Editor’s note: This is the 4th of 5 essays in The Cycle, my series on healthcare’s revenue cycle. Ongoing writing to continue here.

Consider the claim: two charges in one word. A chronicle (the record of the care, what was done to you and why) and a contention (the provider’s sworn assertion that a sum is owed for it). One document that tells the story of your care and opens the argument over its price. Captured, coded, checked, charged, and then, challenged - and a contested claim gets caught in a cycle.

Healthcare finance moves $5.3 trillion a year, most of it as a claim - and a claim is a bill in dispute before it is ever paid. It somehow runs on backwards economic logic: the care first, the price last, the payer keeping the right to dispute the bill after the fact (whoops - wrong code, not medically necessary, out of network). One of my idols, the legendary Princeton economist Uwe Reinhardt, distilled the whole problem down to “It’s the prices, stupid”... but what is often overlooked is that a price has to be agreed, and what healthcare agrees on is the rates - while what is owed is asserted and answered, claim by claim. A credit-card swipe is consent before the fact - a claim is a trial after it.

The initial instinct, always, is to regulate the reimbursement - though not to set the price itself (the anti-capitalist sin American politics won’t commit, leaving the price to the market) - so the rules pile up around it: standards, disclosures, deadlines, courts, each promising to bring the two sides to agreement. And this is how a country came to settle the economics of its largest industrial market through a claim - one contest at a time.

A trial, not a transfer

Nowhere else does moving money require a trial at all. In 2025 the Federal Reserve’s Fedwire system moved $1.15 QUADrillion (a number so incomprehensible I think I’ve typed “quadrillion” twice in my life, including that one) and adjudicated not one of those transfers. The securities clearinghouses settled 3x more than that; the network behind every payroll and direct deposit moved $93 trillion more.1 Sums that dwarf all of American medicine, cleared fast and clean, because every one of those rails has the one thing healthcare does not: a price agreed before the money moved.

Healthcare runs the opposite of other financial infrastructure. A disputed card charge is rare enough that the card networks punish a merchant who lets it cross a fraction of a percent; the automated network caps unauthorized returns at 0.5%; even stock trades fail to settle only 2-3% of the time.1 In healthcare, ACA marketplace insurers denied nearly 20% of in-network claims in 2023 — 86 million denials, of which patients appealed less than 1%. Everywhere else, a disputed payment is the exception; in healthcare, it is the whole job of the revenue cycle.

Insurance looks like the exception - it too pays after the fact and argues constantly. But other insurance fixes its price at underwriting, and even its payout fights (the adjuster’s estimate against the body shop’s) settle against a market that exists outside the policy — the going rate for a fender, a roof, an hour of a contractor’s time. Healthcare fixes only the rates, so it fights on two fronts — whether the care was warranted and what it was worth — with no outside market to referee either one. Other insurance argues the event; healthcare argues the event and the bill.

Everything but what’s owed

Govern a transaction and you reach for the same few tools: a standard (both sides bill in one language), disclosure (the terms sit in the open), a clock (the answer comes in time), and when those fail, a court to settle what is left. Beneath them all should sit one more — an agreement on what is owed, struck before the money moves — and the straight road to that one runs through the cardinal sin of a market economy: setting the price.

Whatever you think of the sin (set the price, so the argument goes, and the money for the next drug dries up), American health policy has kept the commandment.2 So it standardizes the codes, posts the prices, puts the gate on a clock, even builds the court for disputes - everything but settle what is owed. The rates themselves were settled all along (commercial insurers negotiate them into contracts before any patient walks in; Medicare just posts its own). But a rate only names the number; whether it is owed is what every claim re-opens.

What is owed stays unsettled for a plain reason: the country has hated every fix more than it hates the fight. And each rule arrived with a forecast attached — the savings it would book, the behavior it would change — and the forecasts are worth keeping score on as we go: where a prediction breaks, and which way it falls, both point to what the rule actually did.

The map below sets every rule in this essay against the fourteen stations of the cycle - where each one lands, and the gap at the center where none of them do. (It’s interactive - open the live version.)

Start with the one that sounds easiest: the standard - the rule that fixes what a claim is even allowed to say.

A shared language

A standard sets a shared language - the clearest way to get two parties to reach an agreement. In the revenue cycle that language is codes; nothing can be priced until it has one. When a diagnosis, a procedure, a supply each resolve to one standard code that means the same thing on both sides of the bill, the claim can settle almost without a human - the payer reads what the provider wrote, and the two numbers match. That is the whole promise: a standard turns a negotiation into a lookup. It is also the first thing anyone reaches for when the cycle feels like chaos - just standardize it.

And the cycle already runs on a stack of them: ICD for the diagnosis, CPT and HCPCS for what was done, the X12 837 the claim rides on, the UB-04 behind it. All of them converge at coding, the station where someone reads the clinician’s chart and translates the care, line by line, into the codes a claim is built from. And coding is where the biggest code-set change in the industry’s history landed: the roll-out of ICD-10.

Early in my healthcare career, ICD-10 was finally changing over — one narrow technical swap, mandated by HHS under HIPAA’s authority to pick the industry’s code sets — and it still took America a decade of false starts: the go-live slipped, then slipped again when Congress kicked it another year inside a Medicare bill. I was a young consultant at Deloitte then, and one of the services we were pushing was change management for ICD-10. The rest of the world had been running on ICD-10 for some two decades; why wouldn’t we adopt it too?3 When the switch finally landed, on October 1, 2015, the working vocabulary of American diagnosis went from roughly 14,000 codes to 68,000. The trade press had spent two years predicting catastrophe.

Every forecast pointed the same way. The American Medical Association called the switch a “crushing burden,” citing a study that put small-practice conversion as high as $226,000; the trade press warned of coders seizing up, one claim in five rejected, billions in cost industry-wide. But the complaints came too early to be about the workload: providers were calling the standard unfair before a single new code had been filed. They weren’t really fighting the vocabulary; they were fighting the scrutiny - every new code was one more thing a payer could question.

None of the predictions came to pass. Denials held flat - the Medicare rejection rate came in at 9.9%, a hair under its pre-switch baseline - and a 2018 analysis put the real small-practice cost near $1,200, not the hundreds of thousands the AMA had forecast. Modern Healthcare’s verdict on go-live day: “like Y2K, so far.” (The one claim in five did arrive eventually - ACA marketplace insurers deny about that share today - just in a different market, a decade on, with no code switch to blame.)

But a standard is supposed to take friction out - one language, fewer things to argue about. ICD-10 put friction in. Sixty-eight thousand codes meant more to specify, more to document, and more to dispute - every new distinction is one more thing a payer can deny and a provider must defend. An entire documentation industry grew up to work the new specs - a $4.5B market by 2023. And it cut both ways. When the HHS Inspector General audited 200 severe-malnutrition claims, 173 were billed wrong: nearly $1B in overpayments from a single family of codes. So the predicted catastrophe never arrived - but the miss is only half the story.

None of that was in the forecast. The decade’s real product was a new layer of dispute, billions of dollars of it, grown on a rule everyone had filed away as a non-event. The standard was meant to help two sides agree; instead it gave the disagreement more places to live. The providers turned out right, just not for the reason they gave: the catastrophe they predicted never came, but a quieter cost did - the vendors, the documentation industry, the audit machinery, the AI that now codes against the AI that now denies, all of it sitting on top of the transaction. HHS adopted a code set; everything else grew up around it, beyond the forecast and beyond the rule. Over a decade later, that added layer is the ground every later case stands on.

It all lands at one station: coding, where the chart becomes a billable claim. That is where ICD-10 hit hardest, and where the machinery it spun off now sits:

Cards on the table

A shared vocabulary had not settled the fight, so the next mandate went at the price itself - and where ICD-10’s forecast had been all doom, this one was all promise. Post the prices, the thinking went, and the patient could finally see the cards the house had always kept hidden - and play a hand of her own.

The cards did get turned over: the Hospital Price Transparency rule made hospitals post their payer-specific negotiated rates in 2021, and the Transparency in Coverage rule made the insurers post theirs in 2022. The system guards one number above all the rest: what this payer actually pays this hospital for this procedure - the rate written into their contract, the number every bill is measured against. That number became law to disclose, twice over. Both rules chose the same vehicle - the machine-readable file, a giant structured data dump built for computers rather than people, on the theory that someone would build the people-friendly layer on top.

