Nodagentry

Claims desk: an AI workflow for first notice of loss triage

A first notice of loss arrives: a rear-end collision in a parking lot, a $2,800 estimate, photos but no police report, and the third claim on the same vehicle this year. Someone has to decide whether to pay, investigate or deny, and the file note has to explain why.

This workflow does the reading and drafting. It extracts the facts, looks up the policy, routes to one of three drafting agents and then stops. Nothing that moves money or denies a claim leaves the graph until an examiner approves it.

Open this workflow in the editor

Free in Sim. No account or API key.

Sample input

FNOL: rear-end in a parking lot. Bumper + taillight. Insured estimate $2,800. Third claim this year on the same vehicle. Photos mentioned, no police report. Decide pay / investigate / deny.

How the workflow runs, step by step

  1. 1. Start

    Claim in

    The claim note starts the run. The policy snippet is attached alongside it.

    Next: Extract

  2. 2. Agent

    Extract

    A cheap model pulls the structured facts out of the note and the attachment.

    Next: Policy lookup

  3. 3. Tool

    Policy lookup

    Searches a claims playbook, standing in for your policy system.

    Next: Disposition

  4. 4. Router

    Disposition

    Keyword rules choose the disposition. Prior claims, fraud signals or SIU go to investigation, which is also the safe default.

    Next: Pay draft (pay), Investigate brief (investigate), Deny draft (deny)

  5. 5. Agent

    Pay draft

    Drafts a payment worksheet for a clear-liability claim.

    Next: Examiner approve

  6. 6. Agent

    Investigate brief

    Drafts the examiner and SIU questions for a claim that needs a closer look.

    Next: Examiner approve

  7. 7. Agent

    Deny draft

    Drafts denial language. It never sends on its own.

    Next: Examiner approve

  8. 8. Human approval

    Examiner approve

    Every disposition waits here. An examiner approves or rejects before money moves or a denial goes out.

    Next: File note

  9. 9. Result

    File note

    The approved file note, ready for the claim system or Slack.

Why build it as a graph

Claims work needs an audit trail more than it needs clever prompting. A graph makes the decision path visible: which facts were pulled, which branch was taken and who approved it, with the examiner gate built into the workflow rather than left to habit.

Things to try

  • Run the sample in Sim. The third-claim history sends it down the investigate branch.
  • Rewrite the claim note, for example as a clear-liability claim with no prior history, and see which branch the router picks.
  • Compare the cost per step: extraction and routing are cheap next to the drafting agents.
Run it free