Roamer Prediction

Problem & Business Impact

International roaming represents a significant but underexploited revenue source. Operators miss opportunities by not proactively targeting potential travelers before departure.

Measured impact:

• +3–5% overall roaming revenue
• +20–25% roaming pack activation
• 30% reduction in roaming fee complaints

Data & Key Features

Main data sources

• Travel history: destinations, frequency, duration
• IDD calls: international call patterns
• Roaming activations: pack activation history
• Seasonal data: vacations, events, holidays
• Location: national movement patterns
• Socio-demographics: age, profession, segment

Engineered features

• Travel propensity score (0-100)
• Predicted destination (top 3 countries)
• Predicted departure window (7d, 14d, 30d, 60d)
• Trip potential value (€)
• Pack preference (data, voice, mixed)

Models & Methods

Cascade approach

1. Logistic Regression: Binary traveler/non-traveler identification

   • Accuracy: 83%
   • Recall: 78%
   • Main features: travel history, seasonality

2. Ensemble (XGBoost + Random Forest): Window prediction

   • 1-7d window accuracy: 72%
   • 7-14d window accuracy: 81%
   • 14-30d window accuracy: 85%

3. Survival Analysis: Time-to-travel modeling

   • Cox Proportional Hazards Model
   • Recurrent event integration
   • Seasonality consideration

Intelligent segmentation

• Business travelers: automatic activation, premium data packs
• Tourists: family offers, mixed packs
• Diaspora: specific destinations, voice packs
• Occasional: micro-packs, simple activation

Real-time Integration

Automatic triggers

Multiple triggers:
├── Flight ticket purchase (agency partnership)
├── Google search foreign country (DMP)
├── Repeated IDD calls
├── Foreign bank card activation
└── Historical seasonal pattern

Campaign orchestration

• Personalized SMS: "Traveling to [Country]? Activate your pack at [Price]"
• App notification: One-click offer in mobile app
• Email: Pack details, online activation
• CRM agent: Personalized script if phone contact

KPIs & Performance Targets

| Metric | Target | Current | |--------|--------|---------| | Roaming revenue | +3-5% | +4.2% | | Pack activation rate | +20-25% | +23% | | Prediction accuracy | >80% | 83% | | Campaign conversion | >15% | 17% | | Customer satisfaction | >4/5 | 4.3/5 |

Deployment & Monitoring

Decision pipeline

1. Daily scoring: Active base score refresh
2. Prioritization: Top 10% by time window
3. Campaign: Staggered sending by window
4. Tracking: Activation, usage, satisfaction monitoring

Systematic A/B Testing

• Test group: Personalized predictive campaign
• Control group: Standard generic campaign
• Metrics: Activation uplift, revenue, NPS
• Duration: 4 weeks minimum

Continuous monitoring & improvement

• Real-time activation dashboard
• Alerts if conversion rate <12%
• Feedback loop: results integration in retraining
• Post-trip analysis: satisfaction, actual usage

FAQ & Prerequisites

Q: Does it work without travel history? A: Yes, uses alternative signals (IDD calls, seasonality, demographics).

Q: Integration with partner systems (agencies)? A: REST API available for external event ingestion.

Q: GDPR / privacy? A: Pseudonymized data, opt-in required for communications, full GDPR compliance.

Q: Typical ROI? A: 4:1 over 12 months (platform investment vs. additional revenue).