The Problem
A physician submits a prior-authorization request for a biologic medication. Your team manually checks the request against a long clinical policy document: is the diagnosis an approved indication? Is step therapy documented? Is the prescriber valid? Get it wrong and you risk patient harm, provider abrasion, and regulatory exposure. The work is slow, and the No Surprises Act plus tightening CMS rules have raised the stakes on both accuracy and turnaround time.
Why a Black-Box LLM Fails Here
A clinical determination produced by an opaque model is indefensible. You cannot tell a provider "the AI denied it" and you cannot show a regulator the clinical criteria the decision was based on. Healthcare determinations must be traceable to a documented policy. An LLM that fuses reading and deciding gives you neither traceability nor reproducibility.
The Governed Architecture
- The LLM reads. It extracts structured fields from the request — diagnosis code, requested medication, prescriber identifier, documented step therapy. It never makes the coverage determination.
- A DMN policy table decides. Your clinical coverage policy as a deterministic table: approved indications, step-therapy requirements, and prescriber validation map to an approval, a denial, or a referral for clinical review.
- BPMN routes. Clean approvals are issued with an authorization number and validity window; anything that does not clearly meet criteria is routed to a clinical reviewer — a human stays in the loop on the hard cases.
Try It Live
Go to archiet.com/agents and open the Prior Authorization agent. Type:
Patient DOB 1978-03-15. Diagnosis: moderate-to-severe plaque psoriasis,
ICD-10 L40.0. Requested medication: Adalimumab 40mg biweekly. Prescriber NPI
1234567890. Step therapy documented: methotrexate for 6 months with
inadequate response.Click Run governed agent:
1. The LLM extracts (shown read-only):
{
"diagnosis_code": "L40.0",
"requested_drug": "adalimumab",
"prescriber_npi": "1234567890",
"step_therapy_documented": true,
"step_therapy_drug": "methotrexate"
}2. The DMN policy table evaluates and fires: diagnosis matches approved indication AND step therapy documented AND prescriber NPI valid → approve for 12 months. Rule PA-BIOLOGIC-001.
3. BPMN routes to issue an authorization (e.g. PA#2026-PA-88341) valid for 12 months.
4. The audit trail records the firing rule, the clinical criteria met, the extracted fields, and the timestamp — a record you can show a provider or a regulator.
The Regulatory Angle: No Surprises Act and CMS Prior-Auth Rules
The No Surprises Act and CMS interoperability rules tighten both the turnaround-time and the transparency requirements for prior authorization (CMS is mandating electronic prior auth and faster determinations). A governed agent gives you decision turnaround in seconds for clean cases, a documented clinical basis for every determination, and a logged trail proving the criteria applied. Because the determination is a DMN rule tied to your clinical policy — not a model guess — it is auditable and HIPAA-friendly (the decision basis is the policy criteria, kept with minimal necessary data).
Generate This for Your Own System
Describe a prior-authorization or utilization-management workflow in Archiet's Blueprint Wizard and the governed PA agent is generated with your coverage criteria, step-therapy rules, and validity windows extracted from your PRD. Engine, policy model, endpoints, and UI come wired and auditable.
The same governed pattern powers invoice approval, loan adjudication, FNOL triage, refund/retention, and KYC/AML in the gallery.