Upload a First Notice of Loss
Drop a PDF or image here, or click to choose
…or try a sample form
Upload a form or pick a sample. Textract reads it, three models score it, and the routing decision appears here.
Recent triage decisions
Loading…
How it works
- Intake – A claim arrives either as JSON through the API or as an FNOL form (PDF or scanned image). Forms are read by Amazon Textract.
- Model 3 · Narrative analysis – A text classifier reads the loss description and adjuster shorthand. It confirms the loss type and flags injury, attorney involvement and possible total loss. These flags fill in anything the form left blank.
- Model 1 · Fraud risk and Model 2 · Severity run in parallel on the enriched claim. Fraud returns a score plus the factors that drove it. Severity returns the expected cost, a P10–P90 range and a suggested reserve.
- Routing rules combine all three into one of four queues: SIU referral, complex / senior adjuster, standard adjuster, or fast-track. The rules are plain, editable thresholds, not a black box.
- Every decision is stored with its full model output for audit.
All three models are hosted on Amazon SageMaker (serverless inference) and exposed through API Gateway with an API key. They were trained on 5,000 synthetic claims; no real policyholder data is used.