I watched clients comply with Transparency in Coverage (pronounced “tic” in meetings), and just publishing the rates the rule demanded became a months-long technical and operational lift inside the plans, the kind that pulls people off everything else.4 And what the mandate mostly bought was the files themselves - machine-readable, and aptly so (you would need a fairly expensive machine to read one). Single files ran toward a terabyte, formatted so that neither a patient nor most payers could open them, let alone compare what was inside. Hospitals had the same problem, only sharper: a community hospital has no transparency team to spin up, and the smallest independents post genuinely good data but can’t get the file format right. The files went up; the prices inside them never moved.

And the game didn’t stop. Commercial prices kept right on climbing - by 2022 RAND put the average commercial rate at 254% of Medicare, up from 224% two years before. Disclosure didn’t pull prices down; it handed the underpaid hospital a target. The one earning less than its crosstown rival could now see exactly how much less, and exactly what to ask for at the next negotiation. Where rates moved together at all, they moved up, because a posted rate is a target. The most favorable evidence makes the same point: a study of cancer prices found that posting the files moved nothing on its own, and prices dropped only where hospitals went further and posted rates a person could actually read, service by service - and even then slowly, in the most shoppable corner of medicine. The patient never became the player the rule imagined, either. The people working the files are employers, vendors, and researchers - not patients.

Price disclosure was never going to end the contest anyway, and weak compliance isn’t the reason - 7 in 10 hospitals now meet every requirement. The trouble is that posting hasn’t changed behavior, because neither side wants the rule to work. The hospital does not want its discounts shopped, the insurer does not want its rates undercut, and both already knew the numbers, because both negotiated them. Even full compliance just hands each side better ammunition for the next round of contracts. Seeing the other player’s hand makes for a fairer game, not a shorter one. The next mandate would come at it the other way - a hard deadline instead of disclosure - and still reach only half the system.

The permission slip

If you have ever waited days for an insurer to approve a scan your doctor already ordered, you have met prior authorization: the permission slip the payer demands before it will pay — the one check that comes ahead of the care. Prior auth may be the most hated process in American medicine - patients hate the wait, doctors hate the paperwork, and the plans defend it the way you defend a dam: not out of love, but for fear of what floods through without it. In 2024 the government finally moved - it set out, on purpose, to fix a piece of the cycle. The rule (CMS-0057-F, a number you are not meant to remember, but one that keeps a whole industry on edge) put the insurer on a clock: seven days to answer a routine request, seventy-two hours for an urgent one, electronic pipes by 2027, public reporting on how often it says yes, and $15 billion in projected savings over a decade.

But the clock is all it governs. The rule sets how fast the insurer must answer - not what the answer can be. A “faster no” is still a no, and the denial rate doesn’t move. It speeds the payer’s response without touching the patient’s outcome. Compare the pace: a card authorization clears in a second and commits the money; a prior auth can take a week and commits the insurer to nothing - the scan it approves today can still be denied when the claim arrives. Part of the week, to be fair, is real work. Behind many prior auths sits an actual clinical review — a nurse reading the chart, a medical director, sometimes a peer-to-peer call with the ordering doctor — and the plans do not have the clinicians to keep pace with specialty demand. That is the trap in the deadline: you cannot hire your way to seventy-two hours.

And it reaches only half the people it is for. Whether the rule covers your care depends on a line on your insurance card you have probably never read: if your coverage runs through a self-funded employer plan — about two-thirds of workers with employer coverage — then ERISA (a 1974 pension law with nothing to say about scans) puts you outside the rule entirely. On that side of the line the employer designs the benefit and a third-party administrator runs the denials under whatever the plan documents allow - a prior auth that can take as long as the plan likes, a denial that can arrive after the care, the terms written into a benefits booklet only the employer signs. The same scan gets the same denial, but with two different rulebooks - the dividing line set by a law written before the problem existed. A rule that governs half a market is one the market can afford to wait out.

So the deadline lands on a process that was already out of staff, and the way out is software: payers are buying it. Authorization requests that took a nurse a week come back from an algorithm in seconds - approvals instantly, denials instantly, the clinical review compressed into a model neither side trusts. This case is still unfolding as I write: the deadlines hit in 2027, the $15 billion is still a projection, statehouses keep passing their own versions.5 But it is the same miss again - the rule written to tame prior auth is automating it instead. The week-long wait becomes a same-day no, and the person you argue with about it stops being a person.

A room to fight in

Standards, disclosure, a clock - every rule so far had left the two sides to fight it out between themselves. This time the government went further. By the late 2010s the surprise out-of-network bill had become a national scandal — the anesthesiologist you met only on the bill, the ambulance that was somehow out of network — and the anger crossed every line that usually divides healthcare politics: patients, doctors, employers, both parties in Congress. When two sides cannot agree on a price, the oldest answer in the book is a neutral third party - it is how divorces settle and how labor disputes end - and healthcare’s fundamental disagreement looked like exactly the kind of fight you hand to a referee.

Subscribe now

So the No Surprises Act barred those doctors from billing patients and sent the leftover fight to arbitration. Each side names a price, an outside arbitrator picks one, no appeal. The patient steps out of the crossfire, and the payer and the provider argue it out in a back room at the end of the cycle - the provider wanting more, the payer wanting less. On its own terms it worked: patients stopped getting ambushed, and it should have kept working. What the law had built, though, was a private courtroom for the most contested transaction in American medicine - and word got around.

The government anticipated about 17,000 disputes a year; doctors brought 1,200,000 in the first half of 2025 alone, and won roughly 88% of them. A whole industry grew up to feed the machine - firms that exist only to file arbitration claims by the thousand, radiology groups with a hundred and fifty people doing nothing else. The arbitrators did well too: paid by the case, they collected $885 million in three years, and something predictable happens when the referee is paid per fight and prefers one of the fighters. As one health economist put it to the Times: “Arbitrators are people, and the typical person likes physicians.” The insurer walks in with baggage - owing, in every hearing, for every denial it has ever issued - and the doctor walks in as the one who saved a life. Not that every doctor in the queue saved a life: one Manhattan plastic surgeon (who owns the trademark Dr. Penis) has filed six thousand claims through the system and won better than 85% of them.6

You can see where the No Surprises Act hits: billing and AR follow-up, where the out-of-network claim gets fought out long after the patient has gone home:

The people who built it can see it. The congressman who wrote the law is glad it passed and says the arbitration needs reining in; an insurance chair calls it a recipe for higher costs with no checks, no balances, no oversight; the plans have sued by the dozen, and judges keep throwing the suits out, ruling that Congress never meant for courts to second-guess the arbitrator. So the awards stand, premiums rise to cover them, and the fight grinds on inside the room built to end it. Refereeing the fight is the most the US government has ever done about the American medical price - and it leaned on a naive hope that, handed a referee, the two sides would finally settle. The forecast completely whiffed (seriously, only 17k disputes annually?). Even by the standards of policy misses this one was extreme, and it broke one way — toward the providers, who took advantage the moment the doors opened.

Where the rules run out

Every case so far has been a rule that came up short: a standard that bred more to fight over, a disclosure that moved no price, a clock that only speeds the denial, a court that drowned in its own filings. Failure on that scale tempts an obvious conclusion — stop writing rules and let the market sort it out. naviHealth is what the market sorting it out looks like: one side stops waiting for a rule and settles the fight itself.

The soft part of the revenue cycle is recovery — the weeks in a rehab unit or a skilled-nursing bed after the hospital lets you go, where someone has to decide how many more days the plan will pay for. It is a real judgment, and for years no rule watched it closely. So UnitedHealth bought the judgment: in 2020 it acquired naviHealth, a post-acute-care company, and ran its nH Predict algorithm inside the largest Medicare Advantage plan in the country — software that looked at a patient and forecast, almost to the day, how long recovery should take.

The machinery inside a payer runs out of sight - not hidden exactly, just unwatched - and it stays that way until the public finds out. The finding-out, this time, was the work of Casey Ross and Bob Herman at STAT — two years of reporting, internal documents, former employees on the record (the closest thing this industry has had to its own Spotlight team). In March 2023 they published “Denied by AI“ — a Pulitzer finalist, and the reason any of this is on the record at all. The nH Predict forecast how long a patient should need; UnitedHealth then pressed its own nurses and case managers to keep their decisions within 1% of the forecast. A tool meant to advise had become a quota in everything but name, and staff who pushed back were overruled. The story opened on an eighty-five-year-old woman with a shattered shoulder whom the algorithm had given 16.6 days; on day 17, her plan stopped paying. I bet every insurer in the country was shitting their pants wondering if their own operations were doing the same.

