How Plutus Health built ABA revenue cycle management around process: a 30-day diagnostic, 200 KPIs, and a payer intelligence engine that guardrails AI.
Key Takeaways
- Most revenue cycle vendors treat denial symptoms rather than root causes: Thomas John founded Plutus Health in 2008 around an end-to-end RCM model in which the firm is paid only when the provider collects, and every engagement opens with a 30-day diagnostic that benchmarks a practice against MGMA and HFMA medians and sorts a year of denials by root cause across roughly 200 KPIs.
- Plutus will not bill an ABA claim that lacks compliant documentation: John argues that most Medicaid audit exposure in applied behavior analysis traces to poor documentation rather than intent, and says the firm will end a client relationship rather than submit a claim the underlying record cannot support.
- Claims automation runs in layers, and human review still closes the gap: APIs, robotic process automation bots, and voice agents work a claim in sequence, the last 5 to 10 percent routes to staff, and every submission and every appeal passes under human review before it goes out.
- Payers are acquiring the same AI tools while Medicaid rates compress: John expects claim-level AI scrutiny and reimbursement cuts to arrive together, leaving providers almost no room to write off denied revenue and pushing smaller clinics toward intermediaries that can carry the technology cost.
There is a conversation Thomas John has had often enough that it has hardened into policy. A client sends over a batch of claims. Somewhere in the batch sits a session that happened: a child attended, a technician worked, hours were spent. The note behind it is thin. A missing parent signature, a description too vague to support the units, an activity that does not match the code attached to it. Plutus Health, the revenue cycle management firm John founded in Dallas in 2008, sends the batch back.
“We will not bill a claim even at the risk of terminating a contract and stepping away from that relationship,” John said in an interview with Acuity Media Network. “We will never bill a claim if there is no underlying documentation that meets the payer standard.”
Coming from a vendor, that is a strange sentence. Plutus is paid on collections, so refusing to submit is refusing to be paid. It is also the clearest expression of how John built the business, and it lands where the cost of the alternative has become measurable: federal auditors have recommended more than $123 million in refunds across four state Medicaid ABA programs, and the HHS Office of Inspector General audit series that produced them turns almost entirely on records unable to substantiate what was billed.
How Plutus Health Built an Outcome-Based RCM Model on Five Pillars
John did not set out to build a revenue cycle company. He came from technology, and in 2007, after a previous venture of his was acquired, a Dallas physician asked him to look at some problems in his practice. He approached it through a technology lens, then kept going deeper into the mechanics of how the practice actually got paid. A team formed around that one client, then around many more.
Eighteen years later, by John’s account, Plutus collected roughly $1 billion on behalf of its clients in the twelve months ending in April 2026, with a global staff of more than 1,700. The company says it serves providers in more than 25 states, holds denial rates under 5 percent, and posts a 99.5 percent net collections rate. MGMA puts first-pass denials at 8 percent for single-specialty practices and counts anything below 5 percent as best practice. He describes the business as resting on five early decisions. Plutus would run the entire revenue cycle rather than a slice of it, and would be paid on outcomes: “We only got paid if our customer got paid,” he said, which is workable only if you also own the first commitment. Technology would come first, well before anyone called it AI. The firm would specialize rather than serve everyone, and applied behavior analysis was an early choice. And the firm would hold a documentation floor it would not bill beneath, the decision that became the policy of sending thin batches back.
What he saw in his competitors, then and now, was a narrower field of vision. “A lot of my competitors in the marketplace were just addressing the symptoms and not the root cause of the problem,” he said. Sending a claim, posting a payment, working a denial: all necessary, none sufficient. Cash arrives consistently, in his framing, only as a function of what happens upstream, in prior authorization, in payer credentialing, in session note documentation, in whether the schedule matches which clinicians are contracted with which plans.
The 30-Day RCM Diagnostic: Benchmarking ABA Denials Against MGMA and HFMA Medians
Every Plutus engagement opens with a 30-day diagnostic that maps a provider’s operation against MGMA and HFMA medians. What it surfaces, John says, is rarely news at the level of the individual claim. Providers can usually see those. What they cannot see is the shape of twelve months of them.
Denials are the anchor metric, and the work is in decomposing them. A practice running 10 percent denials wants to know which portion is prior authorization, which is eligibility, which is benefits, which is coordination of benefits, which is credentialing. Plutus tracks roughly 200 KPIs and aggregates them by location, by state, by clinician, by payer. “We slice and dice their data six ways to Sunday,” he said. Most provider leaders, in his experience, have never seen their own denials at that level, because nobody has assembled it for them.
