thesis
SaaS Pillar III: Scaled Prevention (Closing the Prediction-Action Gap)
SaaS-specific application of predictive resilience focused on closing the gap between prediction and action. Addresses the SaaS challenge that prediction isn't the problem—many have churn models—but intervention at scale doesn't happen. Enables product-embedded prevention and automated intervention.
Executive Summary
The Shift
From prediction without action (model predicts; humans can't respond) → Scaled prevention (prediction triggers automated intervention)
Core Premise
Most mature SaaS companies already have churn prediction models. The prediction isn't the problem—the action is. When your model flags 500 at-risk accounts and your CS team has capacity for 50, prediction without action creates frustration, not prevention. The breakthrough is closing the prediction-action gap through automated intervention, product-embedded prevention, and AI-powered engagement at scale.
Economic Outcome
- Churn reduction of 20-40% through intervention at scale
- Gross margin improvement (prevention costs less than recovery)
- Infrastructure incidents reduced 50%+ (predictive operations)
- CSM time redirected from firefighting to expansion
Why Now
Three shifts have made scaled prevention achievable:
- Intervention channels matured — In-app messaging, email sequences, product prompts
- AI-powered engagement — Automated outreach that doesn't feel robotic
- Product-embedded intelligence — Prevention becomes product capability, not bolt-on
Who This Is For
- SaaS companies with prediction models but limited intervention capacity
- Companies where churn is concentrated in predictable patterns
- Products with engagement channels (in-app, email, etc.) for intervention
- Teams frustrated by "we saw it coming but couldn't stop it"
Who This Is NOT For
- Companies without any prediction capability (build that first)
- Products with no engagement channels
- Churn that's truly random (no pattern to intervene on)
- Very early stage (<100 customers; manual intervention still works)
The Strategic Argument
The Problem: The Prediction-Action Gap
SaaS companies have invested heavily in prediction. The problem is what happens next.
The Math That Doesn't Work:
Churn model predicts: 500 at-risk accounts this quarter
CS team capacity: 50 high-touch interventions/quarter
Coverage gap: 450 accounts predicted to churn with no intervention
Result: Good prediction, poor prevention
The Pattern:
PREDICTION INVESTMENT ACTION CAPACITY
ML Model ─────────────► CSM Reviews
Data Team (Weekly)
Training │
Validation ▼
Health Score Prioritizes
│ Top 10%
│ │
▼ ▼
500 At-Risk ─────────► 50 Get Help
Accounts 450 Ignored
The Symptoms:
| Symptom | What It Looks Like | Root Cause |
|---|---|---|
| Model exists, churn persists | "Our health scores are accurate but churn is flat" | Action doesn't scale with prediction |
| CSM frustration | "I can see the at-risk list but can't get to everyone" | Human intervention bottleneck |
| Prediction fatigue | "Another dashboard we don't have time to use" | Insight without action |
| Last-minute saves | "We caught it, but barely" | Reactive intervention, not proactive |
| Bolt-on analytics | Risk dashboards separate from product | Prevention not embedded |
Why It Persists:
- Prediction is a data team project; action is a CS team execution
- No organizational muscle for automated intervention
- Fear of automating customer outreach ("will it feel robotic?")
- Product and CS operate as separate functions
The Traditional Response: Why It Failed
SaaS companies have tried three approaches to prevention:
Approach 1: Better Prediction Models
- What they tried: More features in the model; more sophisticated ML; better accuracy
- Why it failed: 85% accuracy vs. 80% accuracy doesn't matter if you can only act on 10%
- The gap: Prediction is sufficient; action is the constraint
Approach 2: Hire More CSMs
- What they tried: Scale intervention capacity through headcount
- Why it failed: Economics don't work; can't afford 1:20 ratios across segments
- The gap: Human intervention doesn't scale with customer growth
Approach 3: Tiered Response (Segment and Abandon)
- What they tried: High-touch for enterprise, self-serve for SMB
- Why it failed: Abandoned segments churn; prediction mocks you by showing what you'll lose
- The gap: Segmentation optimizes scarcity, doesn't solve it
The Common Failure Mode: All three approaches accept that intervention requires humans and optimize around human scarcity. They don't break the constraint—they work within it.