Once it was on the record, the numbers came. The Senate Permanent Subcommittee on Investigations pried loose UnitedHealth’s internal data: post-acute denials up from 8.7% in 2019 to 22.7% in 2022, skilled-nursing denials from 1.4% to 12.6%, a ninefold jump in denial rate. The AHA put Medicare Advantage denials up 55.7% from 2022 to 2023. Inside that data, in 2022 alone, were some 34,000 skilled-nursing denials. One of them was a 91-year-old man named Gene Lokken.

In May 2022, Lokken fell and fractured his leg and ankle, then spent a month in a SNF just healing enough to be cleared for physical therapy. His therapists charted slow, real progress; UnitedHealth paid for 19 days of it, then declared him safe to go home. The algorithm said his recovery was finished. His family looked at a man who still could not walk and disagreed - and kept the therapy going with about $150,000 of their own money over the year that followed, until he died in July 2023. That November his estate sued, and the complaint put it in writing: when patients appealed these algorithmic cutoffs, roughly 90% were reversed — a model wrong nine times in ten, ending tens of thousands of recoveries, counting on how few families would fight. The Lokken estate’s suit has drawn a fraction of the exposé’s press, and it is the sharper instrument: a family asking a court, in plain terms, whether a payer may hand the end of someone’s care to a forecast.

And you can see where these denials cluster: the coverage reviews and denials deep in recovery, the stretch of the cycle no rule reached:

The patient caught in the middle of all this machinery is in week three of a recovery the software has already closed out. Strip the fight down and this is the contested transaction at its rawest. In the rest of the economy, two sides agree on a price before the money moves; here the payer decided, on its own, how much more recovery it would pay for — none — and had the software to make that decision stick across tens of thousands of patients at once. Providers armed back, and the arms race now runs both ways: providers running AI scribes to push their diagnoses up, payers running automation to knock the claims down, Cigna switching on software in October 2025 that automatically downgrades its priciest claims. Underneath the machinery it is the same yes and no the two sides have always traded - automated now, and sounding more official for it. Washington has noticed, and done little: Senator Warren and others pressed CMS for guidance; the agency gestured at the problem and took no power to audit or certify. No federal vehicle, as of this writing, governs the algorithms a payer uses to deny.

It took that lawsuit, because the courts were the last door standing. In February 2025 a Minnesota court threw out the healthcare claims — bad faith, deceptive practices — as preempted by the Medicare Act, and let stand only the contract claims (the breach, and the good faith implied in any contract) — the law that governs a broken fence or an unpaid invoice. The one body that could touch naviHealth could reach it only as a contract dispute - and a suit brought by a dead man’s family is doing the work no regulator did. The case is still moving through the court as of this writing.

So the market sorted it out. One side got a model, the other side got a lawsuit, and the patient in the middle had a say in neither - his family just got the bill. The rules in this essay are inadequate; the stretch with no rules was worse. Imperfect referees still beat an empty field.

The floor

The word has been confessing it all along. Claim comes from the Latin clamare, “to cry out” - the root it shares with clamor, exclaim, proclaim, counterclaim. English took it up in the thirteenth century as a “cry for what one is owed”, and the meaning has held ever since: you claim what someone might withhold. A bill expects payment; a claim expects an argument - the word assumes an audience that can say no. The industry’s own name for its invoice concedes the fight, and every rule so far took the concession at face value: a standard for wording the cry, sunlight on its going rates, a clock on the answer, a courtroom for when the answer is no.

Look back across the regulatory map and the patchwork is plain. The five cases here were drawn from dozens - I lost count past forty, most of them code sets and transaction formats no reader would tell apart (the 276 from the 277). And every one of those rules lands somewhere around the claim - never on the one thing at its core: what is actually owed.

(It’s interactive too - explore the field on Datawrapper.)

Rule after rule, station by station, written for parties that could hardly be less alike -health systems with compliance departments the size of agencies, small hospitals drowning in the same requirements, a thousand-plus payers of every capability, and the patient billed for the friction among them. Which raises the obvious question: why isn’t there just one law that covers the whole revenue cycle?

There was an attempt - and you already know its name. HIPAA, the Health Insurance Portability and Accountability Act, is the most famous statute in American healthcare, mostly for things it did not originally contain (the privacy rules everyone associates with it came years later, by regulation). What Congress actually passed in 1996, by lopsided bipartisan margins, was a bill about insurance: letting workers carry coverage between jobs, limiting preexisting-condition exclusions. The part that matters to this essay rode along almost unnoticed — a title called administrative simplification, added on the theory that if the whole industry billed in one electronic language, the paperwork would stop eating medicine. A floor under every station, rather than one more rule dropped on top - laid in a world of paper claims and fax machines, by people who could not have pictured an algorithm denying a nursing-home day.

Read the name again, because it gives the game away. Insurance, Portability, Accountability... the moving of coverage, the accounting of the transaction, and nowhere in it health, or care, or a patient. A road is built for the traveler; this one was built for the transaction.

HIPAA’s administrative-simplification provisions were meant to do for the cycle what road signs do for a highway: give everyone one language. They standardized the electronic form a claim rides on and the form the payment comes back on, locked the diagnosis and procedure code sets in place, and handed every provider one number the whole system would recognize. It was federal, it was ambitious, and it worked - the rails it laid are the rails every claim in America still runs on. It is the foundation every later rule was built on.

And the fragmentation grew on it. A standard can only settle what the two parties don’t fight about. HIPAA could standardize the form of the claim, since the form was never the fight; it could not standardize the price, because the price is the contest. So it standardized the form of the claim but never the price inside it - and a shared form only let the fighting scale. There are, depending on how you count, somewhere between several hundred and a few thousand health insurers in this country — national carriers, state Blues, Medicaid plans, the administrators running self-funded employers. And each one published a companion guide: the local quirks layered on top of the standard, the extra fields, the house rules, the one form refracted into hundreds of slightly different ones. Vendors productized the exceptions; the clearinghouses built whole businesses on the gap between the format two payers were supposed to share and the ones they actually used. In finance a clearinghouse exists to make settlement certain, netting the trades and guaranteeing the money; healthcare borrowed the name for a business that clears nothing - it translates the dispute and passes it along, because there is nothing underneath to guarantee. But the cracks — the price, the terms, everything the two sides actually fight over — were never built into it, and the fragmentation grew up through them.

The mistake was the same one price transparency would repeat a quarter-century later: it assumed the two sides were fighting because they could not understand each other. Get everyone speaking one language, the thinking went, and the friction falls away. But they were never fighting over language - two parties fighting over money do not stop fighting when you hand them a shared form; they fight more efficiently. A signage standard tells two drivers how to read the same intersection - it does not decide which of them yields.

Subscribe now

Call it what it is

Picking apart the side effects of healthcare policy is a trope at this point, and yet we never examine the defining feature of American healthcare. Every rule in this essay circled the same thing. ICD-10 standardized the language around it, price transparency published the rates beside it, prior authorization put a clock in front of it, the No Surprises Act built a courtroom to fight over it. Each pressed closer than the last, never onto it, and the consequences grew with the proximity — more codes, more filings, more machinery for the dispute. What they skirted was the contention at the bottom: what the care was worth, and so what is owed for it — the disagreement a capitalist market cannot seem to nail down.

And price is more than just a number - it’s a deal closed before the work begins, and when two parties cannot close one, the deal dies there and both walk away. Healthcare’s parties can do neither. The hospital needs the insurer’s members, the insurer needs the hospital’s beds, and the network terminations that make the local news are theatrics before the re-signing. So the contracts get signed and the rates get settled, and the deal still never closes - a rate prices the code, and every claim re-opens whether this care, this code, this coverage earned it. Thousands of these pairings across the country keep re-arguing it, a few hundred dollars at a time, with the question underneath: what was that care worth, and what is owed for it? The provider cannot say what an hour of its work is worth; the payer will only name what it is willing to pay; decades on, the cost of healthcare continues to grow.

The dependence and the resentment are the same bond. So they fight the way a married couple fights - in public, in front of the kids, the same grievance in new words every round - and the rules in this essay are marriage advice: one shared language, open books, a timer on the answers, even a courtroom for the worst of it. They take all of it and stay married.

Only agreement would end it, and agreement cannot be set from above. Fix the price (flat fee, value-based rate, whatever the decade calls it) and the argument just moves to volume. This essay has no clean answer either; offering one would just add a forecast to the pile. The cases point to something narrower and harder: every rule assumed the basic question was already answered - that what is owed had been settled, and the rule only had to organize around it. It never was. There is no settled answer underneath, only the argument, and it has swallowed every rule written to end it.

