02 Underwriting

Insurance ML
Underwriting & Customer Retention.

Client: High-net-worth insurance company
Market: UK
Risk: Customer churn risk
Service: Machine learning for underwriting and customer retention
Product: Validation

Problem: High-net-worth insurers serve a narrow, exceptionally profitable customer base — yet conventional underwriting and retention tools are rarely calibrated to this segment, leaving significant pricing risk unaddressed and churn difficult to predict.

Solution: Machine learning methods applied to underwriting analysis and customer retention — covering prediction, feature engineering, customer segmentation, model interpretation, and price elasticity.

Impact: Enabled targeted retention investment by identifying the most profitable customer segment and the features most predictive of churn, and sharpened underwriting pricing through machine learning–based elasticity analysis.

The challenge

High-net-worth risk is a concentrated market: few carriers compete for a small pool of high-value policyholders, and a single wave of defections to a competitor can materially reduce premium income. The cost of replacing lost clients is high, making churn prediction and proactive retention critical.

Churn prediction with limited data. High-net-worth books are small relative to standard personal lines, restricting the training data available and requiring model approaches that perform reliably in low-sample regimes.
Feature engineering and risk factor selection. Policyholder data is heterogeneous — spanning claims history, property characteristics, relationship tenure, and payment behaviour — requiring careful construction and selection of predictive features before modelling.
Customer segmentation. Identifying distinct policyholder groups with different retention and profitability profiles is not straightforward when the overall book is small and customer characteristics vary widely.
Interpretability. Insurers need models whose outputs underwriters, actuaries, and regulators can interrogate — requiring explainability alongside predictive performance.
Demand for high-net-worth cover is elastic. Premium sensitivity varies significantly across the customer base; pricing decisions that ignore elasticity risk triggering churn among segments where price is the primary driver of renewal.

The solution

The analysis examined customer characteristics relevant to retention and acquisition and provided information for model validation.

Applied a range of machine learning models to identify features associated with customers leaving.
Identified the group of customers most profitable for the company.
Constructed gain charts to identify prospective customers more likely to convert and existing customers less likely to leave.

Deliverables

The engagement provided a clear, data-supported basis for targeting retention spend and sharpening underwriting pricing.

Top customer identification: Ranked policyholders by profitability and identified the highest-value segment — providing a clear target for retention spend and personalised outreach.
Feature importance analysis: Revealed the customer characteristics most strongly associated with churn, giving underwriters and retention teams an actionable basis for intervention.
Gain charts: Produced gain charts to support prioritisation of prospective customers most likely to convert and existing customers at highest risk of churn.
Model validation evidence: Delivered analysis supporting model validation requirements, including stability checks and performance benchmarking.