What challenges did Strands face before AI implementation?

Claims professionals spend excessive time on repetitive FNOL portal work validating unstructured, multimodal evidence before making coverage and severity decisions. During volume spikes, these delays compound into backlogs that slow claim resolution and affect customer experience.

What was the impact of the AI solution?

Adjusters receive context-rich, pre-analyzed submissions instead of raw artifacts. Evidence tags become durable operational data for routing, pattern analysis, and workflow refinement. Claims can be triaged earlier, with simple claims auto-resolved and complex claims routed with structured notes. The approach reduces queue contamination, rework, and late-stage escalation while preserving human oversight.

Rank#288of 3,449

Source

ProductionEvidence: Low40/100

Hands-free FNOL intake with Strands Agents and Amazon Bedrock AgentCore Browser Tool

Use case typeClaims automationUpdated Jun 9, 2026

This article describes a hands-free first notice of loss (FNOL) intake prototype for insurance claims. It combines Strands Agents for insurance-specific reasoning with Amazon Bedrock AgentCore Browser Tool and Amazon Nova Act to operate existing portals, tag multimodal evidence, and produce decision-ready intake for adjusters.

Organization: Strands
Industry: Insurance
Location: United States
Published: June 2026

Reported outcomes

Strategic outcomes

Speed & agilityAccelerated FNOL intake and triageCost efficiencyReduced rework and queue contaminationBetter decisions & insightApplied adjuster expertise earlier in the workflow

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

Customer: Strands
Provider: AWS
Maturity: Production
Linked source: AWS Machine Learning Blog

Evidence tags become durable operational data for routing, pattern analysis, and workflow refinement

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 3 of 3

1Claims automation
2Workflow automation
3Decision support

Claims professionals spend excessive time on repetitive FNOL portal work validating unstructured, multimodal evidence before making coverage and severity decisions.
During volume spikes, these delays compound into backlogs that slow claim resolution and affect customer experience.

Nova Act drives a managed browser session through Amazon Bedrock AgentCore Browser Tool to operate the FNOL portal without portal changes.
Strands Agents apply domain reasoning to interpret, tag, and correlate images, videos, audio transcripts, and claim metadata.
The architecture stores evidence in Amazon S3, claim state and analysis outputs in Amazon DynamoDB, and runs the containerized application on Amazon ECS with AWS Fargate provisioned by AWS Cloud Development Kit.
Browser interactions, prompts, screenshots, and UI state transitions are captured for observability and auditability.

Technologies

Strands Agents SDK Amazon Bedrock AgentCore Browser Tool Amazon Nova Act Amazon Bedrock Amazon ECS AWS Fargate Amazon S3 Amazon DynamoDB AWS Cloud Development Kit Amazon CloudWatch

Adjusters receive context-rich, pre-analyzed submissions instead of raw artifacts.
Evidence tags become durable operational data for routing, pattern analysis, and workflow refinement.
Claims can be triaged earlier, with simple claims auto-resolved and complex claims routed with structured notes.
The approach reduces queue contamination, rework, and late-stage escalation while preserving human oversight.

Architecture

The prototype uses Strands Agents SDK for domain reasoning, Amazon Nova Act for browser reasoning, and Amazon Bedrock AgentCore Browser Tool for a managed Chrome session. The FNOL portal and backend services are containerized on Amazon ECS with AWS Fargate, evidence artifacts are stored in Amazon S3, claim metadata and analysis outputs in Amazon DynamoDB, and AWS CloudWatch provides operational visibility. AWS CDK provisions the infrastructure. Nova Act operates the portal via CDP over WebSocket while screenshots and execution metadata provide auditability.

Sources & evidence1

Evidence: Low40/100Evidence strength