Churn Risk Predictor

Churn prediction

by LeadMagic · MIT license · ★ 62 on GitHub

Builds churn prediction: leading indicators, account risk scores, early-warning alerts, and intervention playbooks tied to each risk level. Use when the user wants to predict which accounts will churn, score churn risk, or set up a…

---
name: churn-prediction
description: >-
  Builds churn prediction: leading indicators, account risk scores,
  early-warning alerts, and intervention playbooks tied to each risk level. Use
  when the user wants to predict which accounts will churn, score churn risk, or
  set up a retention early-warning system. Triggers on: "churn prediction",
  "predict churn", "churn model", "early warning", "risk scoring".
license: MIT
compatibility: Claude Code, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode, Goose, Hermes, Windsurf, Zed
metadata:
  version: "1.0.0"
  author: LeadMagic
  category: lifecycle
  tags: [lifecycle, churn-prediction, risk-scoring, retention, analytics]
  frameworks:
    - "Gainsight — Customer health scores"
    - "Lincoln Murphy — Desired Outcome"

Overview

Retention (stage 6). Canonical index → references/gtm-lifecycle-stages.md. Metrics → references/lifecycle-metrics-by-stage.md (Retention). Scorecard → the gtm-metrics skill (stage-health-scorecard.md) (Retention panel).

Manual install (npx, ZIP)

Claude Codenpx skills add LeadMagic/gtm-skills --skill churn-prediction -g -a claude-code

Claude Code (degit)npx degit LeadMagic/gtm-skills/skills/lifecycle/churn-prediction#main ~/.claude/skills/churn-prediction

Claude.ai / desktopZip the folder and upload it under Settings → Capabilities → Skills.

Codex, Cursor, other agentsPoint them at SKILL.md. The other files are plain text.

Connects with Agent Alley MCP · agentalley.io/mcp

Files included

SKILL.md7.5 KB

Read SKILL.md here

LeadMagic/main1 file shown

SKILL.md7.5 KB

Browse every file on GitHub↗

SKILL.md

Show the full text172 lines

churn-prediction/SKILL.md172 lines · 7.5 KB

PreviewCode

Outline

RawView on GitHub

Churn Prediction

Overview

Churn prediction shifts retention from reactive ("they cancelled — now what?") to proactive ("this account shows 3 risk signals — intervene now"). Early intervention reduces churn by 20-40% compared to waiting for cancellation. This skill covers modeling leading indicators, building risk scores, and designing intervention playbooks.

Authoritative Foundations

Gainsight — Customer health scores — Combine product usage, support signals, relationship depth, and sentiment into a weighted score with clear red, yellow, and green thresholds.
Lincoln Murphy — Desired Outcome — Customers churn when they stop reaching their desired outcome; predict churn from outcome progress, not logins alone.
Patrick Campbell (ProfitWell) — Voluntary vs involuntary churn — Model failed-payment churn separately; it has different leading indicators and fixes than dissatisfaction churn.

When to Use

"Build a churn prediction model"
"Predict which accounts will churn"
"Churn risk scoring"
"Early warning system"
"Retention analytics"

Lifecycle Stage

Retention (stage 6). Canonical index → references/gtm-lifecycle-stages.md.
Metrics → references/lifecycle-metrics-by-stage.md (Retention).
Scorecard → the gtm-metrics skill (stage-health-scorecard.md) (Retention panel).

Core Principle

The best churn signal is not "they stopped paying" — it's the behavior pattern they exhibit 60-90 days before they stop paying. Your job is to find those leading indicators and act on them while there's still time.

