LTV моделирование игроков

LTV (Lifetime Value) — прогнозируемый суммарный revenue от игрока за весь период отношений. Основа для marketing budget decisions.

Формула

Basic: ARPU × Retention months × Margin. Advanced: probabilistic (BG/NBD, Pareto/NBD). ML-based: gradient boosting на 90-day features предсказывает 12-month value. Segmentation: разные модели для casual/VIP/whales. Time-decay для discounting.

Features для ML LTV

Early signals: first 7 days deposits, sessions. Game preferences: high-volatility vs low. Acquisition channel: paid vs organic. Geo, device, demographics. Behavioral: session frequency, time-of-day. Payment method: cards vs crypto (correlate с LTV).

Applications

CPA bidding: max acceptable acquisition cost = LTV × margin. Predicting VIP: early identification для personalized treatment. Refund/deposit limit decisions. Marketing attribution: multi-touch models. Retention investments: differentiate by predicted LTV.

Метрики

MAPE (Mean Absolute Percentage Error): typical 20–40% для 12-month prediction. Calibration: predicted vs actual LTV curves. Segment accuracy: separate metrics для VIP/whales. Recency of prediction: 30/60/90-day accuracy trade-off.