code-ai-installer 2.5.0 → 4.0.0

package/domains/analytics/locales/en/.agents/skills/cohort-analysis/SKILL.md

@@ -1,322 +1,322 @@
----
-name: cohort-analysis
-description: Cohort analysis — retention, revenue per cohort, behavioral patterns by groups
----
-# Cohort Analysis — Retention, revenue, and behavioral patterns by groups
-
-## When to use
-- When analyzing customer retention — to understand how retention changes over time.
-- When evaluating the effectiveness of changes — to compare cohorts before and after the changes.
-- When forecasting LTV — cohort data provides the most accurate estimate.
-- When identifying onboarding problems — early churn in new cohorts.
-
-## Input
-
-| Field | Required | Description |
-|------|:-----------:|----------|
-| Product / service | ✅ | What we are analyzing |
-| Business model | ✅ | SaaS / E-commerce / Marketplace / App |
-| Cohort tracking criteria | ✅ | By sign-up date / first purchase date / other |
-| Cohort data | ⬚ | Actual retention / revenue data (if available) |
-| Analysis period | ⬚ | For what period (default — 12 months) |
-| Granularity | ⬚ | Weeks / months / quarters (default — months) |
-| Cohort size | ⬚ | Number of users in each cohort |
-
-> If required fields are not provided — **request** them from the user, do not generate assumptions.
-
-> If there is no actual data — the skill forms a **methodological framework** with industry benchmarks that can be applied when data appears. Explicitly mark as "Framework (no actual data)".
-
-## Data sources
-1. **Product analytics** — data on sign-ups, activity, payments.
-2. **Web search** — industry retention benchmarks (Mixpanel, Amplitude, Lenny Rachitsky).
-3. **Industry reports** — average retention curves across verticals.
-4. **CRM / Payment data** — data from the user, if provided.
-5. **Public data** — retention cases of public companies (S-1 filings, earnings).
-
-> When working with benchmarks — indicate the source, date, and degree of applicability.
-
-### Connection with other skills
-| Skill | What we take | When to call |
-|------|-----------|----------------|
-| `web-research` | Industry retention benchmarks, public cases | If there is no actual data or benchmarks are needed for comparison |
-| `competitive-analysis` | Competitor retention benchmarks (if public) | For comparison with competitors (step 5) |
-| `rfm-analysis` | RFM segments for cross-analysis with cohorts | If we need to understand which segments within cohorts retain better |
-| `unit-economics` | CAC for calculating cohort payback | For step 6 (revenue per cohort → payback period) |
-| `customer-journey-mapping` | Churn points to explain retention drops | If we need to explain why a specific cohort is losing users |
-
-## Protocol
-
-### Step 0 — Context gathering
-1. Verify the availability of required data.
-2. Determine the mode: **actual data** or **methodological framework**.
-3. Fix the period and granularity.
-4. Determine the activity metric: what we consider "active" (login, transaction, key action — "aha moment").
-5. Fix exceptions: test accounts, bots, anomalies.
-
-### Step 1 — Determining cohort criteria
-
-| Parameter | Value | Justification |
-|----------|----------|-------------|
-| Primary criteria | sign-up date / first purchase / subscription | [Why this was chosen] |
-| Cohort period | week / month / quarter | [Granularity justification] |
-| Activity metric | [login / transaction / key action] | [What reflects value to the user] |
-| Observation period | [6 / 12 / 24 months] | [Depth justification] |
-| Minimum cohort size | ≥ 30 (statistical significance) | Mark cohorts < 30 with ⚠️ |
-
-### Step 2 — Retention Table (Retention Triangle)
-1. Build a retention table by cohorts:
-   - Rows: cohorts (by sign-up months).
-   - Columns: life months (M0, M1, ... MN).
-   - Values: percentage of active users relative to the initial cohort size.
-   - M0 = 100% (by definition).
-2. Highlight cells with color coding:
-   - 🟢 ≥ industry benchmark
-   - 🟡 within ±5pp from benchmark
-   - 🔴 < benchmark by > 5pp
-
-### Step 3 — Retention Curves
-Visualize retention curves:
-1. Overlay the curves of different cohorts for comparison.
-2. Identify the characteristic shape:
-
-   | Shape | Description | Interpretation |
-   |-------|----------|---------------|
-   | **Flattening (plateau)** | Decline → stabilization on a plateau | Healthy product, core of loyal users |
-   | **Continuous decay** | Constant decline without plateau | Problem with value proposition or product-market fit |
-   | **Smile curve** | Decline → growth | Reactivation works, but there is an early retention problem |
-   | **Cliff** | Sharp drop at a specific month | Problem at a specific point (end of trial, first payment) |
-
-3. Identify the stabilization point: the month when retention plateaus (Δ < 1pp between periods).