Step-by-Step Process

Phase 1: Identify Leading Indicators

Analyze churned customers looking back 90 days before cancellation:

Product Usage Signals:

Login frequency declining (weekly → bi-weekly → monthly)
Key feature usage dropping (>30% decline month-over-month)
Time since last key action > 14 days
Multiple users stopped logging in (if multi-seat)
Failed imports/integrations (3+ failures = high frustration)

Engagement Signals:

Stopped opening emails (unengaged for 30+ days)
Support ticket sentiment turning negative
No response to CSM outreach (2+ attempts)
Missed QBR or success review meeting
Stopped attending webinars/events they previously attended

Business Signals:

Champion left the company (detected via LinkedIn)
Company had layoffs or restructuring
Budget cut or freeze in their department
New competitor evaluation (detected via intent data)
Contract renewal window approaching with no engagement

Phase 2: Build Risk Scoring Model

Weight each signal on a 0-100 risk score:

Critical (40 points max):

Champion departure: 40 points
Key feature usage down >50%: 35 points
No login in 30+ days: 30 points

High (25 points max):

Support ticket spike (3+ in 7 days): 20 points
No response to CSM in 14+ days: 15 points
Missed renewal meeting: 10 points

Medium (20 points max):

Email engagement dropped to 0: 10 points
Usage declining 20-30% MoM: 10 points
New competitor intent signal: 5 points (cumulative)

Low (15 points max):

Weather signal: minor usage dip, one missed meeting

Risk Tiers:

Red (70+): Immediate executive intervention required. Cancel risk in <30 days.
Yellow (40-69): CSM intervention this week. Cancel risk in 30-60 days.
Green (<40): Standard monitoring.

Phase 3: Intervention Playbooks

Red Account Playbook:

Day 0: CSM calls within 4 hours. "I noticed [specific signal]. Everything ok?"
Day 0: Executive sponsor reaches out. "Your success is critical to us — let's solve whatever is blocking you."
Day 1: Root cause analysis. Is it product, support, business, or competition?
Day 3: Action plan presented to customer. Specific timeline. Named owner.
Day 7: Check-in. "Is the plan working? What else do you need?"
Day 14: Success review. Have the signals reversed? If not, escalate to VP CS.

Yellow Account Playbook:

Day 0: CSM email: "Noticed [signal] — quick check-in?"
Day 2: If no reply, call.
Day 5: Value reinforcement: usage stats, ROI summary, new features they should try
Day 10: If signals persist, escalate to Red playbook

Phase 4: Automated Monitoring

Daily scan: Run risk model on all accounts every 24 hours
Alert routing: Red → CSM + VP CS + Account Executive. Yellow → CSM only.
Slack/email alerts: "Account ABC moved to RED risk — champion departed, usage down 60%."
Dashboard: Real-time risk heatmap of all accounts. Sort by risk score.
Trend tracking: Is risk score improving or worsening week-over-week?

Phase 5: Feedback Loop

Churn post-mortem: For every churned account, review: what signals were present 90 days before churn? Why weren't they acted on? What signal did we miss?
Model refinement: Every quarter, update signal weights based on which signals actually predicted churn vs false positives
False positive analysis: Accounts flagged red that didn't churn — what made them different? Refine model.
Precision vs recall balance: High precision = fewer false alarms but may miss churn. High recall = catch more churn but more false alarms. Tune based on CSM capacity.

Output Format

Churn prediction model with: leading indicator catalog, risk scoring algorithm, intervention playbooks, automated monitoring setup, and feedback loop.

Quality Check

Before delivering, verify:

All required sections are complete
Output matches the user's stated need
Named frameworks are cited for key recommendations
No vague claims — every recommendation has a specific action
Deliverable is ready for operational use, not just conceptual

Common Pitfalls

Skipping research. Building output without understanding the specific context. Fix: always gather required inputs before producing deliverables.
Generic output. "Improve your process" without concrete steps. Fix: every recommendation must include a specific action, timeline, and owner.
Missing framework citations. Advice without named authorities. Fix: ground every recommendation in a cited framework from a recognized authority.

