Peggy's experience is primarily in applying predictive modeling to personal lines property and casualty insurance. Her expertise includes:
- Statistics and machine learning, including generalized linear models, Bayesian models, survival models, decision trees, and clustering
- Data mining processes, including data preparation, feature extraction, model building, and model evaluation
- Developing risk scores with external data sources such as credit reports, vehicle characteristics, and telematics devices
Peggy has developed a variety of predictive models for profitability, claim risk, retention, and improving efficiency in underwriting and claims processes. These models include:
- Telematics: Developing a telematics-based driving risk score for a large auto insurer.
- Vehicle prediction model: Developing a risk score based on the characteristics of vehicles, such as horsepower, weight, and Electronic Stability Control
- Undisclosed driver pursuit: Identifying policies most likely to have undisclosed youthful operators
- Driver training programs: Measuring efficacy of various programs claiming to reduce crash risk for elderly and teen drivers
- Household retention: Preparing a 360-degree view of customers using transactional and survey data, and identifying signals of defection risk
- Lifetime value modeling: Estimating the expected future economic value of a household, including potential value of future cross-sales.
- By-peril homeowners rating: Exploring homeowners rating by peril and fitting multivariate rating models
- Mortality data mining: Mining hundreds of auto insurance variables for predictors of mortality risk
- Medical claim fraud: Modeling medical provider fraud using medical billing data
Prior to joining Milliman, Peggy was director of analytics and insights at the Allstate Research and Planning Center.
