In the insurance industry, fraud represents a growing challenge that not only increases operating costs, but also erodes customer trust and consumes valuable resources in manual investigations. Traditionally, insurers have relied on systems based on fixed rules, historical patterns, and structured data analytics to detect anomalies. However, these techniques show limitations in the face of sophisticated fraud networks, where the connections between policies, vehicles, suppliers and addresses are hidden under complex relationships that a simple cross-section of fields cannot reveal.
A leading insurer in the United States decided to take a quantum leap by modernizing its fraud detection platform by combining graphs, machine learning, and native AWS services, specifically Amazon EMR Serverless. Not only did the solution improve the accuracy of the models by 50 to 135 percent, but it generated a net present value of more than $5 million over five years. This case demonstrates how a modern data architecture, based on a data lakehouse with Apache Iceberg, AWS Glue Data Catalog, and AWS Lake Formation, can power artificial intelligence for companies looking to go beyond conventional approaches.
The core of the innovation lies in the incorporation of 54 graph-based features that capture the relationships between entities. While traditional models only analyze isolated attributes—such as the amount of the claim or the age of the policy—graphs allow you to uncover hidden links: the same repair shop associated with multiple suspicious claims, addresses shared between different policyholders, or vehicles that appear in claims under different policies. This network vision is what allows fraud detection systems to identify patterns that would go unnoticed.
The technical architecture is deployed on top of AWS using Amazon EMR Serverless to process large volumes of data elastically and efficiently. Ingestion, enrichment, and scoring pipelines are orchestrated with Apache Airflow on Amazon MWAA, ensuring a continuous and reliable flow. Data is stored in Iceberg tables on top of Amazon S3, making it easy to govern and reuse features through your own Feature Store. For the graph layer, AWS-hosted Neo4j is integrated, which feeds the centrality and connectivity metrics that enrich predictive models.
One of the most critical aspects was the integration with the Guidewire Claims claims management system. Each fraud prediction automatically generates an activity in Guidewire with an explanatory description that shows the three main factors that prompted the alert. This allows adjusters to immediately understand why a claim was flagged, increasing confidence in the system and speeding up decision-making. The integration was achieved using Lambda functions that read results from S3, send them to the Guidewire API, and handle retries, dead-letter queues (DLQs), and notifications via Amazon SNS. All credential security is handled with AWS Secrets Manager.
The financial impact was overwhelming: during the pilot phase, savings exceeded initial projections by more than half a million dollars. In 2025, combined savings in auto and home insurance reached $6.81 million. These results not only validated the investment, but demonstrated that combining structured data with graph relationships and machine learning is the way forward to combat organized fraud.
Beyond the numbers, the experience left valuable lessons. The multidisciplinary collaboration between the claims, data, advanced analytics and technology partners teams was critical. The explainability of the models was essential to gain acceptance from end users, who need to understand why a complaint is suspicious in order to act judiciously. Finally, operational resilience—based on continuous monitoring, data quality, and recovery processes—ensured that the models worked reliably in production.
This transformation is not an isolated case. More and more companies are adopting modern cloud architectures to deploy AI solutions that solve complex problems. From automating processes to creating AI agents capable of interacting with legacy systems, the possibilities are enormous. At Q2BSTUDIO, as a software development company, we accompany organizations on this journey, offering AWS and Azure cloud services, custom application development, business intelligence with Power BI and cybersecurity to protect critical data. Our experience in implementing AI-based solutions for businesses allows us to deliver real value at every stage, from conceptualization to production.
The case of the U.S. insurer illustrates that the key is not only in the technology, but in how it integrates with business processes. Fraud detection is no longer a purely reactive exercise; it becomes a predictive and explainable capability that empowers teams. For companies looking to modernize their systems, the roadmap includes migrating to a data lakehouse on AWS, adopting graphs to uncover hidden relationships, and deploying machine learning models with robust MLOps. And, of course, having a technology partner who understands both the technical and business sides.
At Q2BSTUDIO, we help companies build custom software that integrates artificial intelligence, cloud services, and data analytics. Whether they need to develop an anomaly detection system for insurance, optimize supply chains, or create an AI agent for customer service, our team is prepared to design and implement scalable, secure, and goal-aligned solutions. Digital transformation is not a destination, but an ongoing process, and being accompanied by experts makes all the difference.
