What challenges did Zurich Insurance Group face before AI implementation?

Manual assessment methods could not effectively predict flood claims at an individual level across millions of customer assets. Zurich needed a rigorous, auditable path to production for AI/ML models. Risk monitoring had been largely manual and not scalable across thousands of customer assets.

Source

Evidence: Low25/100

Zurich Insurance UK: Predictive parametrics for flood-claim early warning using Amazon SageMaker AI (MLOps)

Use case typePredictive maintenanceUpdated Jun 13, 2026

Zurich Insurance Group (Zurich) in the UK is using Amazon SageMaker AI to anticipate flood claims and provide proactive early warning for customer properties. The program addresses the limitation of manual assessment methods, which cannot effectively predict claims at an individual level across millions of customer assets. Zurich built an end-to-end MLOps workflow with separate training, preproduction, and production accounts, using Terraform, Lambda, SageMaker Studio, a central model registry, and automated promotion/testing. The solution uses anonymized customer and address data stored in Amazon S3 and processes it with SageMaker AI for model development, deployment, inferencing, drift monitoring, and bias monitoring.

Organization: Zurich Insurance Group
Industry: Insurance
Location: United Kingdom
Published: June 2025

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Zurich Insurance Group
Provider: AWS
Maturity: Unknown
Linked source: AWS FSI Blog

No explicit deployment-stage evidence found.

Customer identity supportedSource describes one deploymentMaturity evidence evaluated

Primary read

Use case focus

Showing 1 of 1

1Predictive maintenance

Zurich implemented Amazon SageMaker AI for model development, deployment, inferencing, and MLOps.
A central tooling and registry account coordinates separate Training and Experimentation, PreProd, and Production accounts.
Terraform is used to establish infrastructure, AWS Lambda clones project templates and sets up deployment pipelines, and SageMaker Studio supports model building.
Training data is anonymized before being stored in Amazon S3, then used to train and evaluate models with hyperparameter optimization and hold-out testing before production promotion.
Production models continuously assess customer property flood risk and are monitored for drift and bias.

Technologies

Amazon SageMaker AI Amazon SageMaker Studio Amazon S3 AWS Lambda Terraform Snowflake

Deployment speed reportedly improved from around 26 weeks without reusable patterns to around 8 weeks with reusable patterns.

Architecture

An AWS recommended MLOps architecture with a central tooling and registry AWS account plus separate Training and Experimentation, PreProd, and Production accounts. Terraform provisions infrastructure, SageMaker Studio is used in training, AWS Lambda clones project/deployment templates and creates pipelines, Amazon S3 stores anonymized training and production data, and a centralized model registry governs promotion from training to preproduction testing and then production inference with drift and bias monitoring.

Sources & evidence1

Evidence: Low25/100Evidence strength