Churn Prevention System Design

Purpose

This recipe builds an end-to-end churn prevention system: a multi-dimensional customer health score model, automated intervention triggers tied to score thresholds, a library of save plays mapped to churn reasons, and an involuntary churn recovery pipeline. The output is a live scoring system that flags at-risk accounts 30-90 days before cancellation and routes them to the correct intervention, reducing gross churn by 20-40%. [src1]

Prerequisites

• Product telemetry data (minimum 3 months) — event-level usage data including logins, feature usage, session depth
• Customer master list with contract values, renewal dates, segment tags from CRM/billing system
• Support ticket history from helpdesk platform — ticket volume, categories, resolution times, CSAT ratings
• NPS or CSAT survey data — at least one survey cycle completed (quarterly preferred)
• Payment/billing data — failed payment history, retry outcomes, card expiry dates

Constraints

• Health scores must be segment-specific — a single model across SMB and enterprise produces misleading signals. [src2]
• Minimum 50 churned accounts needed in historical data for statistically meaningful pattern identification. [src4]
• Score recalibration required every 90 days — product and market changes alter healthy usage patterns. [src5]
• Involuntary churn (payment failures) must comply with PCI DSS — never store raw card numbers in CS tooling. [src6]
• Save plays involving discounts require finance approval for offers exceeding 20% of ACV.

Tool Selection Decision

Which path?
├── No CS platform AND < 200 customers
│   └── PATH A: Spreadsheet Model — Google Sheets + manual triggers
├── No CS platform AND 200+ customers
│   └── PATH B: SQL + BI Tool — custom scoring via database queries + Metabase/Looker
├── CS platform available AND < 1000 customers
│   └── PATH C: CS Platform Standard — Gainsight/ChurnZero/Vitally native scoring
└── CS platform available AND 1000+ customers
    └── PATH D: CS Platform + ML — platform scoring augmented with predictive models

Execution Flow

Step 1: Map Leading Indicators by Churn Reason

Duration: 4-8 hours · Tool: Spreadsheet or whiteboard

Analyze historical churn data to identify the top 5-7 reasons customers leave. For each reason, map the behavioral signals that preceded cancellation 30-90 days before the churn event. [src1]

USAGE SIGNALS (leading by 60-90 days):
- Login frequency decline > 40% vs. prior 30-day average
- Core feature adoption drop
- DAU/WAU ratio falling below 0.3
- Session depth declining
- Seat utilization below 50% of licensed seats

ENGAGEMENT SIGNALS (leading by 30-60 days):
- CSM email response time increasing > 48 hours
- Skipped QBRs or check-in meetings
- Champion/admin user goes inactive
- No executive sponsor engagement in 90+ days

SUPPORT SIGNALS (leading by 30-45 days):
- Support ticket volume spike > 2x baseline
- Repeat tickets on same issue
- CSAT on tickets dropping below 3.5/5

FINANCIAL SIGNALS (leading by 15-30 days):
- Payment failures
- Downgrade inquiries or seat reductions
- Requesting cancellation policy

SENTIMENT SIGNALS (leading by 30-90 days):
- NPS score drop > 2 points between surveys
- Detractor NPS score (0-6)
- Negative review sentiment

Verify: At least 5 distinct leading indicators identified per churn reason, confirmed against historical data. · If failed: Conduct 10-15 exit interviews with recently churned customers.

Step 2: Build the Health Score Model

Duration: 1-2 days · Tool: CS platform or spreadsheet

Construct a weighted composite score (0-100) across four dimensions. Weight allocation should reflect predictive power validated against historical churn data. [src2] [src5]

HEALTH SCORE MODEL (0-100):
Dimension 1: Product Engagement (weight: 35%)
  - Login frequency, core feature adoption, breadth, session depth

Dimension 2: Relationship Quality (weight: 25%)
  - CSM engagement, executive sponsor activity, QBR attendance

Dimension 3: Support Health (weight: 20%)
  - Ticket volume trend (inverse), resolution satisfaction, escalation frequency

Dimension 4: Outcome & Sentiment (weight: 20%)
  - NPS/CSAT, stated ROI achievement, renewal likelihood, growth trajectory

COMPOSITE = (D1 × 0.35) + (D2 × 0.25) + (D3 × 0.20) + (D4 × 0.20)

Verify: Back-test against 6 months of data — churned accounts should score below 40 at least 30 days before cancellation in >70% of cases. · If failed: Adjust dimension weights until back-test accuracy exceeds 70%.

Step 3: Set Intervention Triggers and Thresholds

Duration: 4-8 hours · Tool: CS platform automation

Define four health zones with trigger rules and response SLAs. [src2]

THRIVING (81-100): Move to expansion pipeline. SLA: 14 days.
HEALTHY (61-80): Standard cadence. Monthly check-in, quarterly QBR.
AT-RISK (41-60): Activate save play. SLA: outreach within 48 hours.
CRITICAL (0-40): Executive save play. SLA: outreach within 24 hours.

VELOCITY TRIGGERS (override zone-based):
- Score drops > 20 points in 7 days → immediate Critical treatment
- Champion user goes inactive → At-Risk regardless of score
- Payment failure → separate involuntary churn workflow

Verify: Triggers would have fired ≥30 days before actual churn for >60% of historical churned accounts. · If failed: Lower thresholds or add velocity triggers.