Strip the fight down and it is two sides trying to narrow what care is worth and what anyone will pay - a negotiation that never closes. Running it — the coders, the denial software, the arbitration firms, the consultants (I was one) — gets counted as the cost of American healthcare, though the work is arguing, not care. The arguing alone (the billing, the back-and-forth with the insurer) was last counted at $471 billion a year, about a sixth of what the country then spent; administration taken whole runs a quarter to a third, rising with every new rule, because each one hands the fight another surface to happen on. The industry files it under overhead, as if it were friction in a complicated machine; it is the running cost of that open negotiation - and the least we could do is stop sending in the next regulator with the next rule and call it what it is: a permanent quarrel we built our largest industry to run on.

Reconsider the claim: a chronicle and a contention. The chronicle needed nothing from us. The contention — the assertion that a sum is owed — is what the country has spent thirty years governing: codified, compelled, circumscribed, convened, counseled - everything but concluded.

1. I was a young consultant at Deloitte then, doing what was essentially customer success work, trying to upsell client health systems on a pending ICD-10 deadline. We had access to client data and tailored slide decks that would show each system the revenue it stood to lose from coding mishaps - real numbers, their numbers, arranged to alarm. The problem was that the deadline kept moving (go-live slipped two years all told), and you could watch the alarm depreciate in real time: the first delay bought us urgent meetings, the second bought polite ones, and by the end half the client health systems had stopped caring about the deadline at all. Then the cutover came and went, the catastrophe didn’t, and the lesson the industry took wasn’t we prepared well - it was they cried wolf. That, I’d learn, is the standard arc of a compliance deadline in healthcare: announced as an asteroid, postponed into a nuisance, survived as a shrug, and remembered as one more reason to ignore the next warning.↩︎

2. Years later I was on the other side of the table, on an M&A project at a regional Blues plan that suddenly had two top priorities for the year: land the acquisition, and comply with TiC. The plan had a real data department - engineers, data scientists, people hired to build things a member might actually notice - and the roadmap bent around the mandate anyway. Integration work slowed. New product work slowed. The better ideas waited in line behind a compliance artifact, and the team spent its year building machinery whose only purpose was to satisfy the rule. The files went out on time, which got counted as the win; the real cost never appeared on an invoice - it was the year of better work that didn’t happen.↩︎

3. Texas tried to widen the door with its HB 3459 “gold card” law, which lets a physician skip prior auth once she has built a track record of approvals - reward the doctors who reliably order the right care and stop making them ask permission. A sound idea, except that only 3% of Texas physicians ever qualified. The exemption sits out of reach because, for the payer, the chokepoint is the point - friction deters orders, and a scan never requested is a scan never paid for.↩︎

4. ↩︎

Editor’s note: This is the 3rd of 5 essays in The Cycle, my series on healthcare’s revenue cycle. Ongoing writing to continue here.

Every few years, investors find a new market to draw lines through. Take something bundled — a platform, an industry vertical, a system that does too many things under one roof — and split it into specialized companies that each do one thing better. The term, “unbundling”, was coined in a 2010 post unbundling Craigslist (on a Tumblr, no less) - one ugly homepage full of loosely related categories, each a specialized startup waiting to be peeled off. Airbnb took housing, Indeed took jobs, dating apps took personals - and investors followed with billions in capital. The thesis got applied to Excel, LinkedIn, ChatGPT, every bundled incumbent investors could find. A useful mental model, or at least one that keeps the fundraising moving. Now it’s pointed at healthcare’s billing infrastructure.

The administrative function that is the bizarrely-titled Revenue Cycle comprises roughly a baker’s dozen of steps from patient scheduling through final payment. Rev Cycle is a $300 billion industry ripe for unbundling - it is the admin layer between providers and payments worth roughly the GDP of Portugal. For investors, it looks like the perfect target: fragmented, software-underserved, exploding with AI potential. For providers, it looks like the opposite - another set of vendors competing for space on an already-crowded application stack, each one promising to own a piece of something that doesn’t work in pieces.

Market penetration varies across the arc. Cedar handles patient billing and collections. Waystar routes claims between hospitals and insurers. Solventum dominates medical coding — the translation of clinical documentation into billable charges. Cohere Health is building intelligent prior authorization — getting insurer sign-off before care is delivered. Each a specialist, each owning one piece of the arc. Exactly the shape the unbundling thesis looks for.

The biggest names in venture are bought in:

• a16z - coined the healthcare unbundling frame. Their 2021 Akasa thesis: “For every $1 of revenue collected by a hospital, $0.25 is spent on the administrative tasks required to collect it.” Cedar at a $3.2B valuation, Akasa, and Commure all sit in the RCM portfolio.

• Bessemer - State of Health AI 2026: 92% of health systems now deploying AI somewhere in revenue cycle.

• Flare Capital (healthcare-specialist) - “Billions in Play” thesis: AI-enabled category leaders command a 30% valuation premium, station winners 5x. Portfolio: SmarterDx (crossed $100M ARR in 3 years before the New Mountain rollup), Cohere Health, Suki, Layer Health, Axuall.

• Second Opinion / Scrub Capital - Christina Farr’s February 2026 “RCM Market Map” calls RCM “the first smash hit application for AI in healthcare.” Built with Bertelsmann Healthcare Investments ($200M U.S. health-tech fund with RCM as a named focus). Scrub Capital fund backs Elion - a procurement marketplace for hospitals shopping RCM vendors against each other. A meta-bet on the fragmentation itself.

If you’re a founder drawing a line on a market map in a Sand Hill Road conference room, you may walk out the door with a nice seed stage investment check.

But the same money funding the unbundling thesis is writing billion-dollar checks to bolt stations back together. The pitched exit is built, but the realized exit is bought. The market is mispricing one of them.

The Rollup Trade

The investor playbook for a vertical like revenue cycle is well-worn. Find a station where incumbents underperform, build the best point solution, win the wedge, and expand laterally. Unbundling is the playbook. What the pitch decks don’t say is that “expand laterally” is a euphemism for getting rolled up - acquired into a bigger platform spanning more of the arc. And rollups come with costs the market map doesn’t price: integration work, data model translation, change management across every hospital IT operation that adopts them. The playbook works for the founders and their VCs. Their exit isn’t an IPO - it’s the assembly itself, getting absorbed into a bigger platform. Getting bought is the realized thesis, even when it’s not the pitched one. The deck sells “build the category leader“ in something like denials management or clindoc, but the exit is being rolled up before the bilateral cascade catches the wedge. Both can be true at the founder level - but the durability case being sold to investors is doing different work than the liquidity case being executed.

Waystar started as a clearinghouse - routing claims between hospitals and insurers, back-end plumbing. Since the Navicure–ZirMed merger formed the company under Bain Capital in 2017, they’ve made ten acquisitions across the arc:The Iodine deal in 2025 was the first time Waystar stepped upstream into mid-cycle clinical documentation - where the bill actually originates. The seven undisclosed deals between 2018 and 2023 were tuck-ins, almost certainly small. The three priced ones jumped from $450M (Patientco) to $1.3B (eSolutions) to $1.25B (Iodine). The cheap adjacent layers are gone - crossing into a new part of the arc now costs a billion.

Waystar isn’t a back-end vendor climbing a value chain. Three billion dollars of acquisitions across the arc is a company building to be the partner CFOs keep when budgets tighten - the moment specialists get cut in favor of vendors covering more of it.

Cedar ran the same play from the other end of the arc, moving upstream from patient billing into payer data, pre-visit workflows, Medicaid enrollment. Solventum, holding 75% of encoder installations in US hospitals, went the opposite direction - connecting coding downstream into denial prevention. Different starting stations, same move: when the problem you’re solving depends on the adjacent stations, you keep buying until you own them.

Having the best-of-breed point solution doesn’t really matter. The clearest evidence the market doesn’t reward quality is in KLAS ratings (which basically scores how much hospitals like their vendors) of end-to-end RCM outsourcers. Ensemble Health Partners at 95.1, best-in-class, but R1 RCM is at a meager 55.6. There’s a 40-point spread in ratings, but it doesn’t matter for market positioning. Both end-to-end solutions follow the same playbook, same access to durable contracts. R1’s customers just don’t rate them as highly. R1 is still public, still profitable, still acquiring because coverage of the revenue cycle wins the contracts whether staff like the product or not.

The same investors funding station-by-station specialists are writing billion-dollar checks to bundle stations back together. New Mountain Capital just rolled three companies into Smarter Technologies - $800 million in combined revenue, three stations bolted into one platform. Waystar spent $1.25 billion on Iodine for the same move. So, we can see unbundling doesn’t work as well here, but why?