Execution Artifacts

references/framework-notes.md — Named frameworks and reference tables
templates/output-template.md — Deliverable shell for agent output
scripts/check-output.py — Lightweight deliverable validator Canonical lifecycle: references/gtm-lifecycle-stages.md (Retention) · references/lifecycle-metrics-by-stage.md · the gtm-metrics skill (stage-health-scorecard.md)

churn-prevention, cs-playbooks, onboarding-sequences, lifecycle-drips, expansion-selling

1	---
2	name: churn-prediction
3	description: >-
4	Builds churn prediction: leading indicators, account risk scores,
5	early-warning alerts, and intervention playbooks tied to each risk level. Use
6	when the user wants to predict which accounts will churn, score churn risk, or
7	set up a retention early-warning system. Triggers on: "churn prediction",
8	"predict churn", "churn model", "early warning", "risk scoring".
9	license: MIT
10	compatibility: Claude Code, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode, Goose, Hermes, Windsurf, Zed
11	metadata:
12	version: "1.0.0"
13	author: LeadMagic
14	category: lifecycle
15	tags: [lifecycle, churn-prediction, risk-scoring, retention, analytics]
16	frameworks:
17	- "Gainsight — Customer health scores"
18	- "Lincoln Murphy — Desired Outcome"
19	- "Patrick Campbell (ProfitWell) — Voluntary vs involuntary churn"
20	---
21
22	# Churn Prediction
23
24	## Overview
25	Churn prediction shifts retention from reactive ("they cancelled — now what?")
26	to proactive ("this account shows 3 risk signals — intervene now"). Early
27	intervention reduces churn by 20-40% compared to waiting for cancellation.
28	This skill covers modeling leading indicators, building risk scores, and
29	designing intervention playbooks.
30
31	## Authoritative Foundations
32	- Gainsight — Customer health scores — Combine product usage, support signals, relationship depth, and sentiment into a weighted score with clear red, yellow, and green thresholds.
33	- Lincoln Murphy — Desired Outcome — Customers churn when they stop reaching their desired outcome; predict churn from outcome progress, not logins alone.
34	- Patrick Campbell (ProfitWell) — Voluntary vs involuntary churn — Model failed-payment churn separately; it has different leading indicators and fixes than dissatisfaction churn.
35
36	## When to Use
37	- "Build a churn prediction model"
38	- "Predict which accounts will churn"
39	- "Churn risk scoring"
40	- "Early warning system"
41	- "Retention analytics"
42
43	## Lifecycle Stage
44
45	Retention (stage 6). Canonical index → `references/gtm-lifecycle-stages.md`.
46	Metrics → `references/lifecycle-metrics-by-stage.md` (Retention).
47	Scorecard → the `gtm-metrics` skill (stage-health-scorecard.md) (Retention panel).
48
49	## Core Principle
50	> The best churn signal is not "they stopped paying" — it's the behavior
51	> pattern they exhibit 60-90 days before they stop paying. Your job is to
52	> find those leading indicators and act on them while there's still time.
53
54	## Step-by-Step Process
55
56	### Phase 1: Identify Leading Indicators
57	Analyze churned customers looking back 90 days before cancellation:
58
59	Product Usage Signals:
60	- Login frequency declining (weekly → bi-weekly → monthly)
61	- Key feature usage dropping (>30% decline month-over-month)
62	- Time since last key action > 14 days
63	- Multiple users stopped logging in (if multi-seat)
64	- Failed imports/integrations (3+ failures = high frustration)
65
66	Engagement Signals:
67	- Stopped opening emails (unengaged for 30+ days)
68	- Support ticket sentiment turning negative
69	- No response to CSM outreach (2+ attempts)
70	- Missed QBR or success review meeting
71	- Stopped attending webinars/events they previously attended
72
73	Business Signals:
74	- Champion left the company (detected via LinkedIn)
75	- Company had layoffs or restructuring
76	- Budget cut or freeze in their department
77	- New competitor evaluation (detected via intent data)
78	- Contract renewal window approaching with no engagement
79
80	### Phase 2: Build Risk Scoring Model
81	Weight each signal on a 0-100 risk score:
82
83	Critical (40 points max):
84	- Champion departure: 40 points
85	- Key feature usage down >50%: 35 points
