Automate Initial Claims Intake with a Custom AI Agent

Most small brokers rely on their AMS, like Applied Epic or Vertafore, for workflow management. These platforms are excellent databases for structured client and policy information. However, their automation capabilities are typically rule-based. They can create a task when an email with "New Claim" arrives in an inbox, but they cannot read or understand the unstructured text or PDF attachment that contains the actual claim details.

This limitation creates a tedious manual bottleneck. Consider a 10-person agency where a client emails an FNOL report for a commercial auto accident. The report is a 3-page PDF from a standard ACORD form. A customer service representative must open the PDF, find the policy number, manually create a new claim entry in Vertafore, and then copy-paste a dozen fields: driver name, incident date, location, vehicle information, and a description of the event. This process takes at least 15 minutes of focused effort and is prone to data entry errors.

After entering the data, the CSR must then make a judgment call on severity to assign an adjuster. A minor fender-bender and a multi-vehicle incident with injuries look the same to the AMS inbox rule. This initial triage is a critical step that depends entirely on human review, delaying the claims process and introducing inconsistency. The structural problem is that an AMS is built for storing data, not for interpreting it. It expects clean, structured input, but the real world delivers messy, unstructured FNOL reports.

A project would begin with a thorough audit of your current intake process. Syntora would analyze 50-100 of your most recent anonymized FNOL reports to identify all data formats, from plain text emails to multi-page PDFs. We would then map every required field in your AMS, such as Applied Epic, to ensure the AI system extracts precisely the data your team needs to begin their work. This discovery phase produces a clear data map and technical plan.

The technical architecture would be built on serverless components for efficiency and low cost. An AWS Lambda function would trigger whenever a new email or file arrives. This function uses the Claude API to read the document and extract key entities, passing the structured data to a Python-based FastAPI service. This service contains the business logic to score the claim's severity based on your specific rules and routes it by calling the API of your AMS. The entire process, from email receipt to a new claim record appearing in Vertafore, would take under 60 seconds.

The delivered system is a managed automation that runs in the background. Your team would simply see new, accurately-filled claim records appear in their existing AMS workflow, ready for adjuster follow-up. You receive the full source code, a runbook detailing the AWS architecture, and access to a dashboard tracking processing volume and accuracy. A typical build for this system would take 3-5 weeks, with projected hosting costs under $50 per month.