Engineering • Behavioral Model Series
Part 5 of 10

Churn Prediction and LTV Scoring at the Edge

How ClickStream detects early churn signals and predicts lifetime value from behavioral patterns -- enabling real-time retention interventions before customers disengage.

March 2026

Introduction Part 1: Intent, Frustration & Engagement Part 2: Value & Anomaly Part 3: Confusion & Emotion Part 4: Decision & Regret Part 5: Churn & LTV Part 6: Abandonment & Timing Part 7: Affinity, Friction & Next Action Part 8: Momentum, Entropy & Attention Part 9: Conversion, Hover & Scroll Part 10: Price, Loyalty, Micro-Conversion & Bot Detection

What You'll See in the Dashboard

The dashboard surfaces churn risk and lifetime value per visitor — churn as a probability score (0–100) and LTV as a 0–100 score with a tier and confidence. Filter visitors by churn risk to see which customers need attention right now. (Behavioral scoring models are available on Growth plans and above.)

Business Actions: Trigger a retention offer when churn risk climbs. Pipe high-LTV visitors into your CRM for white-glove outreach. Segment audiences by LTV tier to optimize ad spend on your most valuable cohorts.

Model 10: Churn Prediction

The churn prediction model estimates the probability that a visitor or customer will not return. For SaaS products, this predicts subscription cancellation; for e-commerce, it predicts the end of a customer relationship. ClickStream detects churn signals during active sessions, enabling intervention while the user is still present.

The 9 Churn Signal Families (Illustrative)

The signal families below illustrate the kinds of on-site behavior the model weighs. The exact signal list and weights are internal — and tunable per site via scoring configuration.

SignalDescription
Declining session frequencyInterval between sessions is increasing over the last 4 visits
Declining engagement depthEngagement score trending downward across sessions
Reduced feature usageUsing fewer product features per session than historical average
Support/cancel page visitsViewing cancellation, downgrade, or complaint pages
Competitor comparison signalsReferrer from competitor review sites; comparison page visits
Declining session durationSessions getting shorter over time
Error-frustration accumulationFrustration score increasing across multiple sessions
Billing page patternVisiting billing settings without making a purchase
Content disengagementReduced return-visit depth and declining on-site content interaction across sessions

The 4 Churn Levels (Suggested Operating Framework)

The model’s output is a 0–100 churn risk score. The bands and level names below are a suggested operating framework for acting on that score — not labels the product emits.

Score RangeLevelDescriptionIntervention Urgency
0–25HealthyNo churn indicators present. User is engaged and active.None. Continue normal experience.
26–50At RiskEarly warning signals detected. Engagement declining.Low. Proactive engagement nudges.
51–75Likely ChurningMultiple strong churn signals. Clear disengagement pattern.Medium. Retention offer or outreach.
76–100Imminent ChurnActive churn behavior (cancel page, billing review).High. Immediate intervention required.

Churn Risk Factors

TypeScript — illustrative
interface ChurnRiskFactors { sessionFrequencyTrend: number; // negative = declining engagementTrend: number; // slope of engagement over sessions featureUsageBreadth: number; // 0-1, current vs. historical cancelPageVisits: number; // count in last 30 days competitorReferrals: number; // competitor site referrals avgSessionDurationTrend: number; // negative = declining cumulativeFrustration: number; // sum of frustration across sessions billingPageVisits: number; // without transaction contentEngagementDecline: number; // 0-1 decline ratio }

SaaS vs. E-Commerce Churn Signals

Signal CategorySaaS IndicatorE-Commerce Indicator
Usage declineFewer logins, less feature usage, shorter sessionsLess frequent visits, smaller browsing sessions
Active disengagementExport data, cancel page, downgrade pageUnsubscribe page, email opt-out, account deletion
Competitor signalsReferrer from G2/Capterra competitor pagesReferrer from competitor product pages
Support patternIncreasing support tickets with declining satisfactionComplaint page visits, return policy views
Financial signalsBilling page views, plan comparison without upgradeCoupon-only purchasing, declining AOV

Model 11: LTV Score

The LTV score predicts the total value a visitor will generate over their entire customer relationship. Unlike the value estimator (Part 2) which focuses on the current session, LTV scoring projects forward across months or years.

