Predictive Trip-Cancellation Risk Model for a Travel Insurance Broker

Overview

Built a predictive model for a travel insurance broker that estimates the likelihood of a booked trip resulting in a cancellation claim, based on destination, trip timing, traveler history, and broader travel disruption patterns. The broker uses the model's output to inform proactive customer outreach — like reminding travelers to review coverage details ahead of a high-risk travel period — rather than to deny or price individual policies differently. This gave the broker's small operations team an early signal of which upcoming trips were more likely to generate claims and support requests, letting them staff and prepare accordingly. The model was trained and validated against several years of the broker's own historical claims data before being used operationally. A travel insurance broker's small operations team had no early signal for which upcoming trips were more likely to generate cancellation claims, making staffing and support preparation reactive rather than proactive. We built a predictive model estimating cancellation-claim likelihood from destination, trip timing, traveler history, and broader disruption patterns, used to inform proactive outreach — like coverage reminders ahead of high-risk periods — rather than pricing or denying individual policies. We trained and validated the model against several years of the broker's own historical claims data before it was used operationally, comparing predicted risk periods against what actually happened historically. The operations team used its output alongside their own judgment for a full quarter before relying on it for staffing decisions. The operations team now has advance visibility into likely claim volume spikes, improving staffing and response time during high-risk travel periods instead of reacting after claims already started coming in.