An insurance broker spends a significant portion of their day trapped in a cycle of digital transcription. They move a client's address from a PDF application to a CRM, then manually re-enter that same data into an underwriting portal, and finally relay a coverage detail from a carrier's email back to the customer. In an industry managing 8 trillion dollars in global assets, this administrative tax is not just a nuisance; it is a structural bottleneck that limits growth and burns out talent. The friction of redundant data entry creates a ceiling on how many clients a single agent can manage, regardless of their expertise in risk assessment or relationship building.
The Infrastructure of Automated Underwriting
Cara addresses this inefficiency by deploying an AI-native orchestration layer built entirely on Amazon Web Services. The founders, Vic Yeh, Nikhil Kansal, and Jon Patel, developed the solution after operating their own digital brokerage, which they eventually sold to The McGowan Companies, one of the largest private insurance organizations in the United States. Their experience revealed that the primary drain on productivity was not the complexity of insurance law, but the manual movement of data across fragmented systems.
To solve this, Cara utilizes Amazon EKS (Elastic Kubernetes Service) to manage a sophisticated microservices architecture. This setup divides the platform into three primary layers: a data ingestion pipeline, a workflow engine, and an AI inference layer. By distributing these services across multiple Availability Zones, Cara ensures that the system remains resilient during peak insurance renewal seasons when traffic spikes. This containerized approach allows the platform to scale elastically, supporting thousands of concurrent users and complex workflows without degrading performance.
For the intelligence layer, Cara leverages Amazon Bedrock. By using this fully managed service, the team avoids the operational overhead of managing GPU clusters, configuring drivers, or scaling raw compute instances. Bedrock provides direct API access to foundation models, allowing Cara's engineers to shift their focus from infrastructure maintenance to the refinement of insurance-specific prompt engineering and workflow optimization. This architectural choice transforms the AI from a standalone tool into a seamless utility that powers the back-office automation.
Beyond General AI and the Security Wall
While general-purpose large language models can summarize text or draft emails, they often fail when confronted with the rigid requirements of the insurance domain. General AI lacks an inherent understanding of specific insurance data models, the nuanced workflows of brokerage, and the strict regulatory constraints that vary by carrier and jurisdiction. Cara's pivot toward domain-specific AI acknowledges that for a model to be useful in a professional brokerage, it must adhere to a precise set of industry rules rather than predicting the next most likely token in a general dataset.
This need for precision extends to data security. Insurance brokers handle highly sensitive Personally Identifiable Information (PII), financial records, and proprietary underwriting details. To meet enterprise-grade security standards, Cara implements a strict tenant isolation strategy using Amazon EKS namespaces. By assigning each brokerage its own logical namespace within the cluster, Cara ensures that workloads are isolated and data access is physically partitioned between different clients.
To further harden this security posture, Cara employs an AWS account-per-tenant deployment model. This ensures that each brokerage operates within its own dedicated security workspace, providing a level of auditability and isolation that exceeds simple software-level permissions. By isolating the infrastructure itself, Cara eliminates the risk of data leakage between tenants, satisfying the stringent compliance requirements of the financial services industry.
To accelerate the time-to-value for new clients, Cara uses parameterized templates to automate the onboarding process. When a new brokerage joins the platform, the system automatically provisions the isolated namespace, dedicated storage, and inference endpoints. This automation reduces the onboarding window from weeks to a few hours, allowing brokers to launch customized workflows almost immediately. By integrating directly with existing Agency Management Systems (AMS) and Customer Relationship Management (CRM) tools, Cara removes the need for agents to learn a new interface, embedding AI directly into the tools they already use.
Technical implementation details for these patterns are available via the AWS Architecture Center, and developers can begin building similar inference layers using the Amazon Bedrock and Amazon EKS documentation.
The transition from manual data entry to AI-driven orchestration allows insurance professionals to stop acting as data conduits and start acting as strategic advisors.
