Problem & Business Impact

Mobile Number Portability (MNP) represents one of the most critical forms of churn, as it indicates a firm customer decision to migrate to a competitor. Once a customer requests their RIO (Operator Identity Statement) code, the intervention window is extremely short - typically 48 to 72 hours before actual portability. This form of churn is particularly costly as it often affects high-value customers, influenced by aggressive competitive offers.

The average port-out rate varies between 5-8% annually, with high concentration on high-value segments (premium postpaid). Without intervention, 85-90% of RIO requests result in actual portability. Our system combines early detection of pre-RIO signals with ultra-fast post-RIO response, maximizing retention chances in this critical window.

Measured impact:

28% reduction in actual port-out rate
18% save rate on RIO requests (vs 8% baseline)
Preservation of €8.5M in annual revenue
380% ROI on counter-attack offers

Data & Key Features

Main data sources

RIO events: Code requests, timing, channel (web, phone, store)
Pre-RIO behavior: Web/app searches, offer consultation, price comparators
Customer satisfaction: NPS, CSAT, recent complaints, service quality
Value profile: ARPU, tenure, subscribed services, cross-sell potential
Competitive intelligence: Active offers, campaigns, network coverage

Engineered features

Churn intention score (web/app behavior analyzed by NLP)
Service problem severity and recency
Price gap vs competitive offers in same area
Product engagement (mobile app, self-service, consumption)
Promotion sensitivity (response history)
Predicted remaining lifetime score (residual LTV)
Time since RIO request (temporal criticality)
Retention offer success probability

Models & Methods

Dual approach: Predictive + Reactive

XGBoost (Pre-RIO predictive model): Early port-out risk detection
- AUC-ROC: 0.82
- Top 10% precision: 58%
- Key features: Competitive web searches (26%), low NPS (21%), contract end (19%)
- Horizon: 14-30 days before RIO request
Neural Networks (Post-RIO model): Immediate response optimization
- Success rate prediction by offer type
- Optimal offer personalization (price, services, duration)
- Response time: <2 hours after RIO request
Ensemble Methods (Prioritization model): Intervention value scoring
- Combines port-out risk, customer value, save probability
- ROI optimization: Focus on saveable high-value customers
- Dynamic retention budget allocation

Port-out reason segmentation

Price/Competitive offer: Aggressive competitor promotion (48%)
Service/Network quality: Technical dissatisfaction (28%)
Customer service: Negative support/sales experience (14%)
Contract end: Change opportunity without fees (10%)

Real-time Integration

Critical real-time pipeline

RIO request (immediate detection) → Real-time scoring (``<30s``)
    ↓
Context enrichment (history, likely competitor)
    ↓
Optimal offer generation → Priority agent routing
    ↓
Customer contact (``<2h``) → Personalized offer presentation
    ↓
Decision tracking → Model feedback

Intervention windows

Pre-RIO (14-30 days): Soft proactive contact, issue resolution, preventive offers
Immediate post-RIO (0-24h): Priority call, premium offer, manager escalation
Late post-RIO (24-72h): Last attempt, exceptional offers, facilitations

Save offer personalization

Automatic analysis of likely competitive offer
Generation of equivalent or superior offer
Addition of differentiating advantages (exclusive services, loyalty bonus)
Cost calibration vs customer LTV

KPIs & Performance Targets

| Metric | Target | Current | |--------|--------|---------| | Annual port-out rate | <5% | 6.8% → 4.9% | | Post-RIO save rate | >15% | 18% | | Post-RIO contact delay | <2h | 1.5h average | | Revenue preserved | €7M/year | €8.5M | | Pre-RIO model precision | >55% | 58% | | Save offer ROI | >300% | 380% |

Deployment & Monitoring

Real-time architecture

Event detection: Continuous RIO request monitoring (operator API)
Immediate enrichment: Feature aggregation from feature store (<10s)
ML scoring: Risk, value, save probability calculation (<20s)
Offer generation: Personalized optimization via rules engine + ML
Intelligent routing: Automatic assignment to top-performing agent
Outcome tracking: Accept/refuse tracking + actual portability D+7

Performance monitoring

Real-time dashboard: RIO requests, contact rate, save rate
Alerts on save rate degradation (>10% vs baseline)
Continuous A/B analysis on offers and contact strategies
Weekly retraining with outcome feedback

A/B Testing

Test variants: Different offer levels, contact timings, agent scripts
Metrics: Save rate, cost per save, customer satisfaction, post-save LTV
Continuous optimization: Automatic strategy adjustment based on performance

FAQ & Prerequisites

Q: How to detect port-out intention before RIO request? A: Behavioral analysis: web searches on competitors, price comparator consultation, competitor site visits (cookie/app tracking), increased support contacts for unresolved issues.

Q: What is the optimal intervention window? A: Immediate post-RIO (0-24h) offers the best save rate (22%). Beyond 48h, the customer has often already finalized with the competitor. Pre-RIO can prevent the request but requires more false positives.

Q: How to calibrate save offers to remain profitable? A: Cost vs LTV analysis: a customer with residual LTV of €800 justifies a save offer up to €250-300. The model optimizes the minimum offer needed to save the customer (sweet spot).

Q: Can we identify the target competitor? A: Yes, via several signals: web searches, analysis of active offers in geographic area, customer profile (alignment with competitor positioning). Accuracy: ~70%.

Q: What are the technical prerequisites? A: Real-time architecture (event streaming), performant feature store (<10s access), RIO detection API, CRM integration for agent routing, automated offer generation system.

MNP Port-Out Prevention