What is LocScore™?
LocScore™ is Cytrus’s proprietary AI-powered location risk scoring engine, purpose-built for the insurance industry. It combines dozens of real-time and historical data sources — from weather patterns and flood maps to crime statistics and infrastructure data — to generate a single, explainable risk score for any address in the Netherlands.
Unlike traditional actuarial tables that update annually, LocScore™ is a living model that continuously learns from new data, enabling insurers to price risk more accurately and respond faster to changing conditions.
Why we built it
Insurance underwriters have long relied on postcode-level generalisations that fail to capture the real risk profile of individual properties. Two houses on the same street can have dramatically different risk profiles depending on elevation, proximity to water, building age, and dozens of other micro-factors.
We built LocScore™ to solve this: to give underwriters a tool that is as precise as modern data allows, while remaining fully explainable and auditable — a requirement under both EU AI Act guidelines and internal compliance frameworks.
How it helps our customers
- More accurate pricing: Reduce mispricing of risk at the individual property level, improving combined ratios.
- Faster underwriting decisions: Automate the initial risk assessment, cutting decision time from hours to seconds.
- Full explainability: Every score comes with a breakdown of contributing factors, satisfying regulatory requirements.
- Dynamic risk monitoring: Track how a property’s risk profile changes over time as environmental conditions evolve.
- Portfolio insights: Aggregate scores across a portfolio to identify concentration risk and stress-test against climate scenarios.
“LocScore™ gave us the ability to differentiate risk at a granularity we never thought possible. It transformed how we approach pricing for residential portfolios.”
— Head of Personal Lines Underwriting, Dutch insurer
Technical approach
LocScore™ is built on a gradient-boosted ensemble model trained on over 15 years of claims data, combined with real-time data feeds from national weather services, cadastral records, and satellite imagery. The model outputs a composite score from 0 to 100, alongside factor-level attributions using SHAP values for full transparency.
The engine is available via REST API, enabling seamless integration into existing underwriting workbenches, policy management systems, and broker portals.