-
-### Step 4 — Patterns and anomalies
-1. Analyze cohorts for patterns:
-   - **Plateau:** on which month retention stabilizes.
-   - **Decay:** which cohorts lose users faster.
-   - **Revival:** are there any seasonal or trigger-based returns.
-   - **Improvement/deterioration:** new cohorts better or worse than old ones.
-2. Tie anomalies to events:
-
-   | Cohort | Anomaly | Event | Impact | Reliability |
-   |---------|----------|---------|---------|:-------------:|
-   | [Mar 2025] | M1 retention +8pp vs average | Launch of new onboarding | Positive | ✅ / ⚠️ / 🔮 |
-
-### Step 5 — Cohort comparison
-Compare cohorts across a minimum of 2 dimensions:
-
-| Dimension | Group A | Retention MX | Group B | Retention MX | Δ | Significance |
-|------|----------|:------------:|----------|:------------:|:-:|:----------:|
-| By time | Early (Q1) | XX% M6 | Late (Q3) | XX% M6 | Xpp | Significant / No |
-| By channel | Organic | XX% M6 | Paid | XX% M6 | Xpp | Significant / No |
-| By segment | Pro-tier | XX% M6 | Free-tier | XX% M6 | Xpp | Significant / No |
-
-**Significance evaluation of the difference:**
-- Cohort size ≥ 30 and Δ ≥ 5pp → likely significant.
-- Cohort size < 30 or Δ < 5pp → mark with ⚠️, refrain from strong conclusions.
-- If there is access to data: χ²-test or z-test for proportions.
-
-### Step 6 — Revenue per Cohort
-1. Build a cumulative revenue table:
-
-   | Cohort | Size | M0 | M1 | M3 | M6 | M12 | ARPU (M12) | CAC | Payback |
-   |---------|:------:|:--:|:--:|:--:|:--:|:---:|:----------:|:---:|:-------:|
-   | [Jan] | [N] | $XX | $XX | $XX | $XX | $XX | $XX | $XX | M[X] |
-
-2. Calculate ARPU by cohorts (cumulative revenue / cohort size).
-3. Identify payback period: the month when cumulative ARPU ≥ CAC.
-4. Identify cohorts with the best/worst monetization.
-5. Calculate projected LTV based on the retention curve:
-   - `LTV = ARPU × Σ(retention_rate_month_i)` for i = 0..N
-   - Or: `LTV = ARPU / churn_rate` (with stable churn on a plateau).
-
-> If revenue data is unavailable — skip with the mark "Revenue per Cohort: data not provided. Recommended to add when available".
-
-### Step 7 — Actionable Insights and Action Plan
-1. Formulate 3-5 key findings (each with data).
-2. For each finding — root cause (hypothesis) + specific action:
-
-   | # | Finding | Data | Root cause | Action | Metric | Expected impact |
-   |---|---------|--------|------------------|----------|---------|:----------------:|
-   | 1 | [What was derived] | [Numbers] | [Why this is so] | [What to do] | [What to measure] | +Xpp retention |
-
-3. Prioritize actions: P1 (fast impact, low effort) → P3 (slow impact, high effort).
-4. Recommend recalculation frequency (typically monthly).
-
-## Example — B2B SaaS Project Management Platform
-
-**Context:** B2B SaaS, subscription model, 2000 new sign-ups/month. Criterion: sign-up date. Granularity: months. Metric: login ≥ 1 time per month.
-
-### Retention Table
-
-| Cohort | Size | M0 | M1 | M2 | M3 | M6 | M12 |
-|---------|:------:|:---:|:---:|:---:|:---:|:---:|:----:|
-| Jan | 1850 | 100% | 42% 🔴 | 31% | 26% | 20% | 16% |
-| Feb | 2100 | 100% | 44% 🔴 | 33% | 28% | 22% | — |
-| Mar | 2200 | 100% | 55% 🟢 | 44% | 38% | — | — |
-| Apr | 1900 | 100% | 53% 🟢 | 42% | — | — | — |
-
-**B2B SaaS Benchmark:** M1: 50%, M3: 35%, M6: 25%, M12: 18% — *Source: Lenny Rachitsky, 2025*
-
-### Retention Curves
-- **Shape:** Flattening — plateau begins around M4-M5.
-- **Stabilization point:** M5 at ~20% (Jan-Feb cohorts), M5 at ~32% (Mar-Apr cohorts).
