Flood risk models built on historical data are failing. With insured natural catastrophe losses exceeding $137 billion in 2024—the fifth consecutive year above $100 billion—and floods accounting for roughly $13 billion of that total in Europe alone, the industry’s reliance on backward-looking models is becoming an expensive liability.
Enter AI-enhanced catastrophe modelling. In a panel discussion hosted by InsTech in October, Professor Paul Bates—Chairman and Co-Founder of Fathom, the Bristol-based flood intelligence firm acquired by Swiss Re in December 2023—outlined both the promise and the boundaries of applying machine learning to one of insurance’s most complex perils.
The Core Shift: Better Data Before Better Algorithms
The conversation challenged a common assumption: that AI alone will solve flood modelling’s accuracy problem. Bates argued that the foundation of better flood prediction starts with better terrain data—specifically digital elevation models (DEMs)—not just more sophisticated algorithms.
Fathom’s approach combines peer-reviewed physics-based modelling with machine learning in targeted applications. The company’s FathomDEM+ product uses a novel ML approach to produce high-resolution global terrain data, which then feeds into physics-based flood simulations. AI improves the inputs; physics governs the outputs.
This distinction matters. In a market flooded with vendors claiming “AI-powered” capabilities, Fathom’s methodology—published in Environmental Research Letters and other peer-reviewed journals—offers a transparency standard that most competitors lack.
Swiss Re’s Integration: From Acquisition to Operational Impact
The strategic significance extends beyond research. In early 2026, Swiss Re announced it is integrating Fathom’s flood hazard and terrain data directly into its internal catastrophe model. The integration includes building 50,000-year probabilistic flood event sets that leverage AI-enhanced climate models to capture extreme scenarios and remove historical biases.
This is not a pilot program. Fathom’s high-fidelity flood models are already active within Swiss Re’s Risk Data Solutions product suite, and ongoing development is systematically embedding Fathom data across Swiss Re’s underwriting and portfolio management operations.
A McKinsey Global Institute report from December 2025 estimated that nearly half the world’s landmass—home to approximately four billion people—is exposed to flood-related hazards. For Swiss Re, owning the data pipeline from raw terrain intelligence through to probabilistic modelling creates a significant competitive moat in pricing this risk.
Why It Matters: The Validation Gap
The InsTech panel highlighted a critical industry tension: the rush to adopt AI-driven tools without sufficient validation infrastructure. Bates emphasized that open science and peer-reviewed methods should be non-negotiable criteria when evaluating catastrophe models—a position that cuts against the opacity of many proprietary modelling vendors.
This resonates with broader industry dynamics. The European Centre for Medium-Range Weather Forecasts (ECMWF) made its AI Forecasting System operational for deterministic forecasts in February 2025 and extended it to ensemble forecasts by July. DeepMind’s GenCast system, published in Nature, demonstrated the ability to produce global 15-day ensemble weather forecasts in roughly eight minutes. The tools are maturing fast. The validation frameworks are not keeping pace.
For insurers evaluating catastrophe model vendors, this creates a clear decision criterion: demand published methodology, transparent validation, and peer-reviewed science. Models that can’t show their work should raise red flags—regardless of how sophisticated their marketing materials appear.
The Competitive Landscape Is Shifting
Swiss Re’s Fathom integration is part of a broader pattern. Munich Re’s Risk Management Partners unit partnered with ICEYE in late 2025 to integrate satellite-based flood intelligence into its Location Risk Intelligence platform. First Street Foundation has partnered with the Connecticut Insurance Department to provide property-level climate risk data to homeowners—a first-of-its-kind public-private initiative. Aon expanded its Climate Risk Monitor with enhanced Fathom-powered flood capabilities.
The convergence is clear: reinsurers are vertically integrating flood intelligence, and the winners will be those who control the data pipeline from raw observation through to portfolio-level risk assessment.
What Comes Next
Three developments to watch. First, regulatory attention to AI model transparency is accelerating—expect flood model validation standards to tighten, particularly in the EU and UK markets. Second, the gap between reinsurers with proprietary flood intelligence and those relying on third-party models will widen, creating pricing advantages for the vertically integrated players. Third, the NFIP’s temporary shutdown in late 2025 exposed structural vulnerabilities in public flood insurance, accelerating demand for private alternatives powered by precisely the kind of granular AI modelling Fathom delivers.
The insurance industry has spent decades building flood models on what happened. The shift to AI-enhanced modelling is about pricing what will happen—and the carriers that make this transition first will define the market’s next decade.
Source: InsTech, “How AI is changing flood modelling”(November 2025)