The AI Unlock: What Changed
Three capabilities converged to close the prediction-action gap:
1. Automated Contextual Intervention
- AI can initiate outreach that's personalized to customer situation
- Not generic drip campaigns—contextual based on usage, risk signals, history
- Email, in-app, SMS—right channel, right message, right time
- Human quality at machine scale
2. Product-Embedded Prevention
- Prevention built into product experience, not separate CS motion
- At-risk users see different onboarding, prompts, guidance
- Product adapts to prevent failure before CS knows there's a problem
- Prevention as feature, not process
3. Intelligent Escalation
- AI handles routine interventions; surfaces exceptions to humans
- Human time focused on complex cases and high-value accounts
- Clear handoff protocols between AI and CSM
- CSMs elevated to strategic, not drowned in volume
The New Model:
PREDICTION ACTION AT SCALE
Health Score ──────────► AI-Initiated Intervention
│ │
│ ├── Low Risk: Nurture sequence
│ │
│ ├── Medium Risk: Proactive outreach
│ │
│ ├── High Risk: Intensive engagement
│ │
▼ ▼
All At-Risk ──────────► All Get Intervention
Accounts (Right level, right channel)
│ │
│ ▼
└────────────────► Humans for Exceptions
(Complex, high-value, escalated)
The Principle: Prediction creates the list. Automation works the list. Humans work the exceptions.
The Transformation Path
Scaled prevention develops through three parallel capabilities:
Track A: Automated Intervention
Moving from human-dependent to automated response.
Stage A1: Triggered Outreach (Months 1-3)
- Risk signal triggers automated email sequence
- Templates personalized by segment and signal type
- CSM notified; AI handles initial outreach
- Focus: Coverage—ensure every at-risk account hears from you
Stage A2: Contextual Engagement (Months 3-6)
- AI outreach personalized to specific risk signal
- "We noticed you haven't used feature X recently..."
- Dynamic content based on usage pattern
- Focus: Relevance—right message for right situation
Stage A3: Conversational Intervention (Months 6-12)
- AI handles two-way conversation for routine cases
- Can answer questions, schedule calls, provide guidance
- Escalates to human for complex issues
- Focus: Resolution—AI completes intervention, not just initiates
Track B: Product-Embedded Prevention
Building prevention into the product itself.
Stage B1: Risk-Aware UX (Months 2-4)
- At-risk users see different prompts, guidance, onboarding
- "Looks like you haven't completed setup—here's help"
- Product knows user is at risk; adapts experience
- Focus: Prevention before outreach needed
Stage B2: Proactive Feature Nudges (Months 4-8)
- AI identifies features user should adopt but hasn't
- In-app prompts, tooltips, guided workflows
- "Customers like you who use X see better outcomes"
- Focus: Adoption as prevention
Stage B3: Anticipatory Product (Months 8-14)
- Product anticipates needs and removes friction
- Proactively surfaces help before user struggles
- Prevention invisible—user just succeeds
- Focus: Seamless success
Track C: Infrastructure Resilience (For SaaS with infrastructure component)
Preventing operational incidents.
Stage C1: Anomaly Detection (Months 1-4)
- Monitor infrastructure for warning signals
- Latency, error rates, capacity, performance degradation
- Alert before customer impact
Stage C2: Predictive Operations (Months 4-8)
- Predict incidents before they occur
- Auto-scaling based on predicted demand
- Proactive maintenance during low-impact windows
Stage C3: Self-Healing Systems (Months 8-14)
- Automated remediation for known issues
- Failover and recovery without human intervention
- Infrastructure that prevents its own failures
The End State: What Success Looks Like
Operational Reality:
- Every at-risk account receives intervention (not just top 10%)
- CSMs focus on complex cases and expansion; AI handles volume
- Product itself prevents failure for engaged users
- Infrastructure incidents rare; most prevented or auto-remediated
- "We saw it coming and couldn't stop it" → eliminated
Financial Reality:
- Churn reduced 20-40% (all segments)
- Gross margin improved (prevention < recovery cost)
- CAC payback shortened (longer customer lifetime)
- CSM productivity up 2-3x (same team, better outcomes)
Competitive Reality:
- Reputation for reliability and customer care
- Higher NPS/reviews from prevented failures
- "They caught problems before we knew" testimonials
- Lower churn = compounding advantage over leaky competitors
Evidence Base
Quantitative Research
Intervention Impact:
- Proactive intervention on at-risk accounts reduces churn 25-40% (Gainsight benchmark data)
- In-app guidance increases feature adoption 30-50% (Pendo studies)
- Automated engagement achieves 60-70% of human engagement impact at 10% of cost (industry benchmarks)
Product-Embedded Prevention:
- Users who complete onboarding churn at 60% lower rate (SaaS onboarding benchmarks)