-- **Cliff:** M0→M1 = loss of 45-58% — critical transition from trial to active usage.
-- **Trend:** Improvement — Mar-Apr cohorts are +11pp better across M1.
-
-### Anomalies
-| Cohort | Anomaly | Event | Reliability |
-|---------|----------|---------|:-------------:|
-| Mar 2025 | M1 +11pp vs Jan-Feb | Launch of guided onboarding (March 1) | ✅ (A/B test) |
-
-### Revenue per Cohort
-
-| Cohort | ARPU (M6) | CAC | Payback | Projected LTV |
-|---------|:---------:|:---:|:-------:|:-------------:|
-| Jan | $89 | $120 | M8 | $210 |
-| Mar | $124 | $115 | M5 | $320 |
-
-### Key Findings & Action Plan
-
-| # | Finding | Action | Priority | Metric | Impact |
-|---|---------|----------|:---------:|---------|:------:|
-| 1 | M0→M1 cliff: 45-58% loss | Expand guided onboarding (working since March, +11pp) | P1 | M1 retention | +10pp |
-| 2 | Mar+ cohorts perform significantly better | Scale A/B winner to all channels | P1 | M1 retention | +5pp |
-| 3 | Payback Jan = M8 vs Mar = M5 | Focus marketing on channels granting high retention | P2 | Payback period | -2 mo |
-
-## Validation (Quality Gate)
-
-- [ ] Cohort tracking criteria defined with justification
-- [ ] Retention table built (or methodological framework with benchmarks)
-- [ ] Cohorts of size < 30 mapped with a ⚠️, their insights formulated cautiously
-- [ ] Retention curves described indicating shape (plateau / decay / smile / cliff)
-- [ ] Stabilization point tracked (or its absence recognized)
-- [ ] Anomalies directly tied to specific events with their reliability appraised
-- [ ] Cohorts compared against at least 2 dimensions alongside significance evaluation
-- [ ] Revenue per cohort evaluated (or stated as "data not provided")
-- [ ] Actionable findings shaped correctly with sufficient data (minimum 3 matters)
-- [ ] Action plan holds prioritized maneuvers pointing out metrics and expected impact
-- [ ] If data behaves modeling-wise / features benchmark parameters — explicitly noted paired with source and date
-- [ ] Origin data of benchmark items identified; references older than 12 mos marked with a ⚠️
-
-> If validation is missed — follow up continuously until full approval, abstaining from shifting towards the next phase.
-
-## Handoff
-The resulting `$cohort-analysis` becomes an input requirement for:
-- **Strategist / Mediator** — scaling strategies strictly relying on retention logic.
-- **`unit-economics`** — cohort LTV coupled with robust payback period modeling.
-- **`rfm-analysis`** — exploring the intersection spanning cohort metrics alongside RFM segmentation.
-- **`customer-journey-mapping`** — addressing churn inflection landmarks (cliff, decay) targeted towards path enhancement.
-- **`trend-analysis`** — discerning retention drift serving as a hallmark indicator addressing market deviations.
-
-Output package scheme: retention table + retention curves (shape, plateau) + key findings + action plan. Ensure issuing `$handoff` continuously.
-
-## Anti-patterns
-
-| Error | Why it falls short | How to do it right |
-|--------|-------------|---------------|
-| Blended/average retention bypassing cohorts | Obscures structural flaws affecting fresh cohorts (Simpson's paradox) | Inherently breakdown data strictly reflecting cohorts |
-| Raw cohort measurements lacking context | Stats offering zero clues towards "why" represent meaningless artifacts | Bind every anomaly closely reflecting actionable incidents |
-| Analyzing exceedingly thin segments | Missing statistical value → faulty interpretations | Base cohorts spanning a sheer 30 items or mark sparse volumes using ⚠️ |
-| Exclusively viewing tracking for Day 1 / Day 7 / Day 30 | Omission regarding mid-tier behavioral transitions/cliffs | Full-scale retention array deploying continuous measurement spans |
-| Skipping an explicit actionable layout | Vain assessments, rendering nil material enterprise value | Distinct findings generating exclusive and prioritized responses invariably |
-| Disregarding cross-reference scaling with revenue footprints | Displaying behaviors evading fiscal anchoring dilutes the aggregate projection value | Strictly trace gross revenue, explicit ARPU, paired tracking against cohort payback footprints |
