnel stages.
4. **Review data quality and attribution risks** — duplicates, missing events, late events, cross-device issues, and naming drift.
5. **Review privacy constraints** — PII, consent, retention, and deletion expectations.
6. **Recommend the smallest instrumentation slice** — add or refine only what supports a real decision.
7. **Name the validation plan** — how the team will verify the instrumentation is correct after shipping.

## Reporting and charts

- Aggregate reports (dashboards, territory or team roll-ups) use the latest value per metric and date, guard divisions such as attainment = revenue / quota against zero, and state the window (MTD, YTD, trailing N months).
- The question picks the chart: compare categories with sorted bars, trend with a line, distribution with a histogram or box, relation with a scatter, many groups with small multiples.
- Bars start at zero. Avoid dual-axis charts, which can fake a correlation; index the series or use small multiples.
- Do not carry a number in colour alone: use a colourblind-safe palette and add a label or shape. Use categorical, sequential or diverging scales to match the data.
- Test at real data volume; SVG slows at tens of thousands of elements, so move to canvas or aggregate.

## Guardrails

- Do not add events with no clear decision owner.
- Do not stuff unbounded blobs or sensitive data into event properties.
- Do not define funnels without precise event semantics and time windows.
- Do not treat dashboard completeness as proof that the data is correct.

## Output format

```text
Measurement goal:
  Decision to support: [question]
  Primary metric(s): [list]
  Guardrail metric(s): [list]

Event model:
  Core events: [list]
  Required properties: [list]
  Funnel or experiment mapping: [summary]

Risks:
  Attribution/identity: [findings]
  Data quality: [findings]
  Privacy/retention: [findings]

Recommended instrumentation slice:
  [change] — [why] — [how to validate]
```

## Retention and lifecycle metrics

Use this section when the task involves understanding whether users are staying, why they leave, or how to measure product stickiness.

### Cohort analysis design
- A cohort groups users who shared a common experience at a common time (e.g. "users who signed up in March 2026").
- Compare retention across cohorts, not across all users — this isolates product changes from acquisition mix shifts.
- Use weekly or monthly cohort windows for B2B; daily or weekly for high-frequency consumer apps.
- Always normalize cohort size before comparing absolute retention numbers.

### D1/D7/D30 retention patterns
- **D1 (Day 1 retention)**: % of users who returned the day after their first session. Measures first-impression stickiness and onboarding effectiveness.
- **D7 (Week 1 retention)**: % of users who returned within 7 days. Measures whether the core value proposition was realized in the first week.
- **D30 (Month 1 retention)**: % of users still active at 30 days. The primary signal for product-market fit in most consumer products.
- Heuristic, unsourced: Healthy SaaS benchmarks (B2B): D30 > 70%. Consumer apps: D30 > 25% is strong. Below 10% requires urgent product investigation.
- Track the retention curve shape, not just the endpoints: a curve that plateaus (even at 15%) indicates a retained core segment; a curve that keeps falling to zero indicates no retained segment.

### Churn signal identification
- Define churn precisely before measuring it: is it inactivity for N days, explicit cancellation, or payment failure?
- Leading churn indicators appear before churn: declining login frequency, drop in core feature usage, support ticket volume, failed payment attempts.
- Segment churn by acquisition source, plan tier, company size, and cohort — aggregate churn hides the actionable signal.
- Resurrection rate (churned users who return) is a separate metric from retention; conflating them inflates apparent retention.

### Lifecycle stage metrics
Define explicit lifecycle stages and instrument the transitions between them:

| Stage | Definition | Key metric |
|-------|-----------|-----------|
| **Acquisition** | User created account | Signup rate |
| **Activation** | User completed the "aha moment" action | Activation rate |
| **Habit** | User returns without prompting in the first 7 days | D7 retention |
| **Engagement** | User uses the core feature ≥ X times per week | Weekly active usage rate |
| **Expansion** | User upgrades, adds seats, or uses more features | Expansion MRR % |
| **Dormancy** | User has not returned in N days but not churned | Dormancy rate |
| **Churn** | User meets the churn definition | Churn rate |
| **Resurrection** | Churned user returns | Resurrection rate |

Instrument the transition between each adjacent stage, not just the stages themselves — the transition events reveal where users drop off.

## References

- `../../references/checklists/analytics-metrics-checklist.md`

Adapted from engineering/engineering-data-visualization-engineer.md and specialized/data-consolidation-agent.md (agency-agents, MIT).