How Mapfre Insurance modernized fraud claims with Amazon EMR Serverless
Mapfre Insurance · United States
Mapfre Insurance, the number one auto and home insurer in Massachusetts, modernized fraud detection by combining graph-based features from Neo4j with machine learning models deployed on AWS, using Amazon EMR Serverless, Apache Iceberg tables on Amazon S3, AWS Glue Data Catalog, AWS Lake Formation, and Amazon MWAA for orchestration. Fraud predictions integrate directly with Guidewire Claims via AWS Lambda, automatically creating claim activities showing the top model drivers for adjusters. The initiative, covering Massachusetts Auto insurance and later expanded to Home insurance, has delivered more than $5 million in Net Present Value, with detection accuracy improved 50-135 percent compared to baseline methods.
Overview
Mapfre Insurance, the number one auto and home insurer in Massachusetts, modernized fraud detection by combining graph-based features from Neo4j with machine learning models deployed on AWS, using Amazon EMR Serverless, Apache Iceberg tables on Amazon S3, AWS Glue Data Catalog, AWS Lake Formation, and Amazon MWAA for orchestration. Fraud predictions integrate directly with Guidewire Claims via AWS Lambda, automatically creating claim activities showing the top model drivers for adjusters. The initiative, covering Massachusetts Auto insurance and later expanded to Home insurance, has delivered more than $5 million in Net Present Value, with detection accuracy improved 50-135 percent compared to baseline methods.
This entry has 12 published fields tied to exact passages in an immutable source capture.
Inspect the highlighted sourceThe challenge
Traditional fraud detection approaches at Mapfre Insurance relied on rules-based controls, manual investigation triggers, historical claim patterns, and structured-data-only analysis, which struggled to detect sophisticated fraud rings or hidden relationships across claimants, policies, vehicles, providers, addresses, and prior suspicious activities.
The solution
In collaboration with AWS and Neo4j, Mapfre Insurance combined graph-based features from Neo4j with machine learning models deployed on AWS, using Amazon EMR Serverless for processing, Apache Iceberg tables on Amazon S3 with AWS Glue Data Catalog and AWS Lake Formation for governance, and Amazon MWAA for orchestration. Fraud predictions integrate directly with Guidewire Claims via an AWS Lambda function, automatically creating claim activities that show the top three model drivers for adjusters.
Reported business value
The initiative, covering Massachusetts Auto insurance and later expanded to Home insurance, has delivered more than $5 million in Net Present Value, with realized savings outperforming projections, and detection accuracy improved 50-135 percent compared to baseline methods.
Sources
Open any source and check the claim yourself — that is the point of the register.
Other insurance entries in the register.
UNIQA SEE cuts health insurance claims processing times by 90% with Azure AI
UNIQA SEE's claims handlers faced fragmented documents, manual tasks, and heavy call-center reliance that slowed claims processing. Using Microsoft Azure AI alongside Microsoft Foundry, Azure Document Intelligence, and Azure OpenAI, UNIQA SEE built an end-to-end AI-driven medical referral solution that extracts, validates, and integrates referral letter data via APIs into the core claims system. The solution reduced claims processing time from 15-30 minutes to 2-3 minutes (an 85-90% reduction), lowered costs and call volumes, and freed staff for higher-value work. UNIQA SEE also introduced a self-service web and mobile app allowing customers to submit medical referral letters digitally. UNIQA SEE became the first insurer in Southeastern Europe to implement an AI-powered business process solution of this kind.
Efficiency meets data sovereignty: Lector automates document processing with AI
Lector, a platform for intelligent document processing used by more than 60 corporate customers including over 25 health insurers, migrated its productive cloud services from a major American cloud provider to OVHcloud's German data centre to meet GDPR and data-sovereignty requirements. Using OVHcloud Managed Kubernetes, S3-compatible Object Storage, and AI Training/AI Deploy/AI Endpoint services, Lector achieved over 80% time savings in document processing across more than 500 million pages processed to date, while reducing the operational burden of running Kafka and MongoDB themselves.
Generali France Rides the Wave of Microsoft 365 Copilot and Agentic AI with Microsoft
Generali France rolled out Microsoft 365 Copilot, Copilot Studio, and Azure OpenAI as part of its 'Boost 2027' strategic plan. 100% of employees have access to Copilot Chat and 3,700 use Microsoft 365 Copilot, with 70% of the workforce adopting it and generating an average of fifteen prompts per user per week. About fifty agents built with Copilot Studio address specialized cases including helpdesk operations. A 24/7 voice assistant resolves nearly 1.3 million calls (30% of requests) directly, with no human intervention. Generative AI also supports home insurance subscription, compliance beneficiary searches, and claims management, where 2.1 million operations were processed by RPA robots in 2024, including automated analysis of wall cracks during drought episodes (10% needing no human intervention) and recognition of more than 15 document types. Implementation was supported by Microsoft partner Avanade.
Tawuniya transforms CX with unified customer intelligence
Tawuniya Insurance, the largest insurer in the MENA region by gross written premium, implemented Sprinklr Service and Sprinklr Insights to unify customer conversations, sentiment, and service operations from social media, messaging, and other digital channels onto a single platform. The company used sentiment analysis to turn customer feedback into actionable insights and automated key elements of the customer service journey including routing, queue management, and customer communications. Results: 158% improvement in First Contact Resolution (from 31% to 80% within 90 days), a 32-point increase in CSAT (from 51 to 83), and average response times reduced from 45 minutes to 56 seconds.
Was this helpful?
Your feedback helps us improve our use case database