-| Static assessment frames | Cohort matrices morph continuously rendering snapshot tracking functionally expired quickly | Monthly calibration iterations tracking side-by-side progression deltas invariably |
-| Side grading cohorts absent adequate variation confidence check | A 2pp swing possibly constitutes mere noise as opposed to clear signal | Set Δ ≥ 5pp integrated with sample ≥ 30 indicating substantiality; alternatively denote as ⚠️ |
-| Inserting naked benchmarks devoid of authentic source trace | Denies external vetting whilst exhibiting skewed modeling parameters | Pair explicit sources alongside respective publishing dates validating immediate alignment regarding contextual business frameworks |
-
-## Output Template
-
-```
-### Cohort Analysis — [Product / Service]
-
-**Business model:** [model]
-**Cohort tracking criteria:** [criteria]
-**Activity metric:** [what counts as "active"]
-**Granularity:** [weeks / months]
-**Observation period:** [N months]
-**Data type:** Actual / Framework (benchmark: [source])
-**Date of analysis:** [date]
-
----
-
-#### Retention Table (Retention Triangle)
-
-| Cohort | Size | M0 | M1 | M2 | M3 | M6 | M12 |
-|---------|:------:|:---:|:---:|:---:|:---:|:---:|:----:|
-| [Jan] | [N] | 100% | [X]% | [X]% | [X]% | [X]% | [X]% |
-| [Feb] | [N] | 100% | [X]% | [X]% | [X]% | [X]% | — |
-
-Markings: 🟢 ≥ benchmark | 🟡 ±5pp | 🔴 < benchmark by >5pp
-
-**Benchmark [industry]:**
-- M1: [X]% | M3: [X]% | M6: [X]% | M12: [X]% — *Source: [source], [date]*
-
----
-
-#### Retention Curves
-
-| Characteristic | Value |
-|---------------|----------|
-| **Curve shape** | [Plateau / Decay / Smile / Cliff] |
-| **Stabilization point** | Month [X] at [X]% (Δ < 1pp) |
-| **Critical drop (cliff)** | M[X]→M[X]: loss of [X]pp |
-| **Trend across cohorts** | Improvement / Stability / Deterioration |
-
----
-
-#### Patterns and anomalies
-
-| Cohort | Anomaly | Event | Impact | Reliability |
-|---------|----------|---------|---------|:-------------:|
-| [cohort] | [description + Δ] | [event / reason] | [pos. / neg.] | ✅ / ⚠️ / 🔮 |
-
----
-
-#### Cohort comparison
-
-| Dimension | Group A | Ret. MX | Group B | Ret. MX | Δ | Significance |
-|------|----------|:-------:|----------|:-------:|:-:|:----------:|
-| [By time] | [cohort] | [X]% | [cohort] | [X]% | [X]pp | Yes / ⚠️ |
-| [By channel] | [channel] | [X]% | [channel] | [X]% | [X]pp | Yes / ⚠️ |
-
----
-
-#### Revenue per Cohort
-
-| Cohort | Size | Cum. Rev. M6 | Cum. Rev. M12 | ARPU (M12) | CAC | Payback | Proj. LTV |
-|---------|:------:|:------------:|:-------------:|:----------:|:---:|:-------:|:---------:|
-| [Jan] | [N] | $XX | $XX | $XX | $XX | M[X] | $XX |
-
-LTV formula: [ARPU × Σ retention / ARPU ÷ churn_rate]
-
----
-
-#### Key Findings
-
-| # | Finding | Data | Root cause (hypothesis) | Reliability |
-|---|---------|--------|----------------------------|:-------------:|
-| 1 | [What was derived] | [Numbers, Δ, cohorts] | [Why this is so] | ✅ / ⚠️ / 🔮 |
-| 2 | [What was derived] | [Numbers, Δ, cohorts] | [Why this is so] | ✅ / ⚠️ / 🔮 |
-| 3 | [What was derived] | [Numbers, Δ, cohorts] | [Why this is so] | ✅ / ⚠️ / 🔮 |
-
----
-
-#### Action Plan
-
-| Priority | Action | Target metric | Expected impact | Deadline |
-|:---------:|----------|-----------------|:----------------:|------|
-| P1 | [action] | [metric] | +Xpp retention | [deadline] |
-| P2 | [action] | [metric] | +Xpp retention | [deadline] |
-| P3 | [action] | [metric] | +Xpp retention | [deadline] |
-
-**Recommended recalculation frequency:** monthly
-
----
-
-#### Data sources and assumptions
-
-| # | Fact / assumption | Source | Date | Reliability |
-|---|-------------------|----------|------|:-------------:|
-| 1 | [Fact] | [Source] | [Date] | ✅ / ⚠️ / 🔮 |
-
-Legend: ✅ Verified (2+ sources / A/B test) | ⚠️ Estimated (1 source / correlation) | 🔮 Assumed (expert hypothesis)
-```
+---
+name: cohort-analysis
+description: Cohort analysis — retention, revenue per cohort, behavioral patterns by groups
+---
+# Cohort Analysis — Retention, revenue, and behavioral patterns by groups
+
+## When to use
+- When analyzing customer retention — to understand how retention changes over time.
