churn-prediction
О программе
Этот навык помогает разработчикам выявлять клиентов из группы риска, которые могут отказаться от услуг, анализируя поведенческие сигналы, модели взаимодействия и индикаторы состояния. Он полезен для построения систем раннего предупреждения, определения приоритетов в поддержке клиентов и проведения анализа рисков оттока. Инструмент основан на проверенных методиках анализа оттока для обнаружения снижения вовлеченности в продукт и ухудшения взаимоотношений.
Быстрая установка
Claude Code
Рекомендуетсяnpx skills add guia-matthieu/clawfu-skills -a claude-code/plugin add https://github.com/guia-matthieu/clawfu-skillsgit clone https://github.com/guia-matthieu/clawfu-skills.git ~/.claude/skills/churn-predictionСкопируйте и вставьте эту команду в Claude Code для установки этого навыка
Документация
Churn Prediction
Detect early warning signals of customer churn through systematic analysis of usage patterns, support interactions, and relationship health.
When to Use This Skill
- Monthly/quarterly churn risk reviews
- Prioritizing CSM intervention
- Building early warning systems
- Post-mortem analysis on lost customers
- Executive churn reporting
Methodology Foundation
Based on Lincoln Murphy's Churn Analysis and ProfitWell Retention Research, analyzing:
- Product engagement decay
- Support sentiment trends
- Payment behavior changes
- Relationship deterioration
- Competitive signals
What Claude Does vs What You Decide
| Claude Does | You Decide |
|---|---|
| Identifies risk signals | Save vs. let go decisions |
| Calculates risk scores | Resource allocation |
| Suggests interventions | Discount/concession offers |
| Prioritizes at-risk accounts | Executive escalation timing |
| Analyzes churn patterns | Retention strategy changes |
What This Skill Does
- Signal detection - Identify behavioral indicators of churn risk
- Risk scoring - Calculate churn probability
- Root cause analysis - Why are they likely to leave?
- Intervention planning - What actions could save them?
- Pattern recognition - Learn from past churned accounts
How to Use
Assess churn risk for this customer:
Account: [Company Name]
Contract: $[ARR], Renewal: [Date]
Tenure: [Months]
Usage Signals:
- Login frequency: [trend]
- Feature adoption: [% and trend]
- Active users: [current vs licensed]
- Key feature usage: [specific metrics]
Support Signals:
- Recent tickets: [count and nature]
- CSAT trend: [improving/stable/declining]
- Escalations: [any open or recent]
- Sentiment: [last few interactions]
Relationship Signals:
- Champion status: [engaged/disengaged/left]
- Exec sponsor: [status]
- NPS response: [score and comments]
- QBR attendance: [pattern]
Financial Signals:
- Payment status: [current/late]
- Contract discussions: [any mentions of changes]
- Competitor mentions: [any signals]
Instructions
Step 1: Evaluate Leading Indicators
30-60 Day Warning Signs:
| Signal | Risk Level | Weight |
|---|---|---|
| Login drop >50% | High | 15 |
| Feature usage stopped | High | 15 |
| Support tickets spike | Medium | 10 |
| Champion left | Critical | 20 |
| Negative NPS | High | 12 |
| Payment late | Medium | 8 |
| No QBR attendance | Medium | 8 |
| Competitor mentioned | High | 12 |
Step 2: Calculate Churn Probability
Risk Score Formula:
Churn Risk = Sum of weighted signals / 100
Score Ranges:
- 0-20: Low Risk (normal attention)
- 21-40: Moderate Risk (proactive outreach)
- 41-60: High Risk (intervention required)
- 61-80: Critical Risk (executive escalation)
