The average physician’s practice spends **14 hours every week** on prior authorization. Navigating payer portals, re-entering patient data, uploading clinical documents, checking statuses — over and over.

The work is predictable. It’s repeatable. And until recently, it still required a person at every step, because no automation could reliably handle the complexity. Asteroid’s AI browser agents change that. Prior auth is also one stage of a longer chain: the same portal work continues downstream in [medical billing automation](/content/blog/medical-billing-automation/index.html). Both halves of the workflow described here now ship as catalogued templates: [PA submission with attachments and clinical Q&A](/content/workflow-library/prior-auth/pa-submission-attachments-clinical-qa/index.html) and [submission and status on the payer’s own portal](/content/workflow-library/prior-auth/pa-submission-status-payer-portal/index.html).

## The problem

Prior auth looks simple on paper: submit the right information to the right payer, get a decision. In practice, it’s a highly branching workflow full of edge cases.

Every payer has a different portal — dozens of them — each with its own form structure and MFA flow. Forms change dynamically based on procedure type, place of service, and member plan. Some requests route to third-party reviewers mid-submission. Some CPT codes don’t require auth at all, but you only find that out after logging in.

### Every payer is different

Different portals, different forms, different MFA flows. No two submissions are the same.

### Dynamic forms

Fields change based on procedure type, place of service, and member plan — mid-submission.

### RPA breaks constantly

Traditional scripts are built for the happy path. One unexpected field and someone has to step in.

## What AI makes possible

Asteroid’s browser agents handle prior auth the way a trained person would — by understanding the goal, not just following a script.

01

### Handles the edge cases

Agents navigate dynamic forms, handle MFA (including TOTP and email-based codes), detect when auth isn’t required and stop early, and recognize when a case needs to route elsewhere. Edge case handling is configured in natural language — describe how you want exceptions handled, and the agent follows those preferences consistently.

02

### Gets faster over time

As patterns stabilize, Asteroid compiles optimized execution scripts. Agents get more reliable and faster with every run — not slower as portals shift.

03

### Built for a critical process

Prior auth touches PHI at every step. Customers provision their own portal credentials, stored in an encrypted vault. Full execution logs, audit trails, and SOC 2 and HIPAA compliance are standard.

04

### Reliable at scale

Asteroid agents are built for production healthcare workflows. They run continuously, handle volume without degradation, and flag exceptions cleanly rather than failing silently.

## The full prior auth workflow

Asteroid covers the complete prior auth lifecycle with purpose-built agents for each stage (the [prior authorization workflow library](/content/workflow-library/prior-auth/index.html) has the full template list):

01

### Pre-check

The agent logs into the payer portal, verifies whether auth is required for the given procedure codes and member, and returns a per-code result. No unnecessary submissions.

02

### Submission

The agent handles the full form: payer selection, CPT codes, provider NPI, patient demographics, diagnosis codes, service dates, document upload, and final submission. Output is structured data — not a screenshot.

03

### Status tracking

Agents check batches of pending authorizations on a schedule, returning normalized statuses (Approved, Denied, Pending, Partially Approved) via webhook into your existing system.

## Why teams use Asteroid for prior auth

### Any portal

Browser agents work at the UI level. If a human can log in and use it, the agent can too. No API required.

### Natural language

Exception handling, routing logic, and edge case preferences are set in plain English. No engineering work to adjust behavior.

### Infinite scale

The same agent that handles 20 submissions handles 2,000. Volume goes up, staffing costs don’t.

[Davide Locatelli](https://www.linkedin.com/in/davidelocatelli13/)

Research Engineer

Davide Locatelli is a founding research engineer at Asteroid focused on browser and computer use agent behavior. He brings a healthcare background from his previous role at an AI medical-scribe company, where he worked closely with many healthcare organisations.