Home/AI
Eight models · One LLM Gateway · HIPAA technical safeguards

AI for revenue cycle. Eight models. One audit trail.

This is the umbrella. The AI Suite is the catalog. The AI Trust Framework is how we answer the three questions every CFO should ask any AI vendor. Our agentic AI position is where we draw the line between AI suggestion, AI with tool use, and autonomous workflows. Every page from here goes one level deeper. The first thing you should know: there is one audit log behind every model and one senior partner accountable for the outcome.

Get a free 30-day audit → See the three sub-hubs
8 production models 1 LLM Gateway AES-256-GCM PHI at rest
Three sub-hubs

One page is never enough for AI. Pick the lens that fits.

If you want the catalog of models with capability cards and per-model deep dives, go to the AI Suite. If you want the security and governance answer, the AI Trust Framework is the page. If you want our position on agentic AI in healthcare RCM, including where we sit per workflow on the autonomy spectrum, the agentic AI page is for you. All three roll up here.

SUB-HUB 01 · Catalog

The AI Suite

The eight production models with capability cards, per-model SLAs, and links into the deep-dive page for each. HCC Coding, CredPro, Denial Prediction, Eligibility, Coding AI, Insurance Discovery, Reconciliation, AR Workflow.

  • Eight model cards with measured outcomes
  • One LLM Gateway architecture diagram
  • Licensed AI vs managed suite comparison
VISIT THE SUITE → SUB-HUB 02 · Governance

The AI Trust Framework

The three questions every CFO should ask any AI vendor: where does the model run, what data does it see, how is its decision auditable. Each question is expanded with the architecture, the implementation evidence, and the HIPAA technical safeguards mapping.

  • Three CFO-grade questions and our answers
  • LLM Gateway four-layer architecture
  • HIPAA §164.312 control matrix
SEE THE FRAMEWORK → SUB-HUB 03 · Position

Our agentic AI position

Where we sit on the autonomy spectrum per workflow. The five design principles for safe agentic AI in healthcare. What we are willing to automate end to end and what we believe still needs a credentialed human in the loop. Honest about the line.

  • The five-step autonomy spectrum
  • Per-workflow placement on the spectrum
  • Five design principles for safe agentic AI
READ THE POSITION →
Six reasons CFOs trust this suite

Six structural choices that make this AI defensible in front of a board.

When we talk to a CFO about putting AI into the revenue cycle, six questions come up every time. The team made six structural choices upstream so the answers are documented before the questions are asked. Each one is a decision we made differently from the median healthcare AI vendor.

Reason 01 · Models

Purpose-built per moment

Every model is built around a specific operational moment in the revenue cycle, not a horizontal LLM with a healthcare skin. V28 HCC coding, MEAT extraction, denial pattern recognition, 270/271 eligibility, three-way ABA match. Each model knows the rules of the moment it touches.

Reason 02 · Gateway

One audited LLM Gateway

Every AI call passes through a single LLM Gateway. One audit log, one cost meter, one prompt registry, one PHI scrubber, one model router. When finance asks what AI cost this month or compliance asks for the model version behind a code, the answer is one query, not eight vendor tickets.

Reason 03 · PHI

Encrypted at rest, masked by default

AES-256-GCM PHI encryption at rest. SSN, Tax ID, and similar identifiers masked by default in API responses. Reveal endpoints write to a PHI access log with actor, role, IP, and timestamp on every call. The control matrix maps to the HIPAA §164.312 technical safeguards.

Reason 04 · Loop

Certified humans in the loop

AI suggests. Certified specialists confirm or reject. The model retrains on rejections. The audit log captures who validated and when. The 92 percent auto-accept rate is the result of three years of human-in-the-loop training, not a marketing number from a vendor demo.

Reason 05 · Evidence

Every suggestion shows its work

Every code carries the MEAT documentation that supports it. Every denial-risk score carries the payer rule that triggered it. Every credentialing extraction carries the source document line. Every output is paired with the model version, the prompt version, and the confidence score.

Reason 06 · Outcome

Paid on the outcome we promise

The AI runs inside a managed service with a senior partner accountable for a written SLA on the metric the AI is supposed to move. RAF lift, denial drop, recapture rate, credentialing TAT. We earn when your captured revenue closes, not when we hand you a license key.

The belief

AI is the worst thing to bolt on and the best thing to build in.

Most healthcare AI you will encounter in 2026 is bolted on. Someone licensed a horizontal LLM, wrapped it in a workflow that looks vaguely RCM, and shipped a product. The audit trail is an afterthought. The PHI handling is a vendor question. The model version cannot be traced to a specific code. The CFO who asks any of those questions gets a slide deck and a follow-up call.

We built the suite the other way. The audit log existed before the second model shipped. The LLM Gateway existed before we had a billing engine to plug it into. The PHI access log existed before the first reveal endpoint was wired up. The pay-for-outcomes economics existed before we had a contract template to put them on. That order matters. It is the reason a CFO can defend this AI in front of a board without a vendor on speakerphone.