Subscribe now

The EHR Already Ate Half the Map

Let’s start with the landscape where these vendors compete. Investors draw 14 stations on the map, but the real game is narrower. Epic (and other EHRs) have already absorbed the front half of the arc; what remains is defined by what Epic can’t reach - the space between hospital and insurer, where every RCM startup plays and where most of them struggle.

When a hospital goes live on Epic, the front half of the arc comes built in - scheduling, registration, charge capture, the functions closest to the patient chart. The back half doesn’t - coding, clearinghouse, denial management, collections. An HFMA survey of 160 healthcare organizations shows the split as a gradient:

Functions closest to the EHR got absorbed first. Functions that grew up later, in the space between hospitals and insurers, stayed independent - because Epic can manage a patient’s chart, but it can’t manage the negotiation between the hospital and the insurance company over what that chart is worth. Everything remaining on the revenue cycle arc sits in that negotiation.

Third-party vendors live at the mercy of the healthcare economy. When hospitals bleed money, CFOs cut contracts - and that usually means consolidating onto Epic. They look at a vendor inventory with forty-something contracts, a dozen overlapping with something Epic already does, and a CIO pointing out that integration costs are eating the savings. I’ve run a dozen application rationalization projects (CIOs literally only want one thing) - if Epic has a tab for it, the standalone vendor contract will have a hard time getting renewed. This is not because Epic’s version is better, but rather because running two systems that do the same thing is expensive.

Share

Here’s where the money concentrates:

• Coding - $27B market, 51% third-party. Translating clinical documentation into billable codes requires payer-specific reimbursement rules the EHR doesn’t carry.

• Clearinghouse - $14–17B, ~80% third-party. Someone has to route claims between systems that weren’t designed to talk to each other.

• Denial management - $5–9B, 58% third-party. Payers and providers fight over every claim; the EHR can’t see the fight.

Every one of these markets exists because Epic can only see one side of the relationship.

Absorbed stations don’t disappear - they push the market upward. Epic absorbed scheduling (72% native), but the intelligence layer above scheduling didn’t get absorbed with it. Phreesia processes one in six US patient visits and grosses $500 million a year, growing 14-16% annually by owning the wrap-around layer Epic doesn’t bother to build - patient intake, eligibility, payment workflows. Epic took native scheduling. The money migrated to the layer above it.

Epic keeps moving the boundary. In November 2026, Epic Penny launches autonomous coding - an AI agent that turns clinical documentation into billable codes without a human coder in the loop. The same capability standalone AI-coding vendors built businesses around is now an Epic feature, bundled into the EHR contract. If Epic Penny works as advertised, the $27 billion coding market — the largest remaining independent station on the arc — starts becoming absorbed into the EHR. The attackable surface shrinks with every release.

What’s left to unbundle is what Epic can’t see: the stations that live in the space between hospitals and insurers.

Why the Unbundling Analogy Fails

In that remaining space, nothing stands alone. Revenue cycle stations are causally linked in ways Craigslist categories never were. You don’t need to buy a couch to find a date, and the dating section doesn’t affect people shopping for couches. But optimize coding and the patient’s bill goes up - higher-acuity codes mean bigger claims mean bigger coinsurance, a $1,600 patient responsibility becoming $2,200 on the same procedure. Optimize collections and patients defer future care. Every optimization at one station ripples through the others.

Even where categories actually are separable, unbundling underdelivered. “The Unbundling Fallacy” showed that many of Parker’s original Craigslist unbundlers failed because the bundled platform had network effects the verticals couldn’t replicate. “Excel Never Dies” made the case that Excel wasn’t unbundled either - the spreadsheet only got more dominant as SaaS tools proliferated around it. “The Rebundling of Craigslist” documented how vertical unbundlers hit structural limits. The same map got drawn for electronic medical records, and Epic grew market share while the unbundlers stalled.

If unbundling stumbles where the categories are loose, it cannot work where they’re chained together.

Subscribe now

The Independent Market Is the Interdependent Market

It’s not just that one station affects the next. In this remaining space, the stations feed each other in both directions:

• Coding → sets up billing

• Billing → sets up denials

• Denials → trigger appeals

• Appeals → reach back into the original coding documentation

• And prior auth → gates the coding before it even happens

Every station’s output is another station’s input - and the loops run both ways.

These are actual experiences that revenue cycle teams face every month. Walk into a hospital revenue cycle office and watch one of those loops fire: the denial management team — the people fighting rejected claims — traces a rejection back to a coding decision, which traces back to a documentation gap, which traces back to a physician who didn’t know the payer changed its clinical criteria last quarter. The problem crosses four stations and three departments. The coding vendor has no idea the denial rate just spiked because of their optimization.

An investor betting on a coding wedge is betting on a station whose performance depends on what happens at denial management, three stations downstream, at a company they’ve never talked to. There is no wedge that gives you the next wedge. Each station is dependent on stations owned by someone else, which means the wedge-and-expand playbook can’t produce durability through organic growth. The only path to multi-station coverage runs through acquisition, which is why the durable companies in this market are the ones writing checks, not the ones building features.

The Curious Case of Olive AI

Olive AI is the case in point. Between 2018 and 2022, the company raised $900 million on a single-station wedge - automate prior authorization, win the wedge, expand laterally. Peak valuation hit $4 billion in 2021. By late 2023, Olive was wound down and sold for parts. The wedge didn’t hold.

I was *pinches fingers* this close to getting an Exec Director role at Olive in 2021. I made it to the final round, but didn’t make it by not having enough C-level client experience (probably fair) - which, looking back, told me something about what Olive valued. They were optimizing for executive relationships, not operational understanding of the systems they were automating. The public narrative about their collapse is about overpromised AI - brittle RPA, 27 pivots, five-fold headcount growth in eighteen months (all true). But the structural problem ran deeper. Olive tried to automate prior authorization as a standalone function. Prior authorization is a bilateral negotiation:

Hospital request → Insurer evaluation → (if denied) → Hospital appeal → Insurer re-evaluation → (loop until resolved or abandoned)

Automating one side of that exchange doesn’t work when the other side adapts. Olive built a faster fax machine for one side of the conversation.

After the collapse, Jeremy Friese — who’d run Olive’s payer market as president — bought the provider-side prior auth assets and founded Humata Health. The pitch he gave was an admission of error. Where Olive automated only the provider side of prior auth, Humata, in Friese’s words, would build “both sides of the fax machine” - the payer and the provider. He never says Olive got the architecture wrong. The product rebuild says it for him: you can’t win a station in this market from one side of the negotiation. Even the founders who watched it fail had to come back to sit on both sides of the fight.

There’s one type of station that escapes this trap: coding. Nym, Fathom, and CodaMetrix turn clinical documentation into billing codes - one input, one output, one buyer per hospital, clean boundaries. They’ve built standalone businesses because the transaction is one-directional. Every other remaining station is bilateral. The vendor sits between hospital and insurer, mediating a fight neither side can resolve. That’s why the bilateral stations get rolled up: no one can run them alone.

The Friction Market

Revenue cycle is the billing apparatus of a permanent disagreement. Hospitals and insurers don’t agree on what care is worth, and revenue cycle is the system that processes the disagreement.

The money that analysts size as “the revenue cycle market” is the cost of processing claims through a permanent, adversarial negotiation between hospitals and insurers. Every station exists because the two sides can’t agree:

• Coding translates the hospital’s documentation into the insurer’s reimbursement language.

• Denial management processes the insurer’s no’s against the hospital’s yes’s.

• Prior auth is where the insurer gets to veto care before the hospital delivers it - the only stage in the cycle where the negotiation runs before the service exists. The earlier the fight, the more leverage the insurer has.

• Clearinghouse routes claims between hospital systems and payer systems that were never designed to talk to each other. It translates data formats and routing rules every time a claim moves between the two parties - the plumbing that exists because the two sides can’t agree on shared data standards either.

• Collections picks up whatever balance the hospital-insurer negotiation didn’t resolve. When the two sides stalemate, the unpaid portion gets shifted to the patient - the only party in the chain without leverage to fight back.

Each is a tollbooth on the disagreement. The bigger the disagreement, the bigger the tollbooth.

Every vendor in this market is a middleman in that fight. And every optimization — faster coding, smarter denials, automated prior auth — entrenches the fight rather than resolving it. The vendors build rails so dependent on the current terms of the hospital-insurer relationship that they harden the very dynamics they claim to streamline. Cedar’s 2026 Trends in Patient Payments report says it plainly: patients are “the one payer class where providers can still influence outcomes.” When the bilateral fight reaches stalemate, the costs get pushed to the one party that can’t fight back.

