Sistema de Detección de Churn

🚨 ¿Qué es Churn?

Churn es la pérdida de clientes, específicamente cuando dejan de hacer negocios con la empresa. Nuestro sistema detecta patrones de comportamiento que indican riesgo de abandono.

📊 Estados de Churn

1. Active (Activo)

• Definición: Clientes con actividad reciente normal
• Criterio: Compra en últimos 30-60 días (según tenant)
• Acción: Mantener engagement, up-sell/cross-sell

2. At Risk (En Riesgo)

• Definición: Comportamiento anormal, señales de desengagement
• Criterio: Sin compras 60-120 días + factores de riesgo
• Acción: Campañas de reactivación inmediata

3. Churning (Churn en Proceso)

• Definición: Alto riesgo de pérdida inminente
• Criterio: Sin compras 120-180 días + múltiples señales
• Acción: Intervención urgente, ofertas agresivas

4. Churned (Perdido)

• Definición: Cliente efectivamente perdido
• Criterio: Sin actividad >180 días
• Acción: Win-back campaigns especiales o ignorar

🧠 Algoritmo de Predicción

Modelo de Machine Learning Híbrido

def calculate_churn_probability(customer_data):
    # 1. Análisis temporal
    days_since_last_order = calculate_recency(customer_data)
    order_frequency_decline = analyze_frequency_trend(customer_data)
    seasonal_adjustment = apply_seasonal_factors(customer_data)
    
    # 2. Análisis comportamental  
    engagement_score = calculate_engagement(email_opens, site_visits)
    support_interactions = analyze_support_tickets(customer_data)
    payment_issues = check_payment_failures(customer_data)
    
    # 3. Factores RFM
    rfm_risk_factor = get_rfm_churn_multiplier(rfm_segment)
    
    # 4. Modelo predictivo
    feature_vector = [
        days_since_last_order,
        order_frequency_decline,
        engagement_score,
        support_interactions,
        payment_issues,
        rfm_risk_factor
    ]
    
    churn_probability = churn_model.predict_proba(feature_vector)[1]
    
    return {
        'probability': churn_probability,
        'risk_level': categorize_risk(churn_probability),
        'factors': analyze_risk_factors(feature_vector),
        'confidence': calculate_model_confidence()
    }

Factores de Riesgo

Temporales

• Days Since Last Order: Principal indicador
• Order Frequency Decline: Patrón descendente
• Seasonal Patterns: Ajustes por temporada/industria

Comportamentales

• Email Engagement: Opens, clicks, unsubscribes
• Website Activity: Visits, page views, time on site
• Support Tickets: Complaints, issues, satisfaction

Transaccionales

• Order Value Decline: AOV descendente
• Payment Issues: Fallos, retrasos, cambios de método
• Return Rate: Incremento en devoluciones

📈 Métricas de Ejemplo

Distribución de Churn (Tenant 56)

{
  "overview": {
    "total_customers": 29707,
    "churn_rate": 0.15,
    "at_risk_count": 4456,
    "distribution": {
      "active": 20000,      // 67.3%
      "at_risk": 4456,      // 15.0%
      "churning": 3000,     // 10.1%
      "churned": 2251       // 7.6%
    }
  }
}

Análisis por RFM Segment

RFM Segment	Churn Rate	At Risk %	Avg Days to Churn
Champions	2%	5%	180+
Loyal Customers	5%	12%	150
Potential Loyalists	10%	18%	120
At Risk	45%	35%	60
Lost	85%	10%	30

🚀 APIs de Churn

Análisis General de Churn

GET /api/v2/cdp/analytics/churn?tenant_id=56

Respuesta Completa

{
  "success": true,
  "data": {
    "overview": {
      "total_customers": 29707,
      "churn_rate": 0.15,
      "at_risk_count": 4456,
      "avg_days_to_churn": 85.5,
      "distribution": {
        "active": 20000,
        "at_risk": 4456,
        "churning": 3000,
        "churned": 2251
      },
      "last_calculation": "2024-09-16T19:25:00Z"
    },
    "risk_factors": {
      "recency": {
        "weight": 0.35,
        "description": "Days since last order"
      },
      "frequency_decline": {
        "weight": 0.25,
        "description": "Decreasing order frequency"
      },
      "engagement": {
        "weight": 0.20,
        "description": "Email and website engagement"
      },
      "support_issues": {
        "weight": 0.15,
        "description": "Customer service interactions"
      },
      "payment_issues": {
        "weight": 0.05,
        "description": "Payment failures or delays"
      }
    },
    "at_risk_customers": [
      {
        "customer_id": 67890,
        "email": "customer@example.com",
        "last_order_date": "2024-06-15",
        "days_since_last_order": 93,
        "churn_probability": 0.75,
        "churn_risk": "at_risk",
        "rfm_segment": "Need Attention",
        "primary_risk_factors": [
          "recency",
          "frequency_decline"
        ],
        "recommended_actions": [
          "immediate_email_campaign",
          "personalized_offer",
          "account_manager_outreach"
        ]
      }
    ]
  }
}

Filtrar por Nivel de Riesgo

GET /api/v2/cdp/analytics/churn?tenant_id=56&risk_level=at_risk&limit=100

Churn Individual

GET /api/v2/cdp/customers/67890/profile?tenant_id=56

Incluye sección de churn:

