Churn Prediction: предсказание оттока

Churn prediction — идентификация игроков, склонных к прекращению использования платформы. Early intervention значительно дешевле acquisition.

Definition

Definition problem: что считать churn. 30/60/90 дней без активности — типичная threshold. Different definitions для discovery, casual, VIP. Retention metrics: D1, D7, D30. Financial impact: 5% reduction в churn = 25% revenue growth (по Bain).

Features

Recency: days since last activity. Frequency: sessions per week trend. Monetary: deposits vs withdrawals. Engagement: game diversity, session length. Support tickets. Deposit method changes. Communication engagement. Peer group comparison.

Модели

Survival analysis: Kaplan-Meier, Cox proportional hazards. Classification: XGBoost на 30-day churn. Deep learning: LSTM для sequential data. Uplift modeling: identify persuadable users, не just at-risk. Каждая модель serves different intervention strategy.

Interventions

Personalized bonuses (targeted). Communications: email, SMS, in-app. Special promotions. Product changes: preferred games surfacing. Cost-benefit: intervention cost vs LTV rescue. A/B testing critical для measuring lift.