Inspiration Property insurance claims can take 40.7 days on average from First Notice of Loss (FNOL) to final payment. The real challenge isn't always the adjuster. It's fragmentation.

Policyholders, adjusters, repair shops, parts suppliers, and lienholders operate across different systems and communication channels, leaving no single party responsible for coordinating the complete journey.

We built One Touch Resolve around one simple idea: What if an AI agent owned the outcome, not just the conversation?

What it does One Touch Resolve is an agentic omnichannel claims orchestrator that coordinates the entire journey from FNOL to settlement.

A supervisor agent powered by Amazon Bedrock AgentCore coordinates multiple independent parties while maintaining one shared claim state across:

WhatsApp → RCS → SMS → Email

Unlike a traditional chatbot, One Touch Resolve provides:

Multi-party commit gate The claim advances only when every required party has committed. If something blocks the process, such as a back-ordered part, the supervisor re-plans the workflow instead of simply waiting.

Channel-aware continuity Communication can move between channels while preserving claim context. With RCS → SMS fallback, state remains synchronized through AgentCore Memory + Amazon DynamoDB.

Human-in-the-loop settlement approval No money moves without human approval. Settlement remains blocked until an adjuster explicitly approves it, with the rule enforced in code and the workflow state machine.

Multimodal FNOL A policyholder can submit a damage photo through WhatsApp, and Amazon Nova Pro vision evaluates the damage and drivability. The agent then adapts its plan based on what it sees.

How we built it Agentic layer: Amazon Bedrock AgentCore Runtime + Memory, multi-agent collaboration, Strands Agents SDK, and Amazon Nova Pro Communication: Amazon SES, AWS End User Messaging SMS/RCS, and WhatsApp via Meta Cloud API Workflow: Inbound → Amazon SNS → Dispatcher Lambda → AgentCore Runtime → Tool Lambdas → Amazon DynamoDB → Outbound channels Infrastructure: AWS CDK Application: Python, React, and TypeScript Reproducible proof Run:

python3 scripts/send_live_cds.py

The script sends a live Amazon SES email and AWS End User Messaging SMS through the project's adapters and returns the AWS MessageIds.

Challenges we ran into One major challenge was AWS EUM Social WhatsApp onboarding, which was blocked in our account. We kept the architecture moving by implementing WhatsApp through the Meta Cloud API using a drop-in-compatible adapter.

For SMS testing, we used a provisioned simulator number to avoid carrier-registration dependencies.

We also hardened the agent against LLM failure modes with a commit-gate loop guard and one critical rule enforced directly in code:

No settlement without human approval.

What we learned The biggest lesson was that the real opportunity in insurance claims isn't automating a single conversation. It's orchestrating the entire ecosystem around the claim.

Modeling a multi-party process as a multi-agent workflow with shared, gated state makes coordination more reliable.

We also learned that critical business and safety rules, especially those involving payments, should be enforced in code, not just prompts.

What's next We're planning to extend One Touch Resolve with:

Verified RCS sender interactions for adjuster approvals AgentCore Payments for disbursement after human approval Additional workflows for utilities, logistics, healthcare, and financial services AWS Marketplace packaging for industry-specific configurations One orchestration engine. Multiple parties. Multiple channels. One continuous journey from FNOL to resolution.