churn-prediction

流失预测

通过系统分析使用模式、支持互动和客户关系健康状况，识别客户流失的早期预警信号。

何时使用此技能

• 每月／每季度的流失风险评估
• 确定 CSM 干预的优先级
• 构建早期预警系统
• 对已流失客户进行事后分析
• 面向高管的流失报告

方法论基础

基于 Lincoln Murphy 的流失分析和 ProfitWell 留存研究，分析以下方面：

• 产品参与度下降
• 支持情绪趋势
• 付款行为变化
• 客户关系恶化
• 竞争信号

Claude 负责什么，哪些由你决定

Claude 负责	你决定
识别风险信号	挽留还是放弃
计算风险评分	资源分配
提出干预建议	折扣／让步方案
确定高风险客户的优先级	升级至高管处理的时机
分析流失模式	留存策略调整

此技能的功能

1. 信号检测 - 识别流失风险的行为指标
2. 风险评分 - 计算流失概率
3. 根本原因分析 - 客户为什么可能离开？
4. 干预规划 - 可以采取哪些措施挽留客户？
5. 模式识别 - 从过去流失的客户中吸取经验

使用方法

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]

说明

第 1 步：评估领先指标

提前 30-60 天出现的预警信号：

信号	风险等级	权重
登录次数下降 >50%	高	15
停止使用某项功能	高	15
支持工单激增	中	10
客户拥护者离职	严重	20
NPS 为负	高	12
付款逾期	中	8
未参加 QBR	中	8
提及竞争对手	高	12

第 2 步：计算流失概率

风险评分公式：

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)

第 3 步：确定根本原因类别

类别	指标	典型挽留率
产品契合度	采用率低、使用场景不匹配	30%
价值差距	未看到投资回报、预算压力	45%
服务问题	支持服务失败、Bug 未解决	60%
客户关系	客户拥护者离职、缺乏互动	35%
竞争	正在积极评估其他方案	25%
业务变化	并购、预算削减、业务转型	15%

第 4 步：制定干预方案

按根本原因分类：

原因	主要行动	次要行动
产品契合度	制定成功计划	调整合同规模
价值差距	评估投资回报	致电高管赞助人
服务问题	升级处理并解决问题	服务抵扣
客户关系	培养新的客户拥护者	梳理高管关系
竞争	制定竞争防御方案	审查定价
业务变化	灵活条款	暂停选项

第 5 步：制定挽留计划

90 天挽留框架：

• 第 1-7 天：分诊并稳定局面
• 第 8-30 天：解决根本原因
• 第 31-60 天：重塑价值认知
• 第 61-90 天：争取承诺

示例

示例 1：高风险客户

输入：

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

输出：

## 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

示例 2：客户群体流失分析

输入：

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

输出：

## 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

技能边界

此技能擅长的方面

• 系统分析风险信号
• 依据明确逻辑进行概率评分
• 对根本原因进行分类
• 规划干预措施

此技能无法做到的方面

• 访问实际客户数据
• 准确预测流失时间
• 了解客户内部动态
• 取代关系直觉

需要升级至人工处理的情况

• 战略客户（排名前 10%）
• 复杂的多产品客户关系
• 需要法务／高管参与的升级处理
• 定价让步决策

迭代指南

后续提示词

• “为此客户制定一份 30-60-90 天挽留计划。”
• “我们应该准备什么竞争应对方案？”
• “我应该优先处理哪些高风险客户？”
• “分析我所有流失客户的共同模式。”

监控频率

1. 每月为所有账户评分
2. 分数下降超过 15 分时发出警报
3. 每周检查高危和严重风险账户
4. 每季度分析模式

清单和模板

风险账户检查清单

• 分析使用趋势（30/60/90 天）
• 审查支持服务反馈
• 确认拥护者状态
• 收集或请求 NPS
• 检查竞争对手相关信号
• 核实财务状况
• 记录挽留计划

参考资料

• Lincoln Murphy 的客户流失分析框架
• ProfitWell 留存基准
• Gainsight 客户健康度方法
• ChurnZero 预测分析

相关技能

• account-health - 更全面的健康度评分
• health-score-monitor - 持续监控
• renewal-management - 续约流程

技能元数据

• 领域：客户成功
• 复杂度：高级
• 模式：centaur
• 价值实现时间：每个账户 20-30 分钟
• 前置条件：客户数据、历史记录和背景信息