- 81-100: Imminent Churn (save or plan exit)
Step 3: Identify Root Cause Category
| Category | Indicators | Typical Save Rate |
|---|---|---|
| Product Fit | Low adoption, wrong use case | 30% |
| Value Gap | Not seeing ROI, budget pressure | 45% |
| Service Issue | Support failures, unresolved bugs | 60% |
| Relationship | Champion left, no engagement | 35% |
| Competition | Actively evaluating others | 25% |
| Business Change | M&A, budget cuts, pivot | 15% |
Step 4: Prescribe Intervention
By Root Cause:
| Cause | Primary Action | Secondary Action |
|---|---|---|
| Product Fit | Success planning | Right-size contract |
| Value Gap | ROI review | Executive sponsor call |
| Service Issue | Escalation + resolution | Service credits |
| Relationship | New champion dev | Executive mapping |
| Competition | Competitive defense | Pricing review |
| Business | Flexible terms | Pause option |
Step 5: Create Save Plan
90-Day Save Framework:
- Days 1-7: Triage and stabilize
- Days 8-30: Address root cause
- Days 31-60: Rebuild value perception
- Days 61-90: Secure commitment
Examples
Example 1: High-Risk Account
Input:
Account: MediaTech Corp
Contract: $96K ARR, Renewal: April 15 (45 days)
Tenure: 18 months
Usage Signals:
- Logins: Down 65% last 30 days
- Feature adoption: 35% (was 60%)
- Active users: 8 of 25 licensed
- Core feature: Stopped using analytics module
Support Signals:
- Tickets: 8 this month (normally 2)
- CSAT: Dropped from 4.5 to 3.2
- Escalation: 1 open (data export issue)
- Sentiment: Last 3 interactions negative
Relationship Signals:
- Champion: VP Marketing left 6 weeks ago
- Exec sponsor: Never established
- NPS: 4 (was 8 six months ago)
- QBR: Cancelled last two
Financial Signals:
- Payment: Current
- Mentions: "Evaluating options" in last call
- Competitor: Saw CompetitorX demo on their calendar
Output:
## Churn Risk Assessment: MediaTech Corp
### 🔴 CRITICAL RISK - 78/100
**Churn Probability: 75-85%**
**Time to Action: Immediate**
---
### Risk Signal Analysis
| Signal | Finding | Points |
|--------|---------|--------|
| Usage drop 65% | Critical | 15/15 |
| Feature stopped | Analytics abandoned | 15/15 |
| Champion left | No replacement | 20/20 |
| NPS Detractor | Dropped 4 points | 12/12 |
| Competitor eval | Demo scheduled | 12/12 |
| No QBR | 2 cancelled | 8/8 |
| **Total** | | **78/100** |
### Root Cause Analysis
**Primary: Relationship Void** (Champion departure)
- VP Marketing left 6 weeks ago
- No executive sponsor ever established
- Organizational knowledge lost
- No one internally championing value
**Secondary: Value Gap** (Product not delivering)
- Usage collapse suggests not getting value
- Analytics module abandoned = key feature unused
- May have been champion-dependent usage
**Tertiary: Active Competition**
- CompetitorX demo on calendar
- "Evaluating options" language
- Classic exit signals
### Competitor Intelligence
**CompetitorX Positioning:**
- Likely pitching: Simpler interface, lower price
- Their weakness: Less robust analytics
- Our defense: Depth of features, integration
### 90-Day Save Plan
**WEEK 1: Emergency Triage**
| Day | Action | Owner |
|-----|--------|-------|
| 1 | Resolve data export escalation | Support |
| 1 | CSM call to understand situation | CSM |
| 2 | Identify new potential champion | CSM |
| 3 | VP CS call to express commitment | VP CS |
| 5 | Executive sponsor outreach (their CMO) | CRO |
**WEEK 2-4: Stabilize**
- [ ] Close escalation completely
- [ ] Meet new champion (find one!)