  • 01
    Audit trail is a first-class deliverable
    Every model output carries source documentation, model version, prompt version, and confidence score. The same view our specialists use to validate is the view the auditor sees.
  • 02
    One choke point for every AI call
    The LLM Gateway is the single boundary between application code and any model. One log, one cost meter, one prompt registry, one place to update a control.
  • 03
    PHI handling is technical, not policy
    Encrypted at rest, masked by default, reveal endpoints audited. The policy lines up with the controls. The controls do not live in a Word document.
  • 04
    Model risk lives where the model lives
    Per-model version pinning, per-prompt versioning, per-output traceability. When a model is updated, the old version is still queryable for everything it ever produced.
  • 05
    A senior partner signs the SLA
    The AI does not own the outcome. A named senior partner does. The SLA is written on the metric the AI is supposed to move and the partner is the escalation path when it does not.
Common questions

Frequently asked questions: AI for revenue cycle.

What does ASP-RCM mean by AI for revenue cycle?
We mean eight production AI models that each address a specific operational moment in the revenue cycle, all running inside a managed service with certified specialists validating output. The AI is not a chatbot bolted onto billing software. It is purpose-built models for V28 HCC coding, credentialing extraction, denial prediction, eligibility verification, CPT coding, insurance discovery, payment reconciliation, and AR follow-up prioritization. Every model passes through one LLM Gateway that logs every call for cost, audit, and prompt-version traceability.
How do I decide between the AI Suite page and the AI hub?
The AI hub at /ai is the conceptual umbrella. It links to the AI Suite, the AI Trust Framework, our agentic AI point of view, autonomous coding, and denial prediction. The AI Suite at /ai-suite is the product catalog of the eight production models. Start at /ai if you want our position. Go to /ai-suite if you want the tool list with model cards and capabilities.
Is the AI a separate product I can license?
No. The AI runs inside the managed revenue cycle work we deliver. You do not license, install, or staff it. You get the AI capability, certified coders and credentialing coordinators validating output, a named senior partner accountable for the SLA, and a written commitment on the metric the AI is supposed to move. The economics are pay-for-outcomes. We earn when your captured revenue actually closes.
Who built these models and what are their qualifications?
The models were built by our engineering team in collaboration with the certified coders, credentialing coordinators, and AR specialists who run the workflows every day. Our HCC AI tuning lead is a CRC-certified coder. CredPro extraction was validated by a CPCS-credentialed coordinator with 14 years of provider enrollment experience. Denial Prediction was trained on more than 4 million scrubbed claim lines drawn from our own production book across ABA, behavioral health, FQHC, and acute hospital clients.
What is the AI Trust Framework?
Three questions every CFO should ask any AI vendor. Where does the model run. What data does it see. How is its decision auditable. The framework page at /ai/framework walks through how our suite answers each question with implementation evidence, architecture diagrams, and the HIPAA technical safeguards required by §164.312. The short version: HIPAA-eligible AWS, one LLM Gateway, AES-256-GCM PHI encryption, mask-by-default API responses, audited reveal endpoints, and a PHI access log that captures actor, role, IP, and timestamp on every read.
What is your position on agentic AI in healthcare RCM?
We split RCM workflows along a spectrum from scripted automation to fully autonomous agents and place each workflow where the evidence supports. Eligibility verification is scripted automation. Denial prediction is AI suggestion with human acceptance. Credentialing document extraction is AI with constrained tool use and validator handoff. Autonomous coding on tuned specialties runs at 92 percent auto-accept with a coder reviewing the long tail. We do not pretend the entire revenue cycle should be autonomous tomorrow. The agentic AI page at /ai/agentic-ai-rcm explains the spectrum and where we sit per workflow.
How is your AI safer than competitor AI?
Three structural differences. First, every model is purpose-built for a specific operational moment, not a generic horizontal LLM with healthcare branding. Second, every AI suggestion is paired with the source documentation it was based on (MEAT evidence for HCC codes, payer rule text for denials, license expiration for credentialing), the model version that produced it, and the confidence score. Third, certified specialists validate every suggestion before it leaves our hands. The audit log is the same one those specialists see when they confirm or reject.
Can we see the AI working on our data before we sign?
Yes. The free 30-day audit runs three of the eight models on your real data under a same-day BAA. HCC Coding AI runs on a 90-day encounter sample. Denial Prediction AI runs on a 90-day claim sample. CredPro runs on your provider roster. The output is a four-page written audit covering measured RAF, predicted denial rate, current credentialing TAT, and a tailored implementation roadmap. The audit is useful whether or not you hire us.

One audit log. Eight models. A senior partner.

A free 30-day audit running three of the eight models on your real data, under a same-day BAA. The output is a four-page written report covering measured RAF, predicted denial rate, and current credentialing TAT. A senior partner on the call.

Get my free 30-day audit Walk through the suite