- Proactive feature prompts increase adoption 40%+ vs. passive discovery (UserPilot data)
- Personalized in-app experience improves retention 20-30% (Appcues research)
Infrastructure Prediction:
- Predictive operations reduce incidents 50-70% (PagerDuty case studies)
- Mean time to recovery (MTTR) reduced 40-60% with AI-assisted ops
- Proactive scaling prevents 80%+ of capacity-related incidents
Case Studies
Case Study 1: Pendo (Product-Embedded Prevention)
- Context: Product experience platform; thousands of customers; helps others build in-app guidance
- Challenge: Needed to demonstrate scaled prevention on own customer base
- Intervention: Built risk-triggered in-app guidance for own product
- Implementation:
- Health score triggers different in-app experiences
- At-risk users see additional onboarding, prompts, value demonstrations
- Usage patterns trigger relevant feature guides
- CSM alerted only for accounts not responding to automated intervention
- Outcome:
- 30% of at-risk accounts saved through automated in-app intervention
- CSM time refocused to expansion and strategic accounts
- Product itself became prevention engine
- Became proof point for selling to others
- Relevance: Product-embedded prevention; SaaS using own medicine
Case Study 2: Gainsight (Intervention at Scale)
- Context: Customer success platform; mid-market and enterprise; owns the CS tech category
- Challenge: Own CS team faced same scaling challenges as customers
- Intervention: Built automated intervention triggered by health score signals
- Implementation:
- Health signals trigger automated email sequences (risk-specific content)
- CTA (call-to-action) actions monitored; non-response escalates
- CSMs receive pre-prioritized list with context
- Playbooks auto-execute for common scenarios
- Outcome:
- 100% of at-risk accounts receive intervention (vs. previous 20%)
- Churn in automated-intervention cohort down 25%
- CSM capacity expanded; same team handles 40% more accounts
- Response time to risk signal: 48 hours → 4 hours
- Relevance: Automated intervention at scale; closing the prediction-action gap
Case Study 3: PagerDuty (Infrastructure Prediction as Product)
- Context: Incident management platform; serves DevOps teams; reliability is the product
- Challenge: Customers wanted to prevent incidents, not just respond faster
- Intervention: Built predictive capabilities into incident management
- Implementation:
- AI analyzes incident patterns to predict future incidents
- Anomaly detection surfaces degradation before outage
- Runbook automation enables self-healing
- Prediction becomes core product capability
- Outcome:
- Customers using predictive features: 50% reduction in incidents
- MTTR reduced 40% with AI-assisted triage
- Prevention became primary sales differentiator
- Moved from "respond faster" to "prevent incidents"
- Relevance: Infrastructure prediction; prevention as product
Case Study 4: Intercom (Conversational Prevention)
- Context: Customer messaging platform; tens of thousands of customers; broad segment mix
- Challenge: At-risk customers needed intervention; couldn't scale human outreach
- Intervention: Deployed Fin (AI) for proactive risk engagement
- Implementation:
- Risk signals trigger AI-initiated outreach
- AI can hold conversation, answer questions, schedule human call if needed
- Personalized to customer's usage and stated concerns
- Human escalation for unresolved or complex
- Outcome:
- 70% of risk interventions handled by AI without human
- Customer satisfaction with AI intervention: 80%+ positive
- CSM time shifted to expansion conversations
- Churn in AI-intervention cohort: 20% lower than control
- Relevance: AI-powered conversational intervention; human-quality at scale
Case Study 5: Mid-Market Vertical SaaS—Learning Management Platform (Composite)
- Context: $18M ARR; 600 corporate customers; 4 CSMs; annual contracts
- Challenge: Churn concentrated in months 8-10 (pre-renewal); CSMs couldn't reach everyone
- Intervention: Deployed automated intervention system tied to health scores
- Implementation:
- Health model using: login frequency, course completion, admin engagement, support tickets
- At-risk triggers: automated email sequence + in-app prompts
- High-risk escalates to CSM with context package
- Product changes for at-risk users (simplified dashboards, guided next steps)
- Outcome:
- Churn reduced from 18% to 11% annually (39% improvement)
- CSM capacity: same 4 people now handle proactive work vs. firefighting
- Time-to-intervention: average 45 days → 7 days from first risk signal
- Customer feedback: "We felt supported even though we rarely talked to anyone"
- Investment: ~$90K (AI platform + integration + in-app changes)
- Payback: 3 months (from retained revenue)
- Relevance: Mid-market B2B SaaS; full prediction-to-action implementation
Implementation Framework
Prerequisites
Before pursuing scaled prevention, validate:
