Scale acquisition, Conversion maximisation, and LTV growth for UniSalute's digital channels
Context
Italian SMEs face a structural information asymmetry: banks rely on static PDF balance sheets with 12–18 month latency, creating 'blind spots' that lead to false negatives and latent risk across 4.4 million businesses. UniCredit's legacy scoring model lacks the digital operational context needed for accurate SME creditworthiness assessment.
Task
Architect an AI-powered commercial ecosystem to slash acquisition costs through organic networks, maximize lead-to-sale conversion via dynamic routing, and proactively identify and prevent policyholder churn before cancellation.
My Contribution
Developed a Random Forest churn prediction model and engineered a strict unit-economic threshold (14.3%). Built a full-stack interactive agent simulator (React + FastAPI) that recalculates flight risk in real-time based on 16 customer variables.
Outcome
Projected to slash CAC by 86.9% (down to €20), increase lead-to-sale conversion from 4.7% to 6.0-7.0% (generating €1.5M+ incremental revenue), and proactively protect €343,221 in net margin. Delivered a live, functional full-stack web simulator demonstrating real-time risk recalculation for call-center agents.
Key Insights
- —The Unit Economics Paradox: Waiting for a 50% churn probability is a mathematical error when an intervention call costs €7.92 to save a €362 margin. Optimizing the Machine Learning decision threshold to 14.3% mathematically maximized net profit.
- —The Smoke Detector vs. House Fire Model: Shifting retention from reactive to proactive. The AI scans 16 variables (claims, modules, demographics) to identify flight risk weeks before the customer actually cancels.
- —The Claim Value Paradox: A claim event is not just a cost center; it is the highest-trust touchpoint. Data proved that customers who actively use their insurance are less likely to churn, creating the perfect window for module cross-selling.