• A leading U.S.-based health insurer wanted to reduce reliance on invasive medical exams for underwriting
• At the time, 93% of cases required medical tests, and the goal was to find alternative data sources to accurately assess risk

Identified latent risk predictors from past underwriting decisions by combining internal and alternative data sources, including:

• Behavioral data
• Financial and credit health
• Non-invasive self-reported medical information
• Consumer digital footprint

• 10x increase in non-invasive risk classification
• ~50% of cases processed automatically, without manual intervention via continuously updating alternative risk indicators
• $170M in potential savings through reduced costs and improved efficiency

• A global health insurer based in Australia wanted to reduce claims leakage and improve auditing
• Existing models had only an 18% recovery rate, and the goal was to quickly identify inconsistent claims to continuously review the process

• Built an ML model to classify claim severity (DRG)
• Leverage trade flows at national/delivery points 
• Combined internal claims data with ICD and ACHI codes

• Achieved a 30% increase in recoverable claims in pilot data
• Estimated $750K in potential savings with <1-month payback  
• Provided continuously updated insights to empower contracting teams in disputes

• A leading online auto insurer wanted to improve value-based prospecting and digital marketing‍
• They needed a lapse propensity model to identify and explain the likelihood of customer churn and estimate lifetime value (LTV)

Identified over 200 drivers that predict lapse, including:

• Census
• Regional GDP
• OpenStreetMap
• Second-hand car sales data
• Synced predictive lapse, lifetime, and margin scores with a DMP to steer digital marketing and estimate customer LTV

• Developed a highly accurate lapse model with continuous feature updates as customer behaviour evolved
• Cut out media spend for 30% of customers while increasing conversions
• Focused media on high-margin, high-lapse-risk, high-LTV customers

• A leading U.S. life insurer was losing $2.6 billion due to early termination of group retirement policies
• The goal was to identify drivers of churn and improve predictive models to reduce cancellations

Identified churn predictors in historical surrenders using internal and external data sources:

• Website data
• Transactions data
• Product and plans data
• Financial data
• Housing information

• Evaluated 1.2 million predictors and discovered 150 new unknown churn drivers
• Achieved 10% improvement in classification performance over baseline
• Generated $26 million in impact through quick-win retention actions, with ongoing updates to drivers