86	- No login in 30+ days: 30 points
87
88	High (25 points max):
89	- Support ticket spike (3+ in 7 days): 20 points
90	- No response to CSM in 14+ days: 15 points
91	- Missed renewal meeting: 10 points
92
93	Medium (20 points max):
94	- Email engagement dropped to 0: 10 points
95	- Usage declining 20-30% MoM: 10 points
96	- New competitor intent signal: 5 points (cumulative)
97
98	Low (15 points max):
99	- Weather signal: minor usage dip, one missed meeting
100
101	Risk Tiers:
102	- Red (70+): Immediate executive intervention required. Cancel risk in <30 days.
103	- Yellow (40-69): CSM intervention this week. Cancel risk in 30-60 days.
104	- Green (<40): Standard monitoring.
105
106	### Phase 3: Intervention Playbooks
107
108	Red Account Playbook:
109	- Day 0: CSM calls within 4 hours. "I noticed [specific signal]. Everything ok?"
110	- Day 0: Executive sponsor reaches out. "Your success is critical to us — let's
111	solve whatever is blocking you."
112	- Day 1: Root cause analysis. Is it product, support, business, or competition?
113	- Day 3: Action plan presented to customer. Specific timeline. Named owner.
114	- Day 7: Check-in. "Is the plan working? What else do you need?"
115	- Day 14: Success review. Have the signals reversed? If not, escalate to VP CS.
116
117	Yellow Account Playbook:
118	- Day 0: CSM email: "Noticed [signal] — quick check-in?"
119	- Day 2: If no reply, call.
120	- Day 5: Value reinforcement: usage stats, ROI summary, new features they should try
121	- Day 10: If signals persist, escalate to Red playbook
122
123	### Phase 4: Automated Monitoring
124	- Daily scan: Run risk model on all accounts every 24 hours
125	- Alert routing: Red → CSM + VP CS + Account Executive. Yellow → CSM only.
126	- Slack/email alerts: "Account ABC moved to RED risk — champion departed,
127	usage down 60%."
128	- Dashboard: Real-time risk heatmap of all accounts. Sort by risk score.
129	- Trend tracking: Is risk score improving or worsening week-over-week?
130
131	### Phase 5: Feedback Loop
132	- Churn post-mortem: For every churned account, review: what signals were
133	present 90 days before churn? Why weren't they acted on? What signal did we miss?
134	- Model refinement: Every quarter, update signal weights based on which
135	signals actually predicted churn vs false positives
136	- False positive analysis: Accounts flagged red that didn't churn — what
137	made them different? Refine model.
138	- Precision vs recall balance: High precision = fewer false alarms but
139	may miss churn. High recall = catch more churn but more false alarms.
140	Tune based on CSM capacity.
141
142	## Output Format
143	Churn prediction model with: leading indicator catalog, risk scoring algorithm,
144	intervention playbooks, automated monitoring setup, and feedback loop.
145
146
147
148	## Quality Check
149
150	Before delivering, verify:
151	- [ ] All required sections are complete
152	- [ ] Output matches the user's stated need
153	- [ ] Named frameworks are cited for key recommendations
154	- [ ] No vague claims — every recommendation has a specific action
155	- [ ] Deliverable is ready for operational use, not just conceptual
156
157	## Common Pitfalls
158
159	1. Skipping research. Building output without understanding the specific context. Fix: always gather required inputs before producing deliverables.
160	2. Generic output. "Improve your process" without concrete steps. Fix: every recommendation must include a specific action, timeline, and owner.
161	3. Missing framework citations. Advice without named authorities. Fix: ground every recommendation in a cited framework from a recognized authority.
162
163	## Execution Artifacts
164
165	- `references/framework-notes.md` — Named frameworks and reference tables
166	- `templates/output-template.md` — Deliverable shell for agent output
167	- `scripts/check-output.py` — Lightweight deliverable validator
168	Canonical lifecycle: `references/gtm-lifecycle-stages.md` (Retention) · `references/lifecycle-metrics-by-stage.md` · the `gtm-metrics` skill (stage-health-scorecard.md)
169
170	## Related Skills
171	- churn-prevention, cs-playbooks, onboarding-sequences, lifecycle-drips, expansion-selling
172