Automate FNOL Intake and Cut Response Times by 95%

Many agencies try to use their Agency Management System's built-in rules. Vertafore's basic email rules can file an email into a client's record, but they cannot read the content. This means an adjuster still has to manually open the email, read the FNOL report, determine if it is a minor fender bender or a total loss, and then decide who gets the case.

A typical agency with 10 employees gets 40 FNOL emails a day. The office manager spends their first two hours every morning reading each one. They have to decide if a claim is for a commercial or personal line, which adjuster handles that client, and how urgent it is. A complex commercial property claim gets delayed because it arrived after a batch of 15 simple auto glass claims, making a high-value client wait 3 hours for a response.

The core issue is that keyword-based rules and manual sorting cannot understand context. An email containing the word "fire" could be a catastrophic commercial property loss or a minor kitchen flare-up in a rental unit. Without an AI model that understands the narrative of the claim, every report requires human review, creating a permanent bottleneck that scales with claim volume.

Syntora's engagement would begin with a discovery phase to audit your current FNOL intake channels, which commonly include Office 365 or Google Workspace inboxes, web forms, and PDF submissions. We would identify the necessary APIs or data extraction methods. To facilitate prompt engineering and model training, we would work with your team to securely access and preprocess a sample dataset of historical claim reports, typically ranging from 1,000 to 5,000 documents. Our approach would involve using robust Python libraries like imaplib for email parsing and pypdf for text extraction from PDF attachments.

The core of the system would be a custom-built FastAPI service, engineered to receive and process the raw claim text. This service would send the text to the Claude API, utilizing a carefully designed prompt to accurately extract critical entities such as policy number, claimant name, incident type, and a concise narrative summary. The Claude API is capable of returning a structured JSON object, often within 800ms, enabling rapid processing. A subsequent prompt would then assign a severity score, typically on a scale of 1-10, based on the extracted information.

To ensure scalability and cost-efficiency, the FastAPI service would be architected for serverless deployment on platforms like AWS Lambda, with expected operational costs for processing up to 10,000 claims typically under $30/month. Upon scoring, the system would integrate with your Agency Management System, pushing the structured claim data via webhooks or direct API connectors. While we have experience integrating with various AMS platforms, including common ones like Applied Epic, Vertafore, and HawkSoft, custom integration logic would be developed for your specific environment. A configurable routing logic, often implemented as a simple Python configuration, would then assign the claim to the appropriate adjuster's queue based on its severity score and line of business.

For transparency and auditability, every AI decision, along with its confidence score, would be logged to a secure database, such as Supabase. The system can be configured to flag claims exceeding a defined severity threshold (e.g., above 7/10) for mandatory human review. In such cases, a notification could be sent to a designated manager's Slack channel or email, providing a direct link to the claim record and ensuring critical cases receive immediate human oversight. The entire end-to-end process, from initial intake to adjuster notification, would be engineered to complete efficiently, typically within 90 seconds.