- [ ] Re-onboard inactive users
- [ ] Document value delivered to date
- [ ] Competitive battle card preparation
**WEEK 5-8: Rebuild Value**
- [ ] ROI review presentation
- [ ] Success planning session
- [ ] Showcase new features
- [ ] Get 3 quick wins documented
**WEEK 9-12: Secure Renewal**
- [ ] QBR with exec attendance
- [ ] Renewal discussion
- [ ] Multi-year incentive if appropriate
- [ ] Reference/case study request (confidence signal)
### Save Probability Assessment
| If We... | Save Probability |
|----------|------------------|
| Do nothing | 15% |
| Standard outreach | 25% |
| Execute full plan | 45% |
| Add discount/concession | 55% |
| Exec-to-exec + plan | 60% |
### Decision Point
**Recommended: Full save effort**
- $96K ARR worth 60-90 hours of CSM time
- Cost to acquire replacement: ~$30K
- Reputation risk if churns
**If No Traction by Day 30:**
- Prepare for graceful offboarding
- Offer reduced contract if budget issue
- Maintain relationship for potential return
Example 2: Cohort Churn Analysis
Input:
Analyze churn patterns from these 10 churned accounts:
1. AlphaCo: Left after 8 months, champion left, low usage
2. BetaTech: Left after 14 months, pricing, competitor win
3. GammaCorp: Left after 6 months, wrong fit, never adopted
4. DeltaInc: Left after 24 months, budget cuts, loved product
5. EchoSys: Left after 10 months, support issues, 3 escalations
6. FoxtrotLLC: Left after 18 months, competitor, champion left
7. GolfCo: Left after 4 months, implementation failed
8. HotelGrp: Left after 12 months, didn't see ROI
9. IndiaInc: Left after 9 months, champion left, low NPS
10. JulietCorp: Left after 20 months, M&A, product deprecated
Output:
## Churn Cohort Analysis: Q4 Losses
### Summary
- **Total Churned**: 10 accounts
- **Total ARR Lost**: [Calculate from data]
- **Average Tenure**: 12.5 months
---
### Root Cause Distribution
| Root Cause | Count | % | Avg Tenure |
|------------|-------|---|------------|
| Champion Left | 4 | 40% | 11.3 mo |
| Competitor | 3 | 30% | 17.3 mo |
| Product/Fit | 2 | 20% | 5.0 mo |
| Business Change | 2 | 20% | 22.0 mo |
| Service/Support | 1 | 10% | 10.0 mo |
| Value/ROI | 2 | 20% | 10.0 mo |
*Note: Some accounts had multiple causes*
### Key Insights
**1. Champion Dependency is Critical (40%)**
- 4 of 10 churns involved champion departure
- Average: Churned 3-4 months after champion left
- **Action**: Multi-threading program required
**2. Early Churn = Fit Problem**
- 3 accounts churned <6 months
- All had adoption/implementation issues
- **Action**: Improve qualification + onboarding
**3. Competitor Wins Correlate with Tenure**
- Competitor losses at 14, 18, 20 months
- Long enough to evaluate alternatives
- **Action**: Value reinforcement at 12-month mark
**4. Business Change is Uncontrollable**
- 2 churns from M&A/budget cuts
- Both were "happy" customers
- **Action**: Accept, maintain relationship
### Early Warning Signal Validation
| Signal | Present Before Churn | Lead Time |
|--------|---------------------|-----------|
| Champion left | 4/10 (40%) | 3-4 months |
| Usage drop >40% | 7/10 (70%) | 6-8 weeks |
| NPS drop | 6/10 (60%) | 2-3 months |
| Missed QBR | 5/10 (50%) | 3-4 months |
| Support spike | 3/10 (30%) | 4-6 weeks |
**Best Predictor**: Usage drop >40% (70% correlation)
### Preventability Assessment
| Account | Preventable? | What Would Have Helped |
|---------|--------------|------------------------|
| AlphaCo | Likely | Champion backup plan |
| BetaTech | Possibly | Competitive defense earlier |
| GammaCorp | Unlikely | Better qualification |
| DeltaInc | No | Business change |
| EchoSys | Likely | Faster escalation resolution |
| FoxtrotLLC | Possibly | Multi-thread + compete |
| GolfCo | Likely | Implementation oversight |
| HotelGrp | Likely | Proactive ROI review |
| IndiaInc | Likely | Champion backup |
| JulietCorp | No | M&A out of control |
**Preventability Rate**: 60% (6/10 could have been saved)
### Recommendations
**Process Changes:**
1. Implement champion backup contact rule (2+ contacts)
2. Add 12-month value review to CSM playbook
3. Create competitive defense triggers
4. Improve implementation success metrics
**Investment Areas:**
1. CSM capacity for proactive outreach
2. Competitive intelligence
3. Champion development program
**Metrics to Track:**
- Champion backup coverage %
- Time to first value
- Competitive mention alerts
- 12-month NPS trend
Skill Boundaries
What This Skill Does Well
- Systematic risk signal analysis
- Probability scoring with clear logic
- Root cause categorization
- Intervention planning
What This Skill Cannot Do
- Access actual customer data
- Predict exact churn timing
- Know internal customer dynamics
- Replace relationship intuition
When to Escalate to Human
- Strategic accounts (top 10%)
- Complex multi-product relationships
- Escalation requiring legal/exec involvement
- Pricing concession decisions
Iteration Guide
Follow-up Prompts
- "Create a 30-60-90 save plan for this account."