The 8 LTV Signal Families (Illustrative)

As with churn, these signal families illustrate what the model weighs; the exact signal list and weights are internal and tunable per site.

SignalDescription
Purchase frequency trendAccelerating purchase frequency = higher LTV
Average order value trendIncreasing AOV over time = growing relationship
Engagement depth consistencySustained high engagement = loyal customer
Product category breadthCross-category purchasing = deeper relationship
Referral source qualityOrganic/direct visitors have higher average LTV
Session recencyRecent high-engagement sessions boost LTV
Feature adoption velocitySpeed of adopting new features (SaaS-specific)
Social engagementReviews, referrals, community participation

The 4 LTV Tiers (Suggested Operating Framework)

The shipped model outputs a 0–100 LTV score with a tier and a confidence value. The tier names and bands below are a suggested operating framework, not documented product output.

Score RangeTierDescriptionStrategy
0–25Low LTVOne-time or infrequent buyer. Limited growth potential.Efficient self-service. Minimize support cost.
26–50Medium LTVRepeat buyer with growth potential.Cross-sell, loyalty programs, engagement campaigns.
51–75High LTVLoyal customer with consistent purchasing.Priority support, exclusive offers, early access.
76–100WhaleTop-tier customer. Significant revenue contribution.Dedicated account management, VIP treatment.

Real-Time vs. Historical LTV

ClickStream computes two LTV variants:

One illustrative way to blend the two: ltvScore = 0.6 * historicalLTV + 0.4 * sessionLTV for returning visitors, and ltvScore = sessionLTV for new visitors. The shipped model’s exact weighting is internal and tunable per site.

Churn × LTV Priority Matrix

The intersection of churn risk and LTV creates a prioritization framework for customer success teams:

Low LTV (0–25)Medium LTV (26–50)High LTV (51–75)Whale (76–100)
Healthy (0–25)Auto-nurtureCross-sell campaignsLoyalty programVIP monitoring
At Risk (26–50)Automated re-engagementTargeted email + incentiveCSM outreachExecutive escalation
Likely Churning (51–75)Win-back campaignRetention offerPersonal call + special offerC-level intervention
Imminent (76–100)Exit surveyLast-chance offerUrgent 1:1 save attemptAll hands on deck

Real-Time Intervention Timeline

Here is an example of how churn signals evolve during a single session for a SaaS customer who is considering cancellation:

TimeActionChurn ScoreLTV ScoreSystem Response
0:00Login to dashboard3572Monitor: at-risk + high-LTV flagged
0:30Navigates to billing settings5272Alert: CSM notified via your webhook → Slack workflow
1:15Views plan comparison page5872Show: "Need help choosing a plan?" chat widget
2:00Clicks "Cancel Subscription"8572Intercept: retention offer + personal call scheduling
2:30Views retention offer7872Monitor: did they accept?
3:00Accepts offer, returns to dashboard4572Success: tag for follow-up in 7 days

Alert Integration

Churn and LTV alerts can reach your existing tools via ClickStream’s signed webhook destination:

JSON — example payload
{ "alert_type": "churn_risk", "severity": "high", "visitor_id": "v_abc123", "session_id": "s_def456", "churn_score": 78, "ltv_score": 72, "ltv_tier": "high", "priority_action": "urgent_save_attempt", "churn_signals": [ "cancel_page_visit", "billing_page_view", "declining_engagement" ], "recommended_interventions": [ "retention_offer", "personal_outreach", "feature_education" ], "timestamp": "2026-03-10T14:23:45Z" }

The churn-LTV matrix transforms retention from a reactive process (noticing a cancellation after it happens) into a proactive one (intervening during the session where churn behavior first appears). For high-LTV customers, this single capability can justify the entire ClickStream investment.

Previous in Series ← Part 4: Decision Confidence & Regret Risk

Save Revenue You Are Already Losing to Churn

Detect disengaging customers before they cancel and intervene with the right offer at the right time.

Start free