The arms race confirms it:

• Denial rates climbed from 10.15% to 11.8% between 2020 and 2024

• Medicare Advantage prior-auth appeal overturn rates run above 80% - most appealed denials in MA are wrong on the merits

• Volume keeps rising anyway, because the fight is the point, not the accuracy

Both sides spend billions on automation to maintain the same relative position. The vendors are the only clear winners. They bill for every appealed claim, every re-submitted authorization, every dispute that never resolves. Friction itself is the product.

Revenue cycle is too interconnected to unbundle and too sprawling to consolidate. The market exists because of the relationship, not despite it.

Every company operating in this space is optimizing within the current terms of that relationship - and those terms are themselves being rewritten:

• Prior auth is getting a clock on it. The CMS-0057-F mandate takes effect January 2027 - payers have to use FHIR APIs and respond within 72 hours (7 days for urgent). The front-end of the fight just got compressed from a month to days. Every vendor selling prior-auth automation gets a moving target.

• The collections lever just disappeared. The Consumer Financial Protection Bureau’s January 2025 rule bars medical debt from credit reports - providers lose the threat that made patients pay. Cost pressure routes back upstream to the hospital, which routes it back into the rest of the cycle.

• States are picking surprise billing apart one industry at a time. Beyond the federal No Surprises Act, state legislatures are adding rules on ground ambulances, behavioral health, air ambulance pricing, and out-of-network disputes. Every statute moves the line on who eats the loss when the fight ends in stalemate.

This is the ground under every vendor’s feet, moving faster than any of their product roadmaps.

After COVID, I briefed Chuck Christian, CTO at Franciscan Health, on an IT application rationalization (basically, which IT vendor contracts to keep, which to cut). With nearly 40 years of IT experience, he told me the ebbs and flows of the market: best-of-breed point solutions wins for a stretch, then something breaks in the economy and buyers pull everything back under one roof for rationalization. Then the market loosens with new innovations, and the point solutions get back in. But ultimately, the issue wasn’t which tech was best: it was who’s a partner and who’s a vendor - which of these companies will still be here in ten years, when the tech is woven through operations and I can’t rip it out?

This is how institutions think about durability of vendors. A startup’s Series B runway rarely overlaps with a hospital’s implementation timeline. The hospital has twenty specialists to pick from, and most of them won’t survive the next wave. Which means the premiums Flare Capital cited at the top — 30% extra for AI-enabled vendors, 5x for the station winner — aren’t being paid because investors think the company will last. They’re being paid because investors think someone else will buy it. That’s why IPOs happen a lot less in healthtech - the usual outcome is rollup.

The payer-provider fight is structurally dysfunctional, and that dysfunction rewards positioning over performance. CFOs trimming IT budgets pick the partner who covers enough of the arc to be worth keeping, regardless of how staff feels about the product. The worse the underlying market gets, the more valuable broad coverage becomes. Point solutions in this market end up as components in someone else’s stack. The fight stays broken; the broad-coverage vendor stays paid. The moat compounds across cycles, not quarters.

And the fight is what scales. Fix one station and the next one breaks. AI just makes both sides faster - sharper denials meet sharper appeals, prior-auth bots meet payer counter-bots. The market grows because the fight grows.

Which is the thing that should change how this asset class is priced. The $300 billion isn’t the size of the RCM software market. It’s the price the system pays to keep the disagreement going. Investors are pricing it like software. It’s friction - and the durable companies are the ones building for the fight, not its end.

Subscribe now

This piece extends a thread running through Not a Cycle and Cycles and Triangles. Coming next: how policy is rewriting the terms underneath, and what survives when the bundle finally breaks. - A.T.

Nothing here is investment advice. The companies named are illustrative of a market dynamic, not recommendations. I hold no positions in any of them and have no current consulting engagement with any company named.

The companies named throughout this piece are representative, not exhaustive - Stedi’s January 2026 map catalogs 500+ vendors across the arc, Second Opinion / BHI draws another version, and no two agree on where the lines run. Which is its own evidence for the argument above.

Editor's note: This is the 2nd of 5 essays in The Cycle, my series on healthcare's revenue cycle. Ongoing writing to continue here.

I couldn’t seem to attribute the source, but auto repair shops have had this joke since the 70s:

With five simple words, the punchline illustrates the idea of real-life constraints. Yet I’ve always wondered: who wouldn’t pick Good?

It’s a popular idea that spans trades and industries. Engineers built a discipline around this concept and called it the project management triangle, showing the same trade-offs. And in the 1994 book Medicine’s Dilemmas, Dr. William Kissick — a drafter of the original Medicare legislation — named the healthcare version of this: the Iron Triangle, a trilemma of Access, Quality, and Cost. You can have a system that does well at two of them, not all three.

The cruelty of the choice is that Quality is the corner you can’t really trade off. A lower-quality replacement part for my car wears down a few years sooner - which is annoying, but acceptable. A lower-quality stent in a heart valve is not a trade I’d take. Settling for lower-quality healthcare feels foolish, which means the triangle has to bend on the other two - access and cost. Someone gives up the appointment, or someone eats the bill. The triangle isn’t really about three properties; it’s about which trade-off is most acceptable.

Subscribe now

Revenue Cycle as a Triangle

The triangle reads as a policy abstraction. But it gets decided every day, in millions of small decisions, by people who do not think of themselves as policymakers - registration clerks, eligibility specialists, coders, claim scrubbers, denial managers, billers, collectors. The apparatus they work inside is the Revenue Cycle - the financial process in which a patient’s encounter is scheduled, documented, and finally paid. There are three distinct phases: front-end before the encounter, middle during it, back-end in the months after.1

Hover over one of the steps and watch the ripple. The shape isn’t a circle, despite the name - a patient enters from the left, but money walks out on the right. Each phase in the revenue cycle directly affects access, quality, and cost - the central trade-offs in American healthcare.

I’ve seen a lot of healthcare claims, and every claim is a small case study in this trilemma. Most people see boring forms and numbers; I see a story. With nothing but a stack of claims and remittances, I can trace the ups and downs of a patient’s life — what they tried to do, what got in the way, who paid, who didn’t, and who got hurt by the gap in between.

Let me walk you through one. A 52-year-old finally scheduled the colonoscopy her PCP had been nagging her about. Under the Affordable Care Act, screening colonoscopies are preventative care - the patient pays nothing if everything goes according to plan. She thought hers would be free, and on paper she was right. As we trace the case forward, you’ll see how she ended up wrong.

Front-End = Access

From the first step of booking her appointment, our patient begins her journey on the Front-End of the revenue cycle - sometimes called “patient access”. On paper, she does everything right: appointment booked, eligibility checked, no prior auth required at this age, appointment confirmation call, prep packet, clear-liquids sheet, all green lights. But the front-end exists because almost nothing in healthcare goes according to plan.

Two things have already gone wrong that will cascade into months of headaches. Registration entered her insurance information from a photographed card and got one field wrong - a subscriber number off by a digit, enough to break eligibility downstream. And the procedure was booked under the assumption that nothing about the encounter would change during the procedure. Mistakes like these are how routine front-end work becomes the back-end’s problem six weeks later.

This is the Access phase, and its decisions echo through the entire cycle. The wrong field in registration becomes the denial in week six. A prior auth obtained for the wrong code becomes the appeal that ties up two FTEs for a quarter. Front-end work decides who gets care - and decides, in advance, most of what happens to the claim downstream, which has real patient consequences. According to an AMA survey, 34% of physicians watch patients abandon treatment after a prior-auth fight. Each step in the onerous process deters patients from accessing care.

Every step before the procedure is a small bet against something going wrong later - and when those bets lose, the cost doesn’t disappear. It moves to whoever is standing closest to the problem.

Middle = Quality

The procedure goes as scheduled. Under twilight sedation, the gastroenterologist finds two polyps - snares them, drops them in a specimen jar, and sends them to pathology. To the patient on the table, the procedure still feels like the screening she came in for. But on the operative note the doctor writes up afterward — the clinical record of what actually happened in the room — the encounter has already changed. From the dashboard view at the end of the week, this case will look identical to the forty other colonoscopies that ran the same day, until a coder opens that note three days later.

The instant those polyps are biopsied, the encounter changes from screening to diagnostic - different CPT code, different billing rule, different patient liability. Three days later, a coder sitting at a screen forty miles from the procedure room opens the operative note. She has to decide whether to append modifier -332 — a two-character tag added to the billing code, marking the procedure as preventative so the payer keeps treating it as a screening for cost-sharing purposes. A month’s rent is sitting on the knife’s edge of that two-character decision. If the coder appends the modifier, the patient still owes nothing. If she misses it, the patient owes $1,840 — a bill set in motion by someone she will never meet, based on a billing rule she has no idea exists.