- "What competitive response should we prepare?"
- "Which of my at-risk accounts should I prioritize?"
- "Analyze the pattern across all my churned accounts."
Monitoring Cadence
- Score all accounts monthly
- Alert on score drops >15 points
- Weekly review of Critical/High risk
- Quarterly pattern analysis
Checklists & Templates
At-Risk Account Checklist
- Usage trend analyzed (30/60/90 day)
- Support sentiment reviewed
- Champion status confirmed
- NPS collected or requested
- Competitor signals checked
- Financial health verified
- Save plan documented
References
- Lincoln Murphy's Churn Analysis Framework
- ProfitWell Retention Benchmarks
- Gainsight Customer Health Methodology
- ChurnZero Predictive Analytics
Related Skills
account-health- Broader health scoringhealth-score-monitor- Continuous monitoringrenewal-management- Renewal process
Skill Metadata
- Domain: Customer Success
- Complexity: Advanced
- Mode: centaur
- Time to Value: 20-30 min per account
- Prerequisites: Customer data, history, context
GitHub репозиторий
Похожие навыки
llamaguard
ДругоеLlamaGuard — это модель от Meta с 7–8 миллиардами параметров для модерации входных и выходных данных больших языковых моделей по шести категориям безопасности, таким как насилие и разжигание ненависти. Она обеспечивает точность 94–95% и может быть развернута с помощью vLLM, Hugging Face или Amazon SageMaker. Используйте этот навык, чтобы легко интегрировать фильтрацию контента и защитные механизмы в ваши ИИ-приложения.
cost-optimization
ДругоеЭтот навык Claude помогает разработчикам оптимизировать облачные расходы за счет правильного подбора ресурсов, стратегий тегирования и анализа затрат. Он предоставляет framework для сокращения облачных расходов и внедрения управления затратами в AWS, Azure и GCP. Используйте его, когда вам нужно проанализировать расходы на инфраструктуру, оптимизировать ресурсы или уложиться в бюджетные ограничения.
quantizing-models-bitsandbytes
ДругоеЭтот навык выполняет квантизацию LLM до 8-битной или 4-битной точности с использованием библиотеки bitsandbytes, обеспечивая сокращение использования памяти на 50-75% при минимальной потере точности. Он идеально подходит для запуска больших моделей при ограниченной памяти GPU или для ускорения вывода, поддерживая форматы INT8, NF4 и FP4. Навык интегрируется с HuggingFace Transformers и позволяет использовать обучение QLoRA и 8-битные оптимизаторы.
dispatching-parallel-agents
ДругоеЭтот навык Claude распределяет нескольких агентов для исследования и устранения трёх и более независимых проблем параллельно. Он предназначен для сценариев с несвязанными сбоями, которые можно устранить без общего состояния или зависимостей. Ключевая возможность — параллельное решение проблем, где за каждую независимую предметную область назначается отдельный агент для максимальной эффективности.