{
  "churn": {
    "risk": "at_risk",
    "probability": 0.75,
    "confidence": 0.82,
    "calculation_date": "2024-09-16T19:25:00Z",
    "primary_factors": [
      {
        "factor": "recency",
        "value": 93,
        "weight": 0.35,
        "contribution": 0.26
      },
      {
        "factor": "frequency_decline",
        "value": -0.6,
        "weight": 0.25,
        "contribution": 0.18
      }
    ],
    "trend": {
      "direction": "increasing",
      "velocity": 0.05,
      "days_to_predicted_churn": 45
    }
  }
}

🎯 Estrategias Anti-Churn

At Risk Customers

Immediate Actions (24-48h)

• Personal Outreach: Account manager contact
• Exclusive Offers: 20-30% discount
• Surprise & Delight: Unexpected gift/upgrade
• Feedback Request: Survey about experience

Campaign Examples

# Campaña de reactivación inmediata
def urgent_reactivation_campaign(customer_id):
    customer = get_customer_profile(customer_id)
    
    # Personalizar oferta basada en historial
    last_categories = get_last_purchased_categories(customer_id)
    discount_percent = calculate_optimal_discount(customer.clv_tier)
    
    send_email_campaign(
        customer_id=customer_id,
        template="urgent_reactivation",
        discount=discount_percent,
        products=last_categories,
        urgency=True,
        personal_note=True
    )

Churning Customers

Aggressive Intervention

• Win-Back Campaigns: 40-50% discount
• Free Shipping: Remove friction
• Payment Plans: Flexible options
• Exit Survey: Understand reasons

Churned Customers

Win-Back Strategy

• Quarterly Campaigns: Seasonal offers
• New Product Launches: Re-engagement opportunity
• Referral Incentives: Bring-a-friend discounts
• Rebranding Communications: "We've changed"

📊 Churn Prevention KPIs

Effectiveness Metrics

• Retention Rate: 85% target para At Risk campaigns
• Win-Back Rate: 25% target para Churned customers
• Time to Reactivation: <30 días promedio
• ROI de Campaigns: 4:1 mínimo

Early Warning Indicators

-- Dashboard de alertas tempranas
SELECT 
    COUNT(*) as new_at_risk,
    AVG(churn_probability) as avg_risk_score,
    SUM(CASE WHEN clv_tier = 'high' THEN 1 ELSE 0 END) as high_value_at_risk
FROM churn_analysis 
WHERE date_calculated = CURRENT_DATE 
AND churn_risk = 'at_risk' 
AND previous_risk != 'at_risk';

Model Performance

• Accuracy: 87% en predicciones 30 días
• Precision: 83% (pocos falsos positivos)
• Recall: 91% (capturamos mayoría de churners reales)
• AUC-ROC: 0.89 (excelente discriminación)

🔄 Automatización de Churn

Triggers Automáticos

# Monitoreo diario de churn risk
schedule.every().day.at("08:00").do(monitor_churn_changes)

def monitor_churn_changes():
    # Detectar cambios en risk level
    risk_changes = detect_risk_level_changes()
    
    for change in risk_changes:
        if change.new_risk == 'at_risk':
            trigger_immediate_intervention(change.customer_id)
        elif change.new_risk == 'churning':
            escalate_to_retention_team(change.customer_id)
            send_urgent_winback_campaign(change.customer_id)

# Alertas para high-value customers
def alert_high_value_churn():
    high_clv_at_risk = get_customers(
        clv_tier='high',
        churn_risk__in=['at_risk', 'churning']
    )
    
    for customer in high_clv_at_risk:
        notify_account_manager(customer.id)
        create_urgent_task_crm(customer.id)

Integración con Marketing Tools

• Email Platforms: Trigger automático de campaigns
• CRM Systems: Alertas para sales team
• Customer Success: Tasks automáticas
• Support Systems: Priority handling para at-risk

📱 Dashboard de Churn

Vista Ejecutiva

• Churn Rate Trend: Últimos 12 meses
• At Risk Alert: Customers que necesitan atención inmediata
• ROI de Retention: Performance de campaigns
• High-Value Alerts: CLV alto en riesgo

Vista Operativa

• Daily At-Risk Queue: Lista de clientes para contactar
• Campaign Performance: Métricas de win-back campaigns
• Predictive Timeline: Customers próximos a churn
• Success Stories: Clientes reactivados exitosamente

🔮 Predictive Analytics

Seasonal Adjustments

# Ajuste por estacionalidad
seasonal_factors = {
    'fashion': {
        'Q1': 1.2,  # Post-holiday churn spike
        'Q2': 0.9,  # Spring engagement
        'Q3': 0.8,  # Summer stability  
        'Q4': 1.1   # Pre-holiday uncertainty
    }
}

Industry Benchmarks

Industria	Churn Rate Anual	At Risk %	Win-Back Rate
E-commerce	20-25%	15%	20%
SaaS	5-10%	12%	35%
Subscription	15-20%	18%	25%
Retail	25-30%	20%	15%

Advanced Features (Roadmap)

• Cohort Analysis: Churn por cohortes de adquisición
• Propensity Scoring: Probabilidad de respuesta a campaigns
• Channel Attribution: Qué canales previenen mejor el churn
• Sentiment Analysis: Social media y reviews como indicators

---

💡 Tip: La prevención de churn es 5-10x más cost-effective que la adquisición de nuevos clientes. Prioriza siempre la retención sobre la adquisición.