churn-predictor（流失预测器）

Churn Predictor

Expert churn prediction system that identifies at-risk customers before they leave using behavioral signals, engagement patterns, and predictive analytics. This skill provides structured workflows for building churn models, monitoring risk signals, and executing retention interventions.

Churn is the silent killer of growth. By the time a customer announces they're leaving, it's often too late. This skill helps you identify churn risk early when intervention can still make a difference, prioritize retention efforts, and systematically reduce churn.

Built on data science best practices and customer success methodologies, this skill combines leading indicator analysis, risk scoring, and intervention playbooks to predict and prevent churn before it happens.

Core Workflows

Workflow 1: Churn Signal Identification

Map the behaviors that predict churn

1. Behavioral Signals

1. Account Signals

- Champion departure (key user leaves) - Company layoffs or restructuring - Merger/acquisition announcements - Budget cuts affecting your category - Competitor evaluation signals - Contract not renewed on auto-renew

1. Relationship Signals

- NPS score decline (9-10 → 7 or below) - Missed QBRs or check-ins - Unresponsive to outreach - Escalated support issues - Negative sentiment in communications

1. Time-Based Signals

- Approaching renewal (90/60/30 days) - End of trial or pilot - Anniversary of bad experience - Post-implementation plateau - Seasonal usage patterns

Workflow 2: Risk Scoring Model

Build a composite churn risk score

1. Score Components

   Churn Risk Score =
     (Usage Score × 0.30) +
     (Engagement Score × 0.25) +
     (Support Score × 0.20) +
     (Relationship Score × 0.15) +
     (Account Score × 0.10)

   Scale: 0-100 (higher = more at risk)

1. Usage Score Factors

- Login frequency vs. baseline - Feature adoption breadth - Active users vs. licensed seats - Time in product - Core action completion

1. Engagement Score Factors

- Email open/click rates - Meeting attendance - Resource downloads - Training completion - Community participation

1. Risk Categories

Workflow 3: Cohort & Trend Analysis

Understand churn patterns across customer segments

1. Cohort Analysis

- Analyze by signup month/quarter - Track retention curves over time - Identify cohorts with worse retention - Correlate with product/market changes - Find patterns in successful cohorts

1. Segment Analysis

- By customer size (SMB/Mid/Enterprise) - By industry vertical - By use case/persona - By acquisition source - By pricing tier

1. Churn Timing Patterns

- When in customer lifecycle does churn occur? - Renewal vs. mid-contract churn - Time from warning signs to churn - Seasonal patterns - Correlation with contract length

1. Leading Indicator Validation

- Track signals → churn correlation - Calculate signal lead time - Measure false positive rate - Refine scoring weights - A/B test interventions

Workflow 4: Alert & Escalation System

Surface risk at the right time to the right people

1. Alert Triggers

- Score crosses threshold (e.g., into "elevated") - Rapid score increase (10+ points in 7 days) - Critical signal detected (payment failed, champion left) - Renewal approaching with elevated risk - Multiple signals converging

1. Escalation Matrix

1. Alert Content

- Customer name and risk score - Specific signals triggering alert - Score trend (improving/declining) - Renewal date and ARR at risk - Recommended actions

1. Alert Channels

- Slack/Teams notifications - Email digests - CRM dashboards - Weekly risk reports - Executive summaries

Workflow 5: Intervention Playbooks

Systematic approaches to save at-risk customers

1. Intervention Matching

1. Save Play Execution

- Diagnose root cause (don't assume) - Match intervention to cause - Assign owner and resources - Set clear timeline and milestones - Track outcome (saved, lost, reason)

1. Intervention Tactics

- Urgent Call: Same-day executive outreach - Health Check: Comprehensive account review - Training Blitz: Intensive enablement sessions - Success Sprint: Focused value delivery - Executive Alignment: VP/C-level engagement - Commercial Discussion: Pricing/terms adjustment

1. Outcome Tracking

- Save rate by risk level - Save rate by intervention type - Time from intervention to resolution - Reasons for unsuccessful saves - Long-term retention of saved accounts

Quick Reference

Best Practices

Signal Selection

• Focus on behaviors you can observe
• Validate correlation with actual churn
• Use leading indicators (not lagging)
• Combine multiple signal types
• Weight by predictive power

Scoring Model

• Start simple, add complexity gradually
• Calibrate weights with historical data
• Validate with blind holdout testing
• Recalibrate quarterly
• Document methodology

Alert Design

• Don't alert on every score change
• Focus on actionable thresholds
• Include context in alerts
• Route to right person
• Avoid alert fatigue

Intervention

• Diagnose before prescribing
• Match intervention to root cause
• Set clear success criteria
• Track outcomes rigorously
• Learn from failures

Model Maintenance

• Review accuracy monthly
• Retrain with new churn data
• Adjust for product changes
• Update as customer base evolves
• Document false positives/negatives

Churn Signals Library

Usage Signals

Engagement Signals

Support Signals

Relationship Signals

Risk Report Template

Weekly At-Risk Summary

# Churn Risk Report: Week of [Date]

## Summary
- Accounts at elevated risk or above: [X]
- Total ARR at risk: $[Amount]
- New alerts this week: [X]
- Risk trending up: [X accounts]
- Risk trending down: [X accounts]

## Critical Risk (81-100)
| Account | ARR | Score | Key Signals | Owner | Action |
|---------|-----|-------|-------------|-------|--------|
| [Name] | $X | 87 | [Signals] | [CSM] | [Status] |

## High Risk (61-80)
[Same format]

## Elevated Risk (41-60)
[Same format]

## Interventions in Progress
| Account | Started | Intervention | Progress |
|---------|---------|--------------|----------|
| [Name] | [Date] | [Type] | [Status] |

## Outcomes This Week
- Saved: [X accounts, $ARR]
- Lost: [X accounts, $ARR, reasons]
- De-escalated: [X accounts]

Red Flags

• Model overfit: Perfect on training data, poor on new data
• Signal lag: Indicators trigger too late for intervention
• False positive fatigue: Too many alerts that aren't real risk
• Missing signals: Key churn predictors not tracked
• Score opacity: Team doesn't understand why scores change
• Intervention mismatch: Same playbook for different problems
• No feedback loop: Not learning from save attempts
• Data quality: Missing or stale underlying data

Model Validation Metrics