BUSINESS PROBLEM
“Too many policies are not renewed”
          │
          ▼
DECISION TO IMPROVE
“Which policies need early intervention?”
          │
          ▼
PREDICTION DEFINITION
├── Unit: one expiring policy
├── Moment: 30 days before expiry
├── Target: renewal/non-renewal
└── Output: risk probability
          │
          ▼
AVAILABLE DATA
├── Past renewals
├── Claims
├── Complaints
├── Payment delays
└── Customer engagement
          │
          ▼
BASELINE
“Complaint in last 90 days”
          │
          ▼
MODEL DEVELOPMENT
Train → Validate → Test
          │
          ▼
MODEL METRICS
Precision, recall, false positives and false negatives
          │
          ▼
BUSINESS ACTION
Create prioritized retention tasks
          │
          ▼
BUSINESS KPI
Renewal rate and retained revenue
          │
          ▼
PRODUCTION MONITORING
Data quality + model quality + business impact

The lifecycle is iterative:

Define → Collect → Prepare → Train → Evaluate
   ▲                                  │
   └──── Monitor ← Deploy ← Approve ──┘
PreviousNext