| Prerequisite | Why It Matters | How to Assess |
|---|---|---|
| Prediction capability exists | Need to know who's at risk | Do you have health scores or churn prediction? |
| Intervention channels available | Need way to reach at-risk customers | In-app messaging? Email? Product prompts? |
| Patterns in churn | Random churn can't be prevented | Do you know why customers churn? |
| CS team willing to change role | From intervener to exception handler | Will CSMs embrace AI-led intervention? |
| Risk signals identifiable | Prevention requires warning signs | Can you detect at-risk status 30+ days before churn? |
Red Flags (Proceed with Caution):
- No prediction capability (build that first)
- No intervention channels (can't reach customers)
- Churn appears random (no pattern to act on)
- CS team sees AI as threat (change management issue)
- Can't detect risk until too late (signals too late)
Phase 1: Intervention Infrastructure (Months 1-3)
Objectives:
- Connect prediction to action channels
- Launch first automated intervention
- Establish measurement baseline
Key Activities:
- Map health score signals to intervention triggers
- Design intervention playbooks (content, channel, timing)
- Build automation: risk signal → outreach sequence
- Establish control group for measurement
- Train CS team on new workflow (exceptions, escalation)
Success Criteria:
- 100% of at-risk accounts receiving automated intervention
- First intervention within 7 days of risk signal
- CSM escalation path working cleanly
- Baseline established for churn comparison
Phase 2: Contextual & Product-Embedded (Months 3-6)
Objectives:
- Improve intervention relevance
- Build product-embedded prevention
- Prove churn impact
Key Activities:
- Personalize interventions by risk signal type
- Build in-app guidance for at-risk users
- Implement proactive feature prompts
- Deploy conversational AI for two-way engagement
- Measure churn difference vs. control/baseline
Success Criteria:
- Intervention response rate >25%
- At-risk users see different product experience
- Measurable churn reduction (target: 15%+)
- AI handling 50%+ of routine interventions
Phase 3: Scaled Prevention System (Months 6-12)
Objectives:
- Prevention as standard operating model
- Expand beyond churn to all risk types
- Prove ROI and institutionalize
Key Activities:
- Expand to additional risk types (adoption, expansion, champion)
- Build self-improving intervention (learn from outcomes)
- Full integration: prediction → intervention → outcome → learning
- Document ROI and prove business case
- Extend to infrastructure resilience if applicable
Success Criteria:
- Churn reduced 25%+ vs. baseline
- CSM productivity 2x+ (same team, better outcomes)
- Multiple risk types covered by automated intervention
- Clear ROI documentation
Objection Handling
Objection 1: "We already have a churn model—prediction isn't our problem"
Who says it: Data-mature companies with existing prediction; frustrated CS leaders
Underlying concern: True—they've invested in prediction; problem is elsewhere
Response strategy: Agree; pivot to action as the constraint; position as complement to prediction investment
Response:
"Exactly right—and that's actually good news. Your prediction investment isn't wasted; it's underleveraged. The model tells you who's at risk. The question is: what happens next? If you can only intervene on 10-20% of flagged accounts, you're leaving 80% of your prediction value on the table. We help close that gap—take your prediction and connect it to action at scale. The model stays; the intervention capacity multiplies."
Probe questions:
- "Of the accounts your model flags as at-risk, what percentage get meaningful intervention?"
- "What happens to the accounts you can see are at-risk but can't get to?"
- "If you could intervene on all flagged accounts, how much churn would you prevent?"
Evidence to cite:
- Gainsight: 100% coverage (up from 20%) drove 25% churn reduction
- LMS case: same model, new intervention system → 39% churn improvement
Objection 2: "Customers will hate automated outreach—it'll feel robotic"
Who says it: Customer experience-focused leaders; companies with brand sensitivity
Underlying concern: Legitimate worry about customer experience; past bad automation
Response strategy: Show modern AI capability; distinguish from old automation; offer pilot
Response:
"Five years ago, you'd be right. Automated outreach meant generic emails that customers ignored or hated. Modern AI is different—it's contextual. 'We noticed you haven't logged in recently' is generic. 'We noticed you haven't used the reporting feature that helped your team hit goals last quarter—here's how to get started again' is helpful. The difference is AI that knows the customer's context and speaks to their specific situation. And the data shows: customers actually prefer timely, relevant help over waiting for an overloaded CSM to get to them."
