Payor Desk

Northlight Patient Access

Fictional data
Demo built for PRO-spectus by The AI Pipe

The appointment moved twenty days. The letter of agreement that priced it did not.

Ten invented patient cases with an infusion or a fill already on the calendar, the eligibility response the plan returned on each, four payor coverage policies and administrative manuals, and one programme procedure. The agent reads them the way a payor specialist reads them, computes the member exposure in code, dates the next action against the clause that governs it, and drafts the letter that goes to the plan.

What this shows

The problem: A scheduled service moves. Every document in the file stays correct, and one of them quietly stops applying. Nobody re-reads the agreement period, because nothing tells them to.
The result: A verification sourced line by line, the member exposure computed in code rather than asserted, the binding deadline taken from the clause that produces it, and the letter drafted from the same material.
The limit of the proof: Every patient, plan, therapy, code and figure here is invented. What is scored is the deterministic determination engine, 21 of 24 on a labelled set, not the model's own writing. Nothing touches a payor portal, a clearinghouse or a patient record.
What a pilot would need from your team: Twenty to thirty closed cases with the outcome you actually reached, the payor documents and the internal procedure that govern them, and one specialist who signs off every output before it is sent.
What it does not do: No clinical decision, no claim, no appeal after a denial, and no member price where the plan does not cover the therapy.

The desk today

The documents

Live agent, claude-sonnet-5. Every figure comes from a tool call and every citation is checked before it renders.