+- When evaluating the effectiveness of changes — to compare cohorts before and after the changes.
+- When forecasting LTV — cohort data provides the most accurate estimate.
+- When identifying onboarding problems — early churn in new cohorts.
+
+## Input
+
+| Field | Required | Description |
+|------|:-----------:|----------|
+| Product / service | ✅ | What we are analyzing |
+| Business model | ✅ | SaaS / E-commerce / Marketplace / App |
+| Cohort tracking criteria | ✅ | By sign-up date / first purchase date / other |
+| Cohort data | ⬚ | Actual retention / revenue data (if available) |
+| Analysis period | ⬚ | For what period (default — 12 months) |
+| Granularity | ⬚ | Weeks / months / quarters (default — months) |
+| Cohort size | ⬚ | Number of users in each cohort |
+
+> If required fields are not provided — **request** them from the user, do not generate assumptions.
+
+> If there is no actual data — the skill forms a **methodological framework** with industry benchmarks that can be applied when data appears. Explicitly mark as "Framework (no actual data)".
+
+## Data sources
+1. **Product analytics** — data on sign-ups, activity, payments.
+2. **Web search** — industry retention benchmarks (Mixpanel, Amplitude, Lenny Rachitsky).
+3. **Industry reports** — average retention curves across verticals.
+4. **CRM / Payment data** — data from the user, if provided.
+5. **Public data** — retention cases of public companies (S-1 filings, earnings).
+
+> When working with benchmarks — indicate the source, date, and degree of applicability.
+
+### Connection with other skills
+| Skill | What we take | When to call |
+|------|-----------|----------------|
+| `web-research` | Industry retention benchmarks, public cases | If there is no actual data or benchmarks are needed for comparison |
+| `competitive-analysis` | Competitor retention benchmarks (if public) | For comparison with competitors (step 5) |
+| `rfm-analysis` | RFM segments for cross-analysis with cohorts | If we need to understand which segments within cohorts retain better |
+| `unit-economics` | CAC for calculating cohort payback | For step 6 (revenue per cohort → payback period) |
+| `customer-journey-mapping` | Churn points to explain retention drops | If we need to explain why a specific cohort is losing users |
+
+## Protocol
+
+### Step 0 — Context gathering
+1. Verify the availability of required data.
+2. Determine the mode: **actual data** or **methodological framework**.
+3. Fix the period and granularity.
+4. Determine the activity metric: what we consider "active" (login, transaction, key action — "aha moment").
+5. Fix exceptions: test accounts, bots, anomalies.
+
+### Step 1 — Determining cohort criteria
+
+| Parameter | Value | Justification |
+|----------|----------|-------------|
+| Primary criteria | sign-up date / first purchase / subscription | [Why this was chosen] |
+| Cohort period | week / month / quarter | [Granularity justification] |
+| Activity metric | [login / transaction / key action] | [What reflects value to the user] |
+| Observation period | [6 / 12 / 24 months] | [Depth justification] |
+| Minimum cohort size | ≥ 30 (statistical significance) | Mark cohorts < 30 with ⚠️ |
+
+### Step 2 — Retention Table (Retention Triangle)
+1. Build a retention table by cohorts:
+   - Rows: cohorts (by sign-up months).
+   - Columns: life months (M0, M1, ... MN).
+   - Values: percentage of active users relative to the initial cohort size.
+   - M0 = 100% (by definition).
+2. Highlight cells with color coding:
+   - 🟢 ≥ industry benchmark
+   - 🟡 within ±5pp from benchmark
+   - 🔴 < benchmark by > 5pp
+
+### Step 3 — Retention Curves
+Visualize retention curves:
+1. Overlay the curves of different cohorts for comparison.
+2. Identify the characteristic shape:
+
+   | Shape | Description | Interpretation |
+   |-------|----------|---------------|
+   | **Flattening (plateau)** | Decline → stabilization on a plateau | Healthy product, core of loyal users |
+   | **Continuous decay** | Constant decline without plateau | Problem with value proposition or product-market fit |
+   | **Smile curve** | Decline → growth | Reactivation works, but there is an early retention problem |
+   | **Cliff** | Sharp drop at a specific month | Problem at a specific point (end of trial, first payment) |
+
+3. Identify the stabilization point: the month when retention plateaus (Δ < 1pp between periods).
