Amazon Bedrock AgentCore automates initial intake of insurance claims

Insurance companies spend enormous amounts of time on routine data entry during initial registration of insurance claims. AWS has published the architecture of a system that completely eliminates this work: an AI agent independently interacts with insurer web portals, requiring no operator involvement.

What is FNOL and what's the problem

FNOL (First Notice of Loss) — initial notification of loss — is a mandatory first step in any insurance claim: a traffic accident, fire, medical incident, or theft. An operator takes a call from a customer, collects details of the incident, and enters them into insurance systems. The vast majority of this work is strictly mechanical. A specialist copies data from one window to another, fills in dozens of standard fields, checks policy numbers, and switches between multiple systems. Errors during manual entry lead to delays in processing and repeat calls to customers. In large insurance companies, FNOL operations consume tens of thousands of person-hours annually.

How the system works

AWS proposes a solution based on two key components:

• Strands Agents SDK handles logic and reasoning. The agent receives information about the loss, analyzes context, classifies the type of case, and determines a sequence of actions according to the specific insurer's rules.
• Amazon Bedrock AgentCore Browser Tool performs the work in the browser: opens insurance portals, fills out forms, clicks buttons, and reads results — just as a live operator would.

The system works with real web interfaces without additional API integrations. A new portal can be connected without development — the agent independently figures out the interface. Humans step in only for complex or unusual cases.

The fundamental difference from the traditional approach is the rejection of custom integrations. Usually, automating insurance processes requires developing separate connectors to each system. The Browser Tool perceives the screen as a human and controls the browser directly, eliminating this dependency. The browser agent approach is applicable not only in insurance: banking services, government services, and medical documentation face the same problem of manual data entry into web portals.

Strands SDK and the role of the specialist

Strands Agents SDK — an open-source AWS framework for building AI agents. In the context of insurance automation, it bears responsibility for the "mental" part of the work: understanding what happened, determining what data is needed, checking for contradictions in details, and ensuring the case matches policy conditions. The agent works iteratively: asks clarifying questions, validates data, and requests additional information when necessary.

Strands SDK is distributed as open code — companies can add their own processing rules or integrate with existing decision-making systems. The system does not replace insurance experts — it takes on strictly mechanical work. Complex situations requiring professional judgment: potential fraud, unusual losses, disputed circumstances — remain in the domain of humans.

"The approach preserves human expertise while eliminating repetitive screen work,"

AWS describes the system's principle.

Specialists freed from routine can focus on cases where real experience is needed: settling complex claims, working with customers in stressful situations, identifying suspicious patterns in the stream of claims.

What this means

Browser-based AI agents are becoming a working tool for back-office automation in insurance and financial companies. AWS architecture based on Strands SDK and Bedrock AgentCore Browser Tool sets a template that adapts to any portal without rewriting integrations. For the insurance industry, this means lowering the threshold for adoption: instead of months-long integration projects, it takes only tuning the agent for a specific web interface.