Olympus AI and the Order of Operations: APIs, RPA Bots, Voice Agents, Then People
Olympus, the platform Plutus now markets as an AI product, started as something considerably less glamorous. “We needed to have everyone in our organization sing off the same hymn book,” John said. It was a workforce management system: everyone logged into it, every claim moved through it from step to step. The workflow came first, and the workflow is what made everything after it possible.
Automation went in next, one piece at a time. An API to check eligibility. An API to check claim status. Then, as models improved, AI that could assemble a credentialing package and push it into a payer portal, or build a prior authorization package and get it approved. The company now counts 25 specialized AI agents spread across seven stages of the revenue cycle, from eligibility through claims, denials, appeals, and accounts receivable follow-up. The agents run in production across more than $1 billion in annual collections, and roughly a quarter of the firm’s total work is now handled by automation and agentic AI. Any given task now runs in descending order of cheapness. An API goes first and might return half of what is needed. A bot picks up another 10 or 15 percent. A voice agent calls the insurer and talks to a human for another 15 or 20. The last 5 to 10 percent is assigned to a person.
His objection, he is emphatic, is not to AI, which Plutus uses across prior authorization, credentialing, scheduling, and denial analysis. It is to the pricing model that usually arrives with it. “There are vendors out there that say, I’ll give you an AI prior auth tool and charge you 12 bucks for every auth you do,” he said. “We get paid when you get money in the bank.” Solving 20 percent of a provider’s problem and invoicing for it is a different business than the one he is in. “We are not delivering a service, we’re delivering an outcome.” Other entrants have made versions of this argument in ABA billing, including Camber.
The Payer Intelligence Engine: Why Narrow, Real-Time Payer Data Beats a General AI Model
What John seems proudest of is the least visible component. Plutus calls it the Payer Intelligence Engine, or PIE. “Think about it like the brain of Plutus Health and Olympus,” he said.
PIE is a vector database holding what the firm has learned about payer behavior over 18 years: its standard operating procedures, its accumulated best practices, and the payer guidelines it has pulled from portals and bulletins. “We’ve taken tens of thousands of pages of payer policy information and converted them into meaningful rules,” John said. A second layer, built from millions of historical claim transactions, records how specific payers behaved in specific situations and converts that into probability models. Drafting an appeal for a prior authorization denial from Blue Cross Blue Shield of Texas, the system works from what Plutus already knows that plan requires.
The point, as John frames it, is restraint rather than capability. PIE functions as a guardrail on what the model is permitted to conclude. He contrasts it with a general model trained on public data, which would answer a question about Indiana Medicaid’s guidance from last week using information from a year ago, or invent an answer outright. Plutus pulls from payer websites continuously. “We are guardrailing it with the specificity of our business,” he said.
Underneath the engineering sits a claim about the work itself. “RCM is not a very probabilistic business model,” John said. “It’s a very deterministic approach.” Compliance requirements, payer rules, and documentation formats decide outcomes. “You either have prior auth or you don’t.” A field with that little maneuvering room rewards accurate current information over clever inference, which is an argument for narrow, freshly updated models. ABA has been having a version of that argument elsewhere, where CASP practice parameters and state law try to fix where clinical AI stops.
Human-in-the-Loop Claim Review: Where ABA Billing Automation Still Hands Off
Ask John where the automation fails and he does not describe hard problems. He describes logistics.
Small payers without an API. Portals that cannot be queried, requiring a phone call. Insurance representatives who will not engage with a voice agent once they realize what they are talking to. And, occasionally, a payer that changes a rule, mails a letter announcing it, and never posts it to the portal, leaving a window before PIE has scanned and absorbed it. “It’s not the complexity of the problem statement itself,” John said. “It’s just the lag.” Humans get the exceptions because humans can go find the letter.
Beyond the exceptions, Plutus keeps a person in the loop by policy. “We still are validating everything with humans before we send it out,” John said, calling it a difference between his firm and its Silicon Valley peers. Every claim submission and every denial appeal gets human eyes. He is candid that this is partly about customer comfort, and equally candid that it will not hold forever. “Eventually, in a year or maybe in several months, we might keep reducing the number of human eyes on each claim,” he said. In his description, the review concentrates rather than disappears: as confidence data accumulates on routine claims, human attention shifts toward exceptions and newer payer scenarios.