+
+### Step 4 — Patterns and anomalies
+1. Analyze cohorts for patterns:
+   - **Plateau:** on which month retention stabilizes.
+   - **Decay:** which cohorts lose users faster.
+   - **Revival:** are there any seasonal or trigger-based returns.
+   - **Improvement/deterioration:** new cohorts better or worse than old ones.
+2. Tie anomalies to events:
+
+   | Cohort | Anomaly | Event | Impact | Reliability |
+   |---------|----------|---------|---------|:-------------:|
+   | [Mar 2025] | M1 retention +8pp vs average | Launch of new onboarding | Positive | ✅ / ⚠️ / 🔮 |
+
+### Step 5 — Cohort comparison
+Compare cohorts across a minimum of 2 dimensions:
+
+| Dimension | Group A | Retention MX | Group B | Retention MX | Δ | Significance |
+|------|----------|:------------:|----------|:------------:|:-:|:----------:|
+| By time | Early (Q1) | XX% M6 | Late (Q3) | XX% M6 | Xpp | Significant / No |
+| By channel | Organic | XX% M6 | Paid | XX% M6 | Xpp | Significant / No |
+| By segment | Pro-tier | XX% M6 | Free-tier | XX% M6 | Xpp | Significant / No |
+
+**Significance evaluation of the difference:**
+- Cohort size ≥ 30 and Δ ≥ 5pp → likely significant.
+- Cohort size < 30 or Δ < 5pp → mark with ⚠️, refrain from strong conclusions.
+- If there is access to data: χ²-test or z-test for proportions.
+
+### Step 6 — Revenue per Cohort
+1. Build a cumulative revenue table:
+
+   | Cohort | Size | M0 | M1 | M3 | M6 | M12 | ARPU (M12) | CAC | Payback |
+   |---------|:------:|:--:|:--:|:--:|:--:|:---:|:----------:|:---:|:-------:|
+   | [Jan] | [N] | $XX | $XX | $XX | $XX | $XX | $XX | $XX | M[X] |
+
+2. Calculate ARPU by cohorts (cumulative revenue / cohort size).
+3. Identify payback period: the month when cumulative ARPU ≥ CAC.
+4. Identify cohorts with the best/worst monetization.
+5. Calculate projected LTV based on the retention curve:
+   - `LTV = ARPU × Σ(retention_rate_month_i)` for i = 0..N
+   - Or: `LTV = ARPU / churn_rate` (with stable churn on a plateau).
+
+> If revenue data is unavailable — skip with the mark "Revenue per Cohort: data not provided. Recommended to add when available".
+
+### Step 7 — Actionable Insights and Action Plan
+1. Formulate 3-5 key findings (each with data).
+2. For each finding — root cause (hypothesis) + specific action:
+
+   | # | Finding | Data | Root cause | Action | Metric | Expected impact |
+   |---|---------|--------|------------------|----------|---------|:----------------:|
+   | 1 | [What was derived] | [Numbers] | [Why this is so] | [What to do] | [What to measure] | +Xpp retention |
+
+3. Prioritize actions: P1 (fast impact, low effort) → P3 (slow impact, high effort).
+4. Recommend recalculation frequency (typically monthly).
+
+## Example — B2B SaaS Project Management Platform
+
+**Context:** B2B SaaS, subscription model, 2000 new sign-ups/month. Criterion: sign-up date. Granularity: months. Metric: login ≥ 1 time per month.
+
+### Retention Table
+
+| Cohort | Size | M0 | M1 | M2 | M3 | M6 | M12 |
+|---------|:------:|:---:|:---:|:---:|:---:|:---:|:----:|
+| Jan | 1850 | 100% | 42% 🔴 | 31% | 26% | 20% | 16% |
+| Feb | 2100 | 100% | 44% 🔴 | 33% | 28% | 22% | — |
+| Mar | 2200 | 100% | 55% 🟢 | 44% | 38% | — | — |
+| Apr | 1900 | 100% | 53% 🟢 | 42% | — | — | — |
+
+**B2B SaaS Benchmark:** M1: 50%, M3: 35%, M6: 25%, M12: 18% — *Source: Lenny Rachitsky, 2025*
+
+### Retention Curves
+- **Shape:** Flattening — plateau begins around M4-M5.
+- **Stabilization point:** M5 at ~20% (Jan-Feb cohorts), M5 at ~32% (Mar-Apr cohorts).
+- **Cliff:** M0→M1 = loss of 45-58% — critical transition from trial to active usage.
