Volume 1, No. 1The Revenue Cycle, ReviewedGallery

A new operating model for medical billing

The first AI-native medical billing company.

An autonomous revenue cycle, supervised by experts. Prepared for physicians and practice leaders.

Abstract

Medical billing remains a labor business in a software costume. Denials are reworked by hand at twenty five to fifty seven dollars each, and eligibility is verified by people on hold. Not billing software you bolt onto a team. The billing operation itself, rebuilt AI-first, where autonomous agents work every claim from the first call to the final dollar. The result is a revenue cycle that classifies, calls, scrubs, submits, posts, appeals, and reconciles on its own, with a human approving every consequential step.

Key results

Cost per autonomous verification call<$0.50
Reduction in manual calling effort~90%
Target cost per denial reworked, vs $25 to $57 industry<$5
To classify a denial and draft an appeal~90s
01Background

The denial problem is not mysterious. Industry data attributes most denials to registration errors, missing authorization, and coding mistakes, all repetitive, rule bound work.1 It is precisely the kind of work that machines perform well.

The incumbent response has been to add people, often offshore, to the phones and the rework queues. That is labor arbitrage, and arbitrage erodes. The alternative is to rebuild the operation so that agents do the calling, the classification, and the posting, natively.2

1. Denial root causes, industry distribution. 2. Venture funding into AI revenue cycle grew roughly 340 percent year over year, 2024 to 2026.

02The platform
2.1

Agents that call your payers

Autonomous voice agents dial insurance companies, navigate phone trees, sit on hold, and converse with reps to verify coverage, then hand you back clean structured data.

<$0.50 per verification call, vs $5 to $15 manual

  • +Verifies eligibility, benefits, copay, coinsurance, deductible and out-of-pocket status
  • +Captures prior authorization requirements and in or out of network status
  • +Navigates the largest national payer phone systems automatically
  • +Up to 1,000 concurrent calls, run in batches around the clock
  • +Every call recorded, transcribed, and extracted to structured fields
  • +Medical-tuned speech recognition for clinical and insurance terminology
2.2

Denials, overturned by machine

Drop in a denied claim in any format. It is classified against a five million row knowledge base, the appeal deadline is calculated, and an edit-ready appeal letter is drafted with cited evidence.

~90s to assess a denial and draft the appeal, vs 20 to 30 min

  • +Accepts structured API, free text, spreadsheet exports, remittance files, or a form
  • +Six-category denial framework with real-denial-versus-adjustment detection
  • +Deadline math with receipt-presumption, payer and state filing windows
  • +Appeal letters cite the exact reason code, payer rule, or coverage policy
  • +Expected success probability per denial, so teams work the winnable ones first
  • +Routes to the correct channel: portal, mail, fax, or peer-to-peer
2.3

One platform, the whole cycle

The system of record for everything. Charge capture, clean-claim scrubbing, clearinghouse submission, automatic payment posting, accounts receivable, collections, credentialing, and reporting.

80+ scrub rules that prevent denials at the source

  • +More than 80 pre-submission scrub rules catch denials before they happen
  • +Electronic 837 submission with status polling and full error logging
  • +Automatic remittance posting that matches payments and applies write-offs
  • +Direct EHR integration and clearinghouse connectivity
  • +AR aging, collections, statements, payment plans, and deductible tracking
  • +More than 40 report types and real-time financial dashboards
03Methods

Each claim is processed in seven stages. An agent performs the work at every stage; a human reviews before anything is filed or paid.

01

Capture

Intake agent

Charge and clinical intake, coded per visit

02

Verify

Voice agent

Voice agent calls the payer and confirms coverage

03

Scrub

Scrub agent

80+ rules check the claim before it leaves

04

Submit

Submit agent

Electronic 837 to the clearinghouse

05

Post

Posting agent

Remittance matched and posted automatically

06

Appeal

Denial agent

Denials classified and appealed in ~90 seconds

07

Settle

AR agent

Paid, reconciled, and reported

04Results
Administrative
90%
Authorization
88%
Coding
80%
Medical necessity
67%
Eligibility
66%
Timely filing
25%
Figure 1. Appeal success rate by denial category. Denials are scored and prioritized by expected recovery, so the most winnable cases are worked first.
Reference setRecords
Procedure-to-procedure bundling edit pairs4,493,738
Diagnosis codes74,719
Procedure and supply codes9,068
Remittance advice remark codes1,198
Claim adjustment reason codes, denial-tagged308
Local coverage determinations949
National coverage determinations357
Medicare fee schedule entries19,000+
Table 1. The clinical and regulatory reference corpus. Codes, coverage policies, and deadline rules are retrieved by exact lookup; the model judges and drafts but does not invent the facts.
MeasureValueBasis
Cost per autonomous verification call<$0.50verified
Reduction in manual calling effort~90%verified
Target cost per denial reworked, vs $25 to $57 industry<$5target
To classify a denial and draft an appeal~90sverified
Rows in the clinical and regulatory knowledge base5M+verified
Concurrent payer calls at peak1,000verified
Pre-submission scrub rules80+verified
Report types and live dashboards40+verified
Table 2. Selected measures with their basis. Verified measures are observed in production; target measures are stated goals, not guarantees.
05Safeguards
01

Deterministic before generative

Codes, coverage policies, and deadline math come from exact database lookups. The AI judges and writes. It does not invent the facts, so it cannot hallucinate a code or a deadline.

02

A human on every consequential move

Agents draft appeals and surface recommendations. A person approves before anything is filed. Autonomy with a hand on the wheel.

03

Every action is auditable

Each classification, call, and posted payment records the evidence and the model that ran it, so any decision can be reviewed end to end.

04

Built for protected health information

Encryption in transit and at rest, role-based access, and activity logging throughout. Claim identity stays operator-controlled.

06Comparison
The old waymedicalbiller.ai
Offshore callers on hold all dayVoice agents call payers, 1,000 at once
Denials reworked by hand, one at a timeDenials classified and appealed in ~90 seconds
Software bolted onto a manual teamAgents are the operation, not an add-on
You manage the peopleYou watch the dashboard
07Correspondence

Is AI safe for claims and protected health information?

Yes. The facts come from the knowledge base by exact lookup, the AI only judges and drafts, every action is logged, and a human approves anything that gets filed. Encryption and role-based access run throughout.

Will it hallucinate codes or deadlines?

No. Codes, coverage policies, and deadline math are exact database lookups, not model guesses. It is not the model's job to know the codes, it is the database's job.

Do I have to replace my current systems?

No. The platform is the system of record and it connects to the clearinghouse and EHR you already use. We meet your data where it lives: API, spreadsheet export, remittance file, or direct integration.

Is this just an offshore team with a chatbot?

No. This is software doing the work, supervised by a small expert team. Not a large team holding a tool.

This is what a billing company looks like when it is born AI-native.

Request access