Automate First Notice of Loss Intake with Custom AI
Insurance agencies automate FNOL claim intake by using AI to parse emails and web forms. The AI extracts key details, scores claim severity, and routes it to the correct adjuster.
This process involves connecting your intake channels to a language model API like Claude. The system reads unstructured reports, identifies policy numbers, and summarizes the incident. Complexity depends on the number of intake sources and integration with your specific Agency Management System.
We built a triage system for a 6-adjuster agency that was struggling with a 4-hour average first-response time. Their new system, integrated with Vertafore, now processes and routes incoming FNOL reports in 12 minutes. The entire build took 4 weeks.
What Problem Does This Solve?
Most agencies rely on the built-in workflow tools in their Agency Management System. But rule engines in Applied Epic or Vertafore depend on structured data fields. They cannot parse the body of an email to decide if 'water damage in basement' is more urgent than 'fender bender in parking lot'. Adjusters still have to read every single email to assess severity.
This forces teams to use manual email forwarding and shared inboxes. A senior adjuster becomes a human router, spending 2-3 hours a day just reading and assigning new claims instead of managing complex files. This manual process is slow and error-prone, especially after hours or on weekends when staffing is low.
Consider an agency with 8 producers and 2 adjusters. An email arrives at 4:45 PM on a Friday titled 'Claim'. Inside is a two-paragraph description of a multi-car accident with potential injuries. Because the subject line is generic, it sits in the general claims inbox. The on-call adjuster does not see it until Monday morning, 64 hours after the First Notice of Loss. The client is already frustrated by the delay.
How Does It Work?
We connect to your FNOL sources, typically a Microsoft 365 inbox via the Graph API and web form submissions via webhook. We pull the last 3 months of claim reports, around 500-1000 documents, to fine-tune prompts for the Claude API. The goal is to create a reliable JSON output for every report type, from property damage to auto liability.
We build a FastAPI service that orchestrates the workflow. An incoming FNOL report triggers an AWS Lambda function. This function calls the Claude API to parse the text, extracting entities like policy number, claimant name, and incident description. It then scores severity on a 1-5 scale based on keywords and context. The entire process from email receipt to severity score takes under 30 seconds.
The FastAPI service then pushes the structured data into your AMS. We use the official APIs for Vertafore and Applied Epic, or webhooks for systems like HawkSoft. The claim summary, severity score, and recommended next steps are written to a new activity or note. Routing logic, now based on the AI-generated score, assigns the claim to the correct adjuster; for example, scores 4-5 go to a senior adjuster, while scores 1-3 go to the general pool.
Every AI decision is logged in a Supabase database table with its confidence score. For claims scored above a severity threshold of 3, a human review gate is created. A notification is sent to a manager's Slack channel with a link to the claim and the AI-generated summary, requiring a 1-click approval before assignment. This ensures oversight on high-stakes claims while allowing 80% of routine claims to process automatically.
What Are the Key Benefits?
From 4 Hours to 12 Minutes
Our system processes, scores, and routes an incoming FNOL report in under 12 minutes, down from a typical 4-hour manual review cycle. Urgent claims get attention immediately.
Stop Paying Adjusters to Triage
Free up your most experienced adjusters from routing emails. Redirecting 2 hours of a senior adjuster's time per day creates capacity for handling more complex, high-value claims.
You Get the Full Source Code
We deliver the complete Python codebase in your private GitHub repository. You are not locked into a platform and can have any developer maintain or extend the system.
Alerts for Failed API Calls
We use structlog for structured logging and configure CloudWatch alerts. If the Claude API fails or an AMS connection drops, you get an immediate Slack notification with the error details.
Native to Your Agency's AMS
The system writes data directly into Applied Epic, Vertafore, or HawkSoft. Adjusters work within the AMS they already know, with no new software to learn.
What Does the Process Look Like?
Week 1: System Access and Discovery
You grant read-only access to FNOL intake channels and your AMS. We map the data flow and define the severity scoring and routing logic. You receive a technical spec document for approval.
Weeks 2-3: Core System Build
We write the parsing, scoring, and integration code in a shared development environment. You receive weekly progress updates with demos of the system processing your sample claim data.
Week 4: Deployment and Testing
We deploy the system to AWS Lambda and connect it to your live intake channels in a monitoring-only mode. You receive a runbook detailing the architecture and maintenance procedures.
Weeks 5-8: Go-Live and Support
The system goes live, actively routing claims. We monitor performance and accuracy for 30 days, providing support for any issues. You receive final documentation and full ownership of the system.
Frequently Asked Questions
- How much does a custom FNOL system cost?
- Pricing is based on the number of intake sources and the complexity of your AMS integration. A typical engagement for an agency with 1-2 intake channels and a standard Vertafore API integration is a fixed-price project. We provide a firm quote after a 30-minute discovery call where we review your exact requirements. Hosting costs on AWS are typically under $50/month.
- What happens if the AI misinterprets a claim?
- The system logs every decision with a confidence score. We implement a human review gate for any claim with high severity or low confidence. This sends a summary to a manager for approval before assignment. This hybrid approach prevents critical errors while still automating the bulk of routine claims. The system learns from corrections over time.
- How is this different from the AI features in our AMS?
- AMS AI features are generally limited to analyzing data already inside the system. They do not typically handle unstructured intake from emails or web forms. Syntora builds the bridge from the outside world into your AMS, parsing and structuring the FNOL data so your AMS workflows can actually use it. We extend your system, not replace a feature within it.
- Can this system handle attachments like photos or PDFs?
- Yes. The system can extract text from PDF attachments using OCR libraries like PyMuPDF and analyze images for context using multi-modal models. It can identify photos of vehicle damage versus property damage, for example. The extracted information is added to the claim summary. This requires additional scope which we would discuss during discovery.
- Who provides support after the initial 30-day period?
- You own the code and can have any developer support it. For agencies without technical staff, Syntora offers a simple monthly retainer. This covers monitoring, dependency updates, and a 4-hour response time for any production issues. We also handle retraining the model prompts if your claim patterns change significantly.
- What security measures are in place for sensitive claimant data?
- Data is encrypted in transit and at rest. We use AWS Secrets Manager for all API keys and credentials, never hardcoding them in the application. The system only processes the data needed for triage and does not store full PII in its own database long-term. All processing happens within your own dedicated cloud environment, ensuring data isolation.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call