+- **Trend:** Improvement — Mar-Apr cohorts are +11pp better across M1.
+
+### Anomalies
+| Cohort | Anomaly | Event | Reliability |
+|---------|----------|---------|:-------------:|
+| Mar 2025 | M1 +11pp vs Jan-Feb | Launch of guided onboarding (March 1) | ✅ (A/B test) |
+
+### Revenue per Cohort
+
+| Cohort | ARPU (M6) | CAC | Payback | Projected LTV |
+|---------|:---------:|:---:|:-------:|:-------------:|
+| Jan | $89 | $120 | M8 | $210 |
+| Mar | $124 | $115 | M5 | $320 |
+
+### Key Findings & Action Plan
+
+| # | Finding | Action | Priority | Metric | Impact |
+|---|---------|----------|:---------:|---------|:------:|
+| 1 | M0→M1 cliff: 45-58% loss | Expand guided onboarding (working since March, +11pp) | P1 | M1 retention | +10pp |
+| 2 | Mar+ cohorts perform significantly better | Scale A/B winner to all channels | P1 | M1 retention | +5pp |
+| 3 | Payback Jan = M8 vs Mar = M5 | Focus marketing on channels granting high retention | P2 | Payback period | -2 mo |
+
+## Validation (Quality Gate)
+
+- [ ] Cohort tracking criteria defined with justification
+- [ ] Retention table built (or methodological framework with benchmarks)
+- [ ] Cohorts of size < 30 mapped with a ⚠️, their insights formulated cautiously
+- [ ] Retention curves described indicating shape (plateau / decay / smile / cliff)
+- [ ] Stabilization point tracked (or its absence recognized)
+- [ ] Anomalies directly tied to specific events with their reliability appraised
+- [ ] Cohorts compared against at least 2 dimensions alongside significance evaluation
+- [ ] Revenue per cohort evaluated (or stated as "data not provided")
+- [ ] Actionable findings shaped correctly with sufficient data (minimum 3 matters)
+- [ ] Action plan holds prioritized maneuvers pointing out metrics and expected impact
+- [ ] If data behaves modeling-wise / features benchmark parameters — explicitly noted paired with source and date
+- [ ] Origin data of benchmark items identified; references older than 12 mos marked with a ⚠️
+
+> If validation is missed — follow up continuously until full approval, abstaining from shifting towards the next phase.
+
+## Handoff
+The resulting `$cohort-analysis` becomes an input requirement for:
+- **Strategist / Mediator** — scaling strategies strictly relying on retention logic.
+- **`unit-economics`** — cohort LTV coupled with robust payback period modeling.
+- **`rfm-analysis`** — exploring the intersection spanning cohort metrics alongside RFM segmentation.
+- **`customer-journey-mapping`** — addressing churn inflection landmarks (cliff, decay) targeted towards path enhancement.
+- **`trend-analysis`** — discerning retention drift serving as a hallmark indicator addressing market deviations.
+
+Output package scheme: retention table + retention curves (shape, plateau) + key findings + action plan. Ensure issuing `$handoff` continuously.
+
+## Anti-patterns
+
+| Error | Why it falls short | How to do it right |
+|--------|-------------|---------------|
+| Blended/average retention bypassing cohorts | Obscures structural flaws affecting fresh cohorts (Simpson's paradox) | Inherently breakdown data strictly reflecting cohorts |
+| Raw cohort measurements lacking context | Stats offering zero clues towards "why" represent meaningless artifacts | Bind every anomaly closely reflecting actionable incidents |
+| Analyzing exceedingly thin segments | Missing statistical value → faulty interpretations | Base cohorts spanning a sheer 30 items or mark sparse volumes using ⚠️ |
+| Exclusively viewing tracking for Day 1 / Day 7 / Day 30 | Omission regarding mid-tier behavioral transitions/cliffs | Full-scale retention array deploying continuous measurement spans |
+| Skipping an explicit actionable layout | Vain assessments, rendering nil material enterprise value | Distinct findings generating exclusive and prioritized responses invariably |