Probe questions:
- "What's worse: relevant automated outreach or no outreach at all?"
- "If the message was genuinely helpful and contextual, would that change things?"
- "Could we pilot with a subset of accounts and measure customer response?"
Evidence to cite:
- Intercom: 80%+ positive satisfaction with AI intervention
- LMS case: "We felt supported even though we rarely talked to anyone"
Objection 3: "Our product is too complex for automated prevention"
Who says it: Enterprise SaaS; complex multi-stakeholder products; technical products
Underlying concern: Real complexity; fear that simple interventions won't work
Response strategy: Acknowledge complexity; show layered approach; AI for volume, humans for complexity
Response:
"Complex products actually need scaled prevention more, not less. Here's why: in simple products, customers figure it out or leave quickly. In complex products, they slowly disengage over months while your CSMs can't see it. The intervention doesn't have to solve the complex problem—it has to surface it. AI might detect that your champion hasn't logged in for 3 weeks and say 'Let's schedule a call with your success manager.' That's not robotic; that's appropriate escalation. AI handles detection and triage; humans handle the complex conversation."
Probe questions:
- "What percentage of your at-risk situations are actually complex vs. routine?"
- "Would it help if AI triaged and prioritized, even if humans handled complex cases?"
- "What if the intervention was 'let's talk' rather than 'here's a self-serve solution'?"
Evidence to cite:
- Layered model: AI for volume, humans for exceptions
- Even complex products have routine risk patterns
Objection 4: "We don't have the intervention channels—our product has no in-app messaging"
Who says it: Products without engagement layer; older tech stacks
Underlying concern: Real infrastructure gap; would require development
Response strategy: Show email-first approach; phased investment in product channels
Response:
"You don't need in-app messaging to start. Most intervention impact comes from email—reaching the right person with the right message at the right time. In-app is phase 2 once you've proven email intervention works. The sequence is: (1) email-based automated intervention, (2) measure impact, (3) invest in product-embedded prevention once ROI is clear. Many companies see significant churn reduction from email alone—it's not the channel, it's the timeliness and relevance."
Probe questions:
- "Could you reach at-risk customers via email effectively?"
- "What would it take to add basic in-app messaging in phase 2?"
- "Is channel the constraint, or is it knowing what to say and when?"
Evidence to cite:
- Email-first implementations show 60-70% of impact
- Product-embedded prevention as phase 2, not prerequisite
Success Metrics
Leading Indicators (0-6 months)
| Metric | Definition | Target | Warning Sign |
|---|---|---|---|
| Intervention Coverage | % of at-risk accounts receiving intervention | 100% | <50% |
| Time to Intervention | Days from risk signal to first outreach | <7 days | >30 days |
| Intervention Response Rate | % of automated outreach that gets engagement | >25% | <10% |
| AI Resolution Rate | % of interventions completed without human | >50% | <25% |
| Escalation Quality | CSM satisfaction with AI-surfaced exceptions | >4/5 | <3/5 |
Lagging Indicators (12-24 months)
| Metric | Definition | Target | Warning Sign |
|---|---|---|---|
| Churn Reduction | Change in churn rate vs. baseline | -25% or better | No change |
| Intervention ROI | Value of prevented churn ÷ intervention cost | >10:1 | <3:1 |
| CSM Productivity | Outcomes per CSM (saves, expansions) | 2x+ baseline | No improvement |
| Save Rate | % of at-risk accounts retained | >60% | <40% |
| Time to Save | Days from intervention to "saved" status | Decreasing | Increasing |
Warning Signs
Technical:
- Intervention triggers not firing reliably
- Email deliverability problems
- In-app prompts being dismissed/ignored
- AI responses generating negative feedback
Business:
- Churn not improving despite intervention
- Customers complaining about automation
- CSMs not trusting AI triage
- Save conversations happening too late
Pillar Interactions (SaaS Context)
How SaaS P3 Enables Other Pillars
P3 → P1 (Scaled Success):
- Prevention is proactive success
- Risk intervention IS the success motion for many accounts
- Saves create expansion opportunity moments
- Prevention frees CSMs for strategic success work
P3 → P2 (Platform Value):
- Risk patterns become data product (risk scoring for customers' customers)