This is the Quality phase, and the word is doing strange work. Quality in revenue cycle is not the same thing as clinical quality, but they’re directly related: it’s not whether the patient got better, it’s how accurately the encounter is rendered on paper. The middle is where the clinical reality the doctor experienced gets translated into the billable artifact the payer will see. What was this encounter? The coder’s answer flows downstream into the payer’s records, the appeal queue, and the patient’s bill — and you only get to answer it once.

What the middle decides is what the back-end inherits. The coder’s modifier choice will determine whether the appeal three months from now is a fight or a formality. The coders, the documentation specialists, the charge auditors — they’re not setting healthcare policy. They are deciding patient experience, one encounter at a time.

The claim goes out. A clean claim, in revenue-cycle speak, is one the payer accepts on first read — no missing fields, no codes in conflict, no eligibility mismatches — and about 84% of claims at a typical hospital clear that bar. This one will not. Two assumptions from week one are about to come due in week six, and the burden of resolving them is about to land on someone else’s desk.

Back-End = Cost

The Denial

Two weeks later, the denial comes back. The patient will not hear about it for another six weeks - she will be eating Thanksgiving dinner with the bill she does not know about sitting in a folder on her counter. The hospital hears it immediately. In the 835 file — the clunky, electronic remittance file, which uses the EDI (Electronic Data Interchange) format designed in an era optimized for kilobytes3 — the denial is a single row of text:

CARC 31. Patient cannot be identified as our insured.

CARC stands for Claim Adjustment Reason Code - there are over 200 of them in active use. Code 31 is the insurance company’s reading that they have no record of this person under this plan, and that single row is the entire communication: no human voice, no explanation, no record of why. Behind it is the subscriber number off by a digit from registration day — the insurer can’t find her — so the claim is rejected before the modifier-33 question even matters. One row of data, a month of human labor to reverse it, and $1,840 the patient owes until that work is done.

The denial management team picks up the case on the other end. Tier 1 work — the templated denials with known patterns and known fixes — is handled by a vendor in Manila, twelve time zones from the hospital’s billing office, working through queues that already have a few thousand items in them when the shift starts.

Outsourced revenue cycle is mostly arbitrage: cheaper labor doing predictable rework, which works well for templated denials and breaks at the edges where context matters — the unusual modifier, the half-documented op note, the eligibility detail that doesn’t match anything in the playbook. The mismatch from registration day, fortunately for everyone, is exactly the kind of thing they look for. They request medical records to support the modifier-33 documentation, attach them to the case, and file an appeal. The appeal cycle is slow: twenty to forty days, depending on the payer. The cost to file — staff time, documentation pulls, follow-up calls — runs $45 per claim. Across the industry, roughly half of appealed denials get overturned on the first try; with persistence, around two-thirds eventually get paid.

Meanwhile, the patient is about to find out about the bill. The cost of resolving it is already mounting - just on someone else’s ledger.

Job to be Done

A letter arrives in the mail.

Your insurance has determined that this service is not covered under your preventative benefit and you are responsible for $1,840.

It’s wrong (not just morally), but she doesn’t know it’s wrong. She doesn’t know what an EOB is — that it’s a notification, not a bill — and the document was not designed to clarify the difference. So she calls the insurance company. They tell her to call the hospital. She calls the hospital. Patient billing tells her the claim is in appeal and please don’t pay anything yet. Then patient billing, running on its own schedule, mails her a statement because the system flagged a balance. Three departments at two organizations are now arguing on her behalf about $1,840 — hours of paid labor at the hospital, hours of unpaid labor by her, all of it set in motion by a mis-typed subscriber number from week one.

Unfortunately, neither the insurer nor the hospital will be able to help her through this easily. The work of navigating this system is a job-to-be-done, but nobody is incentivized to pay for this to happen.

The Cost of the Cost

The industry has a name for what hospitals spend just to get paid: cost to collect. In 2024, US hospitals spent roughly $43 billion on it, per the American Hospital Association — money for staff, software, vendors, and appeals that never touches a patient.

For a mid-sized academic medical center, that’s hundreds of FTEs, tens of millions of dollars a year, and a building full of people whose entire job is chasing claims that should not have been denied in the first place. Every quarter, the hospital submits tens of thousands of claims, watches an industry-average 11% bounce back as denials, and runs a back office sized to fight the rest back into payment. None of those numbers describe a hospital in distress. They describe an average hospital, on a Tuesday, handling its share of the roughly 500 million claim denials American insurers issue annually. The cycle does not look like a crisis from the inside. It looks like a job.

This is the Cost corner of the triangle, and it isn’t the cost of the procedure — it’s the cost of deciding who pays for it. Almost all of it is invisible to the patient who started the cascade by mis-typing a subscriber number. The rest gets folded back into premiums and bills, paid in small increments by everyone else who carries the system’s costs.

Resolved

As we reach the resolution of this claim, nearly twelve weeks after the procedure, the appeal lands. The hospital eats most of the difference; she owes the coinsurance her plan attaches to a diagnostic procedure. She pays $200 — less than the $1,840 the letter threatened, more than the $0 the ACA promised. The final statement reads balance: $200. And while she doesn’t know what changed or why, she is not scheduling another preventative appointment any time soon. Something about the word free in the preventative care that she has sworn off.

Front-end, middle, back-end. Access, Quality, Cost. And our patient got access and quality, but has paid the cost: first in dollars, then in trust.

Another Iron Triangle

Now turn the triangle.

The corners we put on the slide — Access, Quality, Cost — describe what’s getting traded off. They don’t describe who’s doing the trading. Behind each one is a person who absorbed a share of the $1,840 colonoscopy we’ve just traced.

• The patient wanted fast access to preventative care that was supposed to be free, and walked away owing money she hadn’t budgeted for - because a subscriber number was off by a digit on day one.

• The provider delivered good quality care, and had to trust that a coder forty miles from the procedure room would translate two polyps into a billing code that survived a payer’s review.

• The payer wanted cheap cost for the care, and answered the claim with a single line of CAR code - then spent a month watching a denial team drag it back into payment anyway.

Patient, provider, payer — three parties, each capable of bearing some share of the load, with a constraint that the load has to fall on at least one of them. The cycle is the apparatus that moves it between them.

Every fix-the-revenue-cycle initiative I’ve worked on hits the same ceiling. You can scrub claims cleaner, deny less aggressively, schedule more efficiently, code more accurately. You can deploy AI on either side of the payer/provider line and shave a few days off A/R. What you cannot do is make all three parties’ lives easier at the same time. Burden of work is conserved.

Exceptions that Prove the Rule

To steelman my metaphors, there are two objections worth addressing.

The first: there are really more than three parties. Behind every payer stands an employer (commercial coverage) or a taxpayer (Medicare and Medicaid), and the load the payer absorbs eventually flows back to them as premiums or taxes. Factually correct, and it strengthens the metaphor instead of weakening it. The employer paying premiums doesn’t see the denied claim. The taxpayer funding Medicaid doesn’t see the appeal (neither do I). The patient sees the EOB but can’t read it. Meanwhile, the hospital’s denial team and a payer’s medical director negotiate a roughly $1,800 line-item on a first-name basis. The system keeps the people most exposed to the load farthest from the work that decides who eats it.

The second: the trilemma is an artifact of US payment fragmentation. Single-payer countries don’t have it. Kaiser doesn’t have it. Eliminate the cycle and the constraint disappears.

The cycle disappears, but the constraint doesn’t. Integrated and single-payer systems internalize the load instead of litigating it, constrained by the same triangle but at a different angle:

• Access — NHS England. NHS England doesn’t run prior-auth fights, but it does ration access - through wait times. The elective list peaked past 7.5 million patients in 2024. The access constraint exists in single-payer just as it does in fee-for-service; it just looks like a queue instead of a denial.

• Quality — Kaiser Permanente. Kaiser is integrated payer-provider, so the medical-necessity decision happens inside one wall instead of across two. The trade-off about which procedures get approved still gets made - by Kaiser’s physicians, using Kaiser’s protocols. Integration doesn’t eliminate the gatekeeping, it merely relocates it.

• Cost — Canada. Canada’s provinces don’t run claim denials, but they do ration cost - but it’s upstream, through formularies, wait times, and provincial capital budgets that decide which hospitals get which scanners. The cost trade-off happens at budget time, not at billing time.

