Docs/Retention Engine/Churn Prediction

Churn Prediction

Identify at-risk players before they leave with AI-powered churn prediction and automated interventions.

10 min read

Overview

Ilara uses machine learning to predict which players are likely to churn. By identifying at-risk players early, you can take proactive steps to re-engage them before they leave.

Key Benefits

Early Warning: Identify at-risk players 7-14 days before churn
Actionable Insights: Understand why players are leaving
Automated Intervention: Trigger retention campaigns automatically
ROI Tracking: Measure retention improvements

How It Works

Data Collection

Ilara analyzes behavioral signals:

Session frequency and duration trends
Declining playtime patterns
Reduced feature engagement
Purchase velocity changes
Social activity decline
Failed attempts and frustration signals

Risk Scoring

Each player receives a churn risk score from 0.0 to 1.0:

Score Range	Risk Level	Action
0.0 - 0.3	Low	Normal engagement
0.3 - 0.6	Medium	Monitor closely
0.6 - 0.8	High	Proactive intervention
0.8 - 1.0	Critical	Urgent action required

Churn Factors

Ilara identifies specific factors contributing to churn risk:

Factor	Description	Weight
`session_decline`	Sessions dropped 50%+ from average	High
`playtime_decline`	Playtime decreased significantly	High
`no_purchase_30d`	No purchases in 30 days (was payer)	Medium
`failed_attempts`	Multiple level failures without retry	Medium
`social_inactive`	No guild/friend activity	Low
`feature_abandonment`	Stopped using key features	Medium

Churn Risk Response

json

GET /v1/players/players/{id}/churn-risk
 
{
  class="code-string">"player_id": class="code-string">"550e8400-e29b-41d4-a716-446655440000",
  class="code-string">"risk_score": 0.72,
  class="code-string">"risk_level": class="code-string">"high",
  class="code-string">"factors": [
    {
      class="code-string">"factor": class="code-string">"session_decline",
      class="code-string">"description": class="code-string">"Sessions dropped 65% over 14 days",
      class="code-string">"contribution": 0.35
    },
    {
      class="code-string">"factor": class="code-string">"playtime_decline",
      class="code-string">"description": class="code-string">"Average session length dropped from 25min to 8min",
      class="code-string">"contribution": 0.22
    },
    {
      class="code-string">"factor": class="code-string">"no_purchase_30d",
      class="code-string">"description": class="code-string">"Last purchase was 45 days ago",
      class="code-string">"contribution": 0.15
    }
  ],
  class="code-string">"prediction_date": class="code-string">"2025-01-25T00:00:00Z",
  class="code-string">"confidence": 0.85
}

SDK Usage

Unity - Get Churn Risk

csharp

class=class="code-string">"code-comment">// Get churn risk for current player
var churnRisk = await IlaraClient.Instance.GetChurnRisk();
 
Debug.Log($class="code-string">"Risk Score: {churnRisk.RiskScore}");
Debug.Log($class="code-string">"Risk Level: {churnRisk.RiskLevel}");
 
class=class="code-string">"code-comment">// Check if player is at risk
if (churnRisk.IsAtRisk)
{
    class=class="code-string">"code-comment">// Show retention offer
    ShowSpecialOffer(churnRisk.RiskLevel);
 
    class=class="code-string">"code-comment">// Log factors
    foreach(var factor in churnRisk.Factors)
    {
        Debug.Log($class="code-string">"Churn factor: {factor}");
    }
}

Godot - Get Churn Risk

gdscript

var risk = await Ilara.get_churn_risk()
 
print(class="code-string">"Risk Score: ", risk.risk_score)
print(class="code-string">"Risk Level: ", risk.risk_level)
print(class="code-string">"Factors: ", risk.factors)
 
if risk.is_at_risk():
    show_retention_offer(risk.risk_level)

Automated Interventions

Configure automatic actions when players become at-risk:

Intervention Rules

json

POST /v1/retention/interventions
{
  class="code-string">"name": class="code-string">"High-Risk Player Intervention",
  class="code-string">"trigger": {
    class="code-string">"type": class="code-string">"churn_risk",
    class="code-string">"risk_level": class="code-string">"high",
    class="code-string">"cooldown_days": 7
  },
  class="code-string">"actions": [
    {
      class="code-string">"type": class="code-string">"notification",
      class="code-string">"channel": class="code-string">"push",
      class="code-string">"template_id": class="code-string">"retention-offer-v2",
      class="code-string">"delay_hours": 0
    },
    {
      class="code-string">"type": class="code-string">"offer",
      class="code-string">"offer_id": class="code-string">"comeback-bonus-50",
      class="code-string">"duration_hours": 72
    },
    {
      class="code-string">"type": class="code-string">"segment_add",
      class="code-string">"segment_id": class="code-string">"at-risk-intervention"
    }
  ]
}

Intervention Types

Type	Description
`notification`	Send push/email/in-app message
`offer`	Grant special offer or discount
`bonus`	Add bonus currency/items
`feature_unlock`	Temporarily unlock premium features
`segment_add`	Add to segment for targeting
`webhook`	Trigger custom webhook

Churn Dashboard

Monitor churn trends in Dashboard → Retention → Churn:

Risk Distribution: Players by risk level over time
Churn Rate: Actual churn vs. predicted
Intervention Success: Players saved by interventions
Factor Analysis: Most common churn factors
Cohort Comparison: Churn by acquisition cohort

Model Accuracy

Ilara tracks prediction accuracy to continuously improve:

Metric	Description	Target
Precision	Of predicted churns, how many actually churned	>75%
Recall	Of actual churns, how many were predicted	>80%
AUC	Area under ROC curve	>0.85
Lead Time	Days before churn prediction is made	7-14 days

Model Training

The model improves as more data is collected. Initial predictions start after ~1000 churned players in your game.

Churn-Based Segments

Ilara creates automatic segments based on churn risk:

Segment	Criteria
`at-risk-players`	Risk score > 0.6
`critical-risk`	Risk score > 0.8
`declining-engagement`	Session decline > 50%
`saved-from-churn`	Was at-risk, now recovered

Best Practices

Act early: Intervene at medium risk, don't wait for critical
Personalize offers: Match offer to player value and preferences
Don't over-notify: One intervention per risk event, not daily
Track outcomes: Measure which interventions work best
Iterate: Test different offers and timing

Next Steps

AI Message Generation — Personalize retention messages
Segments — Target at-risk segments
Analytics Dashboard — Monitor retention metrics

Multi-Channel Setup

Dashboard Overview