Step 4: Design Save Plays by Churn Reason

Duration: 1-2 days · Tool: Document + CS platform playbook builder

Create a save play for each top churn reason with trigger, owner, actions, timeline, and escalation path. [src1] [src4]

PLAY 1: Low Product Adoption
  Trigger: Product Engagement < 30
  Actions: Re-onboarding, best-practice workflows, solutions engineer sprint
  Success: Product Engagement > 50 within 30 days

PLAY 2: Champion/Sponsor Departure
  Trigger: Admin inactive > 14 days or marked as departed
  Actions: New stakeholder identification, exec-to-exec outreach, ROI report
  Success: New champion engaged within 21 days

PLAY 3: Support Frustration
  Trigger: Support Health < 30 or 3+ escalations in 30 days
  Actions: Ticket review, engineering escalation, executive apology, service credit
  Success: All critical tickets resolved, CSAT > 4.0 within 14 days

PLAY 4: Price/Value Mismatch
  Trigger: Downgrade inquiry or "too expensive" feedback
  Actions: Value audit, ROI report, right-sized plan, retention offer
  Success: Customer confirms renewal or accepts adjusted plan

PLAY 5: Competitive Threat
  Trigger: Competitor mention or G2 comparison activity
  Actions: Competitive intel, feature comparison, roadmap preview
  Success: Competitor evaluation abandoned within 21 days

Verify: At least 5 save plays covering the most common churn reasons, each with clear triggers and success metrics.

Step 5: Implement Involuntary Churn Prevention (Dunning)

Duration: 4-8 hours · Tool: Billing platform + dunning tool

Payment failures account for up to 48% of all churn. Build a multi-channel dunning sequence with intelligent retry logic. [src6]

DAY 0: Classify decline (soft vs. hard). Soft: auto-retry in 24h.
DAY 1: Pre-dunning email with one-click payment update link.
DAY 3: Smart retry #2. If recovered, exit sequence.
DAY 5: In-app notification + email #2.
DAY 7: Smart retry #3 (final automated attempt).
DAY 10: Urgency email + SMS. Offer alternative payment methods.
DAY 14: Downgrade to free tier or pause (do NOT cancel).
DAY 30: Win-back email highlighting missed features.
DAY 60: Final win-back with incentive for high-value accounts.

Verify: Dunning recovers >50% of failed payments within 14 days. · If failed: Add SMS channel, test alternative retry timing, or switch to AI-powered retry engine.

Step 6: Build the Monitoring Dashboard

Duration: 4-8 hours · Tool: CS platform dashboard or BI tool

Build a live dashboard with health distribution, at-risk account queue, intervention effectiveness, churn metrics, leading indicator trends, and dunning recovery rates.

Verify: Dashboard updates daily. All four health zones represented with accurate counts. · If failed: Check data pipeline connections between analytics, CRM, and CS platform.

Output Schema

{
  "output_type": "churn_prevention_system",
  "format": "JSON",
  "columns": [
    {"name": "account_id", "type": "string", "required": true},
    {"name": "health_score", "type": "number", "description": "Composite 0-100", "required": true},
    {"name": "health_zone", "type": "string", "description": "thriving|healthy|at_risk|critical", "required": true},
    {"name": "product_engagement_score", "type": "number", "required": true},
    {"name": "relationship_quality_score", "type": "number", "required": true},
    {"name": "support_health_score", "type": "number", "required": true},
    {"name": "outcome_sentiment_score", "type": "number", "required": true},
    {"name": "score_trend", "type": "string", "description": "improving|stable|declining", "required": true},
    {"name": "active_save_play", "type": "string", "required": false},
    {"name": "renewal_date", "type": "date", "required": true},
    {"name": "arr", "type": "number", "required": true}
  ],
  "sort_order": "health_score ascending",
  "deduplication_key": "account_id"
}

Quality Benchmarks

If below minimum: Re-examine dimension weights. Most common cause is over-weighting lagging indicators (NPS) vs. leading indicators (usage trends). [src5]

Error Handling

Cost Breakdown

Anti-Patterns

Wrong: Single Universal Health Score for All Segments

Applying the same scoring model to SMB and enterprise accounts produces misleading results. Segment-specific scores achieve 15-20% higher prediction accuracy. [src1]

Correct: Segment-Specific Models

Build separate scoring models or normalization baselines per segment. Enterprise weights relationship quality and outcome metrics higher; SMB weights product engagement higher.

Wrong: Reacting to Score Drops Without Root Cause Analysis

Triggering save plays based on score drops alone leads to generic outreach that customers ignore. A drop from 75 to 55 could mean lost champion, product bug, or budget freeze. [src2]

Correct: Diagnose Before Intervening

When a score drops into At-Risk, examine which dimension(s) drove the decline. Route to the specific save play matching the root cause.

Wrong: Treating Involuntary Churn as Voluntary

Sending "we're sorry to see you go" messaging to customers with expired credit cards destroys trust. Payment failures account for up to 48% of churn and are almost entirely preventable. [src6]

Correct: Separate Involuntary Churn Pipeline

Route all payment failures to the dunning workflow. Tone should be helpful and transactional. Offer multiple payment alternatives with frictionless update links.

Churn Rate Benchmarks (2025-2026)

Median B2B SaaS annual churn is 3.5% for companies above $10M ARR. [src3] [src4]

When This Matters

Use this recipe when a SaaS company has at least 50 customers, 3+ months of product telemetry, and is experiencing churn above segment benchmarks — or wants to build proactive retention infrastructure before churn becomes a crisis. This produces a working health scoring system, intervention playbooks, and dunning sequences that reduce churn within 90 days.