+| Disregarding cross-reference scaling with revenue footprints | Displaying behaviors evading fiscal anchoring dilutes the aggregate projection value | Strictly trace gross revenue, explicit ARPU, paired tracking against cohort payback footprints |
+| Static assessment frames | Cohort matrices morph continuously rendering snapshot tracking functionally expired quickly | Monthly calibration iterations tracking side-by-side progression deltas invariably |
+| Side grading cohorts absent adequate variation confidence check | A 2pp swing possibly constitutes mere noise as opposed to clear signal | Set Δ ≥ 5pp integrated with sample ≥ 30 indicating substantiality; alternatively denote as ⚠️ |
+| Inserting naked benchmarks devoid of authentic source trace | Denies external vetting whilst exhibiting skewed modeling parameters | Pair explicit sources alongside respective publishing dates validating immediate alignment regarding contextual business frameworks |
+
+## Output Template
+
+```
+### Cohort Analysis — [Product / Service]
+
+**Business model:** [model]
+**Cohort tracking criteria:** [criteria]
+**Activity metric:** [what counts as "active"]
+**Granularity:** [weeks / months]
+**Observation period:** [N months]
+**Data type:** Actual / Framework (benchmark: [source])
+**Date of analysis:** [date]
+
+---
+
+#### Retention Table (Retention Triangle)
+
+| Cohort | Size | M0 | M1 | M2 | M3 | M6 | M12 |
+|---------|:------:|:---:|:---:|:---:|:---:|:---:|:----:|
+| [Jan] | [N] | 100% | [X]% | [X]% | [X]% | [X]% | [X]% |
+| [Feb] | [N] | 100% | [X]% | [X]% | [X]% | [X]% | — |
+
+Markings: 🟢 ≥ benchmark | 🟡 ±5pp | 🔴 < benchmark by >5pp
+
+**Benchmark [industry]:**
+- M1: [X]% | M3: [X]% | M6: [X]% | M12: [X]% — *Source: [source], [date]*
+
+---
+
+#### Retention Curves
+
+| Characteristic | Value |
+|---------------|----------|
+| **Curve shape** | [Plateau / Decay / Smile / Cliff] |
+| **Stabilization point** | Month [X] at [X]% (Δ < 1pp) |
+| **Critical drop (cliff)** | M[X]→M[X]: loss of [X]pp |
+| **Trend across cohorts** | Improvement / Stability / Deterioration |
+
+---
+
+#### Patterns and anomalies
+
+| Cohort | Anomaly | Event | Impact | Reliability |
+|---------|----------|---------|---------|:-------------:|
+| [cohort] | [description + Δ] | [event / reason] | [pos. / neg.] | ✅ / ⚠️ / 🔮 |
+
+---
+
+#### Cohort comparison
+
+| Dimension | Group A | Ret. MX | Group B | Ret. MX | Δ | Significance |
+|------|----------|:-------:|----------|:-------:|:-:|:----------:|
+| [By time] | [cohort] | [X]% | [cohort] | [X]% | [X]pp | Yes / ⚠️ |
+| [By channel] | [channel] | [X]% | [channel] | [X]% | [X]pp | Yes / ⚠️ |
+
+---
+
+#### Revenue per Cohort
+
+| Cohort | Size | Cum. Rev. M6 | Cum. Rev. M12 | ARPU (M12) | CAC | Payback | Proj. LTV |
+|---------|:------:|:------------:|:-------------:|:----------:|:---:|:-------:|:---------:|
+| [Jan] | [N] | $XX | $XX | $XX | $XX | M[X] | $XX |
+
+LTV formula: [ARPU × Σ retention / ARPU ÷ churn_rate]
+
+---
+
+#### Key Findings
+
+| # | Finding | Data | Root cause (hypothesis) | Reliability |
+|---|---------|--------|----------------------------|:-------------:|
+| 1 | [What was derived] | [Numbers, Δ, cohorts] | [Why this is so] | ✅ / ⚠️ / 🔮 |
+| 2 | [What was derived] | [Numbers, Δ, cohorts] | [Why this is so] | ✅ / ⚠️ / 🔮 |
+| 3 | [What was derived] | [Numbers, Δ, cohorts] | [Why this is so] | ✅ / ⚠️ / 🔮 |
+
+---
+
+#### Action Plan
+
+| Priority | Action | Target metric | Expected impact | Deadline |
+|:---------:|----------|-----------------|:----------------:|------|
+| P1 | [action] | [metric] | +Xpp retention | [deadline] |
+| P2 | [action] | [metric] | +Xpp retention | [deadline] |
+| P3 | [action] | [metric] | +Xpp retention | [deadline] |
+
+**Recommended recalculation frequency:** monthly
+
+---
+
+#### Data sources and assumptions
+
+| # | Fact / assumption | Source | Date | Reliability |
+|---|-------------------|----------|------|:-------------:|
+| 1 | [Fact] | [Source] | [Date] | ✅ / ⚠️ / 🔮 |
+
+Legend: ✅ Verified (2+ sources / A/B test) | ⚠️ Estimated (1 source / correlation) | 🔮 Assumed (expert hypothesis)
+```