Medicaid Fraud Scrutiny in ABA Is Mostly a Documentation Problem
Automation in ABA billing lands in a charged moment. Federal auditors have been through four state Medicaid programs, House investigators have opened inquiries in eleven states, and the word fraud attaches itself to the sector in coverage that often conflates a documentation deficiency with a criminal one. It can tighten documentation, or it can propagate a single error across thousands of claims before anyone notices.
John separates the two problems. Genuine bad actors exist in ABA as they exist in home health and in laboratory services, and no system eliminates them. But that is not what he sees most of the time. “They’ve delivered service, but they just did a bad job of documenting it,” he said of the majority of providers who run into trouble, and he treats that as a training gap rather than a criminal one.
John sits on both ends of the documentation problem. Through Artemis ABA, a practice management and billing platform he also founded, his companies own the layer where session notes are created, not just the layer that bills them. His remedy moves the intervention as far upstream as it will go. Plutus runs AI agents that watch Registered Behavior Technicians as they document and prompt them in the moment: this payer requires that element, you have missed a signature here. Feedback arriving six weeks late arrives at the wrong time. The same logic drove the practice of pre-billing audits before any of it was automated, when human billers read every document behind every claim. Today AI takes the first pass and a person still reviews it against an eight-to-ten-point check. A claim that fails goes back to the Board Certified Behavior Analyst, the technician, and the clinic owner. That upstream discipline is the one program integrity reviewers say providers most often skip.
Waiting is the alternative, and it is a conversation nobody wants. “You don’t want to wait six months from now when an insurance company comes for your medical documentation and then you are scrambling, because the RBT that did that activity has already quit and left,” he said. “That’s when people get into trouble.”
Payer AI, Medicaid Rate Compression, and the Future of ABA Revenue Cycle Management
John dates his own change of mind to roughly a year ago. Robotic process automation had been shifting the industry from labor toward technology for five years, but large language models capable of reading an entire claim file were, he says, a different order of thing. Plutus signed business associate agreements with its AI model providers, among the earlier RCM firms to do so, and built a pipeline that masks protected health information before anything reaches a model, then unmasks the output on return, all under those agreements. The firm now runs several models depending on the task.
Scope is what impressed him. No human, he points out, can read every record and session note attached to a claim, compare it against payer policies running to tens of thousands of pages, and do it at volume. “The key thing is to bring all of that together and orchestrate it at the right time, at the right combination of factors,” he said. He was equally quick to name the failure mode, hallucination, which is what PIE exists to constrain. His summary of the moment was blunt: “If we don’t pivot ourselves and get to the forefront of this business, we’re going to become toast.”
His forecast rests on two trends arriving together. Payers are buying the same capability, and John expects them to move from rule-based auto-adjudication to mapping underlying documentation and an organization’s history. “Their likelihood of scrutinizing your claim at an individual claim level will significantly increase,” he said. Meanwhile Medicaid budgets are contracting, most visibly in ABA, where Indiana cut ABA reimbursement 6 percent in April 2026 and, under bulletin BT202627, will apply another 4 percent in April 2027 while operating costs hold steady or rise. “They’ll be out of business if they can’t squeeze out every penny from every claim,” he said.
That failure mode is not hypothetical. An Indianapolis ABA provider shut down in June after what its closure letter described as a three-month Medicaid payment hold, a managed care network termination, and credentialing backlogs that left it delivering care without payment.
Where that leaves a firm like his, in John’s telling, is closer to infrastructure than to services. He reaches for the credit card processor as an analogy: millions of transactions at low margin, screened continuously for errors, invisible when it works. Smaller ABA providers will not be able to buy the technology this environment demands, and he sees the RCM company’s role as lending them scale. “There is basically a fundamentally adversarial relationship between a payer and provider,” he said, and both sides are now arming for it.
He ends where the money ends up. “If you don’t get paid, you’re not going to be able to pay your employees, and if you can’t pay your employees, they’re not going to show up to work,” John said. “And if that doesn’t happen, then how do you take care of the kids?” It is a self-serving argument in that it describes his own market. It is also the chain of causation that decides whether a child keeps a therapy appointment.
Disclosure: Plutus Health is a sponsor of Acuity Media Network. Sponsorship does not confer editorial approval over framing, headlines, or the contributions of other sources.