- Prevention effectiveness data valuable for benchmarking
- Prediction infrastructure supports data products
- Resilience is proof of platform maturity
How Other Pillars Enable SaaS P3
P1 → P3:
- Success engagement generates risk signals
- CSM relationships enable intervention delivery
- Success programs create prevention touchpoints
- Health scoring requires success context
P2 → P3:
- Usage data powers prediction models
- Benchmark deviations surface risk
- Cross-customer patterns improve prediction accuracy
- Data infrastructure enables real-time detection
Sequencing Considerations for SaaS
P3 First (when it makes sense):
- Churn is the acute pain point
- Prediction exists but action doesn't
- NRR < 100% and declining
- "We can see it coming but can't stop it"
P3 After P1:
- Need success infrastructure for intervention delivery
- CSM relationships required for escalation
- Success programs create intervention channels
P3 After P2:
- Need usage data for prediction models
- Data infrastructure required for real-time detection
SaaS Segment Patterns
By Company Stage
| Stage | P3 Challenge | P3 Opportunity | Approach |
|---|---|---|---|
| Series A-B | Manual intervention still possible | Build prediction foundation | Light health scoring; founder-led saves |
| Series C | Scale breaking intervention capacity | Automated intervention MVP | Connect prediction to automated action |
| Series D+ | Sophisticated prediction, insufficient action | Full scaled prevention | Complete prediction-action infrastructure |
| Late/Pre-IPO | Churn hurting metrics narratives | Prove prevention capability | ROI documentation; predictable retention |
By Business Model
| Model | P3 Challenge | P3 Opportunity |
|---|---|---|
| PLG Self-Serve | No human intervention at all | Product-embedded prevention; in-app only |
| PLG + Sales | Hybrid intervention model | Automated for PLG; human for sales-led |
| Sales-Led SMB | Volume + economics challenge | Automated intervention essential; humans unaffordable |
| Sales-Led Mid-Market | Prediction-action gap classic case | Full scaled prevention implementation |
| Sales-Led Enterprise | Complex, human required | AI for triage and prep; humans for intervention |
By Risk Profile
| Risk Pattern | P3 Approach |
|---|---|
| Early churn (0-90 days) | Product-embedded onboarding prevention |
| Mid-lifecycle churn (4-12 months) | Automated engagement + adoption prompts |
| Renewal-period churn (pre-contract) | Triggered sequences + CSM escalation |
| Champion departure | Stakeholder diversification prompts |
| Usage cliff | Immediate re-engagement sequences |
| Competitive threat | Value reinforcement + executive engagement |
Conversation Guide
Opening Frame
"Most SaaS companies can predict churn pretty well at this point. The model isn't the problem—the action is. When your model flags 500 at-risk accounts and your team has capacity for 50 interventions, you're watching 450 customers churn while knowing it's coming. That's not a prediction problem; it's a prediction-action gap. We help close that gap—connecting your prediction to intervention at scale, so every at-risk account gets attention, not just the top 10%."
Discovery Questions
- "You have churn prediction—what happens after an account is flagged?"
- "Of the accounts your model identifies as at-risk, what percentage get meaningful intervention?"
- "When was your last 'save' conversation? Was it 60 days before churn, or 6 days?"
- "What would happen to your churn rate if you could intervene on every at-risk account?"
- "Does your product know when a user is struggling? Does it do anything about it?"
- "What's the cost of your current churn vs. the cost of prevention capability?"
Proof Points to Cite
- Gainsight: 100% intervention coverage (vs. 20%); 25% churn reduction
- Pendo: 30% of at-risk accounts saved through automated in-app intervention
- Intercom: 70% of risk interventions handled by AI; 80%+ satisfaction
- LMS vertical SaaS: 39% churn improvement; 3-month payback
Readiness Signals
Ready (green lights):
- "We can see churn coming but can't stop it"
- Prediction exists; action doesn't scale
- CSMs overwhelmed and firefighting
- NRR < 100% with concern about it
- "We only intervene on our top accounts"
Not Ready (yellow/red lights):
- No prediction capability (need that first)
- Churn appears truly random
- <100 customers (manual still works)
- No intervention channels at all
- CSM team resistant to AI involvement
Version History
| Version | Date | Changes |
|---|---|---|
| 1.0 | 2026-01-03 | Initial SaaS P3 documentation |