The problem
Payer AI denies at machine speed, and the fight has gone public (the nH Predict disclosure order in March 2026 put algorithmic denials under legal scrutiny, and the overwhelming majority of appealed AI denials get overturned). The cheapest, fastest reversal point is not the appeal letter, it is the peer-to-peer: a 15-minute call between the treating clinician and a payer medical director.
And clinicians walk into it cold. The slot lands between patients, the payer's medical-necessity criteria are a PDF nobody has open, the supporting evidence is scattered across the chart, and the first pointed objection ("conservative treatment was not tried") goes unanswered because nobody had time to look. Denial vendors show up after the P2P is lost, with a letter.
How it works
You hand the autopilot the denial and the case. From there the span does the work:
- triages the denial (reason codes, deadline, whether P2P is even the right lane),
- retrieves the payer's own medical-necessity criteria for the denied service,
- pulls the relevant clinical documents, behind a consent gate that fails closed before any chart access,
- maps the record evidence to each criterion, and flags the criteria with no support (honestly),
- compresses the case into the opening clinical narrative,
- anticipates the objections the medical director is likely to raise, each with a record-grounded response,
- gives an honest viability read (strengths, gaps, overturn-likelihood band),
- scrubs PHI, then generates the one-page briefing.
The whole thing is composition on AYA's verification spine (nine existing capabilities wired together as data, the same rails already serving our prior-auth and denial line). The span never contacts the payer and never joins the call. The conversation stays yours.
What you get
A one-page P2P briefing per scheduled call: the opening narrative, the payer's criteria with your evidence mapped to each, the anticipated objections with responses, the honest gaps stated plainly, and the specific ask. Plus a separate viability read so you know before the call whether this one is strong, salvageable, or a documentation problem. PHI de-identified on egress.
Who it is for
Specialty practices with heavy prior-auth burden (oncology, cardiology, ortho and spine, imaging, infusion), utilization-review and physician-advisor teams at hospitals, and the RCM and denial-management vendors that staff P2P calendars. It sits directly on the Luigi prior-auth line, downstream of the PA Wizard and beside Denial Defense.
Pricing
Pricing is a set of hypotheses we are validating, not a committed price sheet.
- Per briefing, one briefing per scheduled P2P, priced well below the revenue at stake in the denied service.
- PA desk, a monthly per-clinician or per-desk subscription, so every scheduled P2P on the calendar gets a briefing automatically.
- Platform, wired into the full PA and denial pipeline (PA Wizard, Denial Defense, and P2P Prep as one line), denial in, briefing out, outcomes tracked across payers.
Proof (dogfood)
We run the autopilot on ourselves first. AYA feeds its own synthetic denial casebook (synthetic denials with reason codes, synthetic charts, and on-file payer criteria from the Luigi test fixtures) through the exact deliverable a customer would buy: triage, criteria, evidence map, narrative, objections, viability, scrub, briefing. All of it as a structural dry run with no real patient data and no payer contact.
What is actually true today: the deliverable pattern (AYA-3SPN-LUIGI-P2P-PREP-AUTOPILOT-v1) is on disk, PQS-scored at 91 (Grade A), and all nine of its composed capabilities resolve to existing atoms already serving the Luigi PA and denial estate (denial triage, appeal evidence, viability assessment, PHI scrub).
Honest note
This is a structural build, not a live-customer result. When a payer's guideline is not published or on file, the briefing declares the criteria gap instead of quietly substituting a generic one. The widely cited overturn statistic is about AI denials in general, not a promise about this tool, and our own overturn-rate impact is unmeasured until a design-partner run. Real runs need EMR access under consent, a payer-criteria corpus, and a booted system. The call itself, and the clinical judgment in it, belong to the clinician. We would rather tell you that than sell you a script.
*This page is a specification. The capability it describes is not built yet, and nothing here is a claim that it runs today.*