Integrated and single-payer systems do shrink one part of the load — the deadweight of two organizations litigating the same claim goes away when there’s only one organization. The underlying constraint doesn’t. The iron triangle isn’t an artifact of US payment fragmentation; it’s the geometry of finite-resource healthcare itself. Pick a system and you can see which corner it bends. You won’t find one that bends none of them.

Try Angles

Back to the United States, where the cycle is the apparatus. There are three configurations, and all three are hard:

• Easier for the patient. The load moves to the payer-provider fight. The fight gets longer, more expensive, more byzantine. The patient gets a smoother experience and a more opaque one - no real idea what just happened, who paid, or what she would have owed.

• Easier for the provider. The load moves to the payer (who tightens the screws) and to the patient (who absorbs more of the friction by filling out more forms or confirming more appointment reminders). This is what happens when a hospital outsources its revenue cycle to a third party with aggressive collections.

• Easier for the payer. The load moves to the provider (more documentation, more denials to appeal) and to the patient (more coinsurance, narrower networks, more prior-auth steps).

You’ll notice I haven’t written a fourth option: there isn’t one. You can grow the pile, which is roughly where the country has been heading for thirty years, or shrink one party’s load by moving it to another desk (basically what I do at work to improve my productivity).

This is why “automation will fix the revenue cycle” doesn’t quite work. Some specific frictions do shrink. Real-time eligibility checks prevent denials from getting written. Electronic remittance advice eliminated a category of paper that used to consume FTEs.

But the last decade of automation has mostly moved the pile, not shrunk it. Payers automate denials; providers automate appeals; the work goes faster, the work is still there. HFMA calls the pattern the battle of the bots. Whoever automates faster sends more load to the slower party’s desk. Total cost — in dollars, hours, patient anxiety, and clinician moral injury — has not fallen.

There is an unstated problem here: if you want it easier for the patient, the payer and the provider have to absorb more of the load - and neither has any structural reason to. The American revenue cycle is what you get when no one with the power to shrink the pile is paid to shrink it. Payers are public companies with margin targets. Providers are squeezed between rising labor costs and flat Medicare rates. Patients have no leverage individually, no lobby collectively, and no clear path to push back. The people who could change the cycle are the same people who would lose if it changed.

Subscribe now

So the next time you read about a fix — a new AI tool, a payer initiative, an interoperability mandate, an EHR upgrade promising to “transform” the revenue cycle — don’t ask whether it works. Ask which party absorbs the new load, and whether that party has any reason to.

Editor's note: This is the 1st of 5 essays in The Cycle, my series on healthcare's revenue cycle. Ongoing writing to continue here.

I don’t know whose MBA-wet-dream coined the term “revenue cycle”1. It’s a jargonist term that’s not a cycle at all! The name is so disorienting that it obfuscates almost all semantic meaning - no hint of what it does, no reference to healthcare, no indication that a human being is involved anywhere in the process. If someone tried to impress you with “revenue cycle management” at a cocktail party, you’d politely sip your drink and walk away, thinking how full of shit they were.

First, it’s not a “cycle” in any operational sense. It has a start (scheduling) and an end (bad debt write-offs). But it is a “cycle” in the way that life and death and rebirth are a “cycle”: something from which we can’t escape. From the moment you’re born in an American hospital, your first breath creates a billable event. And long after you’re put in the ground, some poor soul will be writing off the insurance balance that goes uncollected.

American healthcare sits in an absurd position: the most advanced medical science funded by an enormous economic engine - paired with the worst consumer healthcare experience in the developed world. We are caught in the cycle without even knowing it. And despite the good intentions of healthcare providers striving to deliver patient care, the revenue cycle is largely seen as an administrative burden, with an estimated $350 Billion cottage industry2 built on extending the reach of the healthcare industrial complex. The common refrain is to cut the administrative waste - automate the tasks, remove the labor costs, shrink the bureaucracy that perpetuates this behemoth - and in doing so, fix American healthcare.

But that’s the wrong diagnosis.

The entire revenue cycle - the scheduling, the eligibility, the coding, the billing, the appealing, the collecting, every financial and administrative interaction that wraps around clinical care - is not some tumor to be excised from an otherwise healthy body. The clinical encounter (the part everyone thinks of as “healthcare”) is the briefest part of the whole experience. The average doctor’s visit is eighteen minutes of face time. And if you’ve ever been on hold with a hospital billing department long enough to hear the hold music loop three times, congratulations - you’ve spent more time in revenue cycle than your last doctor’s visit.

Revenue cycle pervades all parts of the healthcare experience. Patients spend hours nervously researching how to access care. They stew with anxiety hoping that their insurance coverage will take care of the labyrinthian accounting rules required to pay for it. They spend more time filling out forms in waiting rooms, more time opening mail explaining their financial burden, more time on hold and on the phone - than they ever spend with a physician. Seventy-three percent of insured Americans dealt with at least one administrative task last year - scheduling, prior auth, billing disputes, insurance questions - and one in four delayed or skipped care entirely because the paperwork was too much. Yet behind the scenes, someone is feeding pages through a fax machine that whirs and jams, sending the same prior auth form for the third time this week, for a single patient.

The industry sees revenue cycle as a back-office function. Patients experience it as the front door.

And the industry’s response to this - the most patient-facing part of the entire system - is to automate it. Some of that automation genuinely helps. Auto-posting payments, eligibility verification, claim status checks - these are tasks that don’t need a human, and removing them frees up the people who do the work that does. Nobody mourns the loss of mind-numbingly manual claim status checks. Work like that makes people update their LinkedIn. But that’s not where the industry is stopping.

Automation is only helpful if it's not adversarial. Payers deploy AI to deny claims more efficiently, and providers respond by deploying AI to catch those denials and appeal them. The HFMA calls it "the battle of the bots" - an arms race where both sides spend more to maintain the same stalemate, and the patient's claim is the territory being fought over.

All the meanwhile, healthcare industry spending grows unabated. The RCM market is projected to grow 10-13% annually through 2030. The industry that wants to shrink the revenue cycle is pouring money into it - just not into the people who do the work. The investment goes to software that replaces the human, not software that helps them. The humans are being taken out of the loop at exactly the moments a human matters most - when a patient is scared, confused, broke, or staring at a bill they can’t read.

And ultimately, who is this for? It’s delusional to think that AI agent robodialers interrupting patients at suppertime is progress.

The assumption underneath all of it - the automation, the data exchanges, the interoperability initiatives - is that the patient experience is a friction problem. That if you make the process fast enough, smooth enough, invisible enough, it stops being painful. But the pain isn’t friction. The pain is opacity, contempt, and the feeling that the system wasn’t built for you.

All of it gets built while the deeper existential question goes unanswered: why do we have to do this at all?

Not why do we need the revenue cycle - in the system we’ve built, it’s load-bearing. But why have we built a system where the thing patients experience most is the thing the industry is most embarrassed by? Where the primary interface between healthcare and the human being it’s supposed to serve is a process so opaque, so adversarial, and so poorly resourced that a hundred million Americans carry medical debt.

Nearly every health system in America has a mission statement about “patient-centered care” (if you’ve read one, you’ve read them all - something about compassion, innovation, and a stock photo of a diverse group of concerned clinicians). The healthcare industry focuses on the Triple Aim - access, quality, cost - as though these are aspirations. The revenue cycle is where those aspirations get tested - and mostly fail. Access becomes a prior auth fax sent to an unchecked machine. Quality becomes a code selection that determines payment, not outcomes. Cost becomes a cruel billing call, striking fear in the hearts of family members who have already jumped through enough hoops as it is. What healthcare says it is and what the revenue cycle actually does is the gap between the brochure and the building.

The people who work inside revenue cycle see more of this than anyone. The scheduler helping the daughter of the Vietnamese-speaking patient find the right appointment. The coder making sure the documentation captures the full complexity of a patient’s condition, because the difference between two codes is thousands of dollars in reimbursement for the same care. The biller whose follow-up call becomes a counseling session, easing the fear of a dad recovering from a surgery, staring at a number he can’t pay. These are the people with the most comprehensive view of what’s broken in American healthcare - where documentation fails, where payer rules punish the wrong people, where patients fall through gaps. And the industry’s plan for them is to spend less.

The revenue cycle isn’t a cycle. It’s not a department, or a cost center, or a line item buried in the back pages of a hospital’s annual report. It’s the chassis American healthcare is built on, whether the industry wants to admit it or not. And right now, the plan is to run it with fewer people, less money, and more machines, without ever asking the people on the other end of the phone whether they wanted to talk to a machine in the first place.