Revenue cycle management is how clinical work turns into cash. It has a reputation for being tedious, and it earns it — but almost none of it is arbitrary. Every form, code and modifier exists because somebody had to prove something to somebody else. This is an illustrated walk through the whole machine, from booking an appointment to a balanced ledger.
Click anything underlined in teal. Terms like adjudication or CARC open a short definition wherever they appear. Inside the panels, most things are clickable too — table rows, diagram boxes, form fields, ledger entries. If something looks like it might explain itself, it does.
Why so many fields? A claim carries about forty fields to describe a forty-minute appointment. That sounds absurd until you see what the fields do. They do not describe the treatment. They prove entitlement: who was licensed to do this, on whose order, for which diagnosis, with what permission, at what place, under which contract. The clinical event is small. The proof around it is large.
One metaphor runs throughout. A practice is a freight terminal. Every claim is a shipment that must clear customs before anyone gets paid.
This is a mental model, not a rulebook. The real rules live in the CMS Medicare Claims Processing Manual, your MAC's coverage determinations, the X12 guides and your own payer contracts — and they change every January. Numbers here are illustrative. Verify before you build.
In one line: in retail, one person receives, decides and pays. In healthcare those are three different parties — and every form exists to reconcile them.
Buying a coffee is one transaction. You hand over money, you get a coffee, and the books record one line. Nobody asks whether the barista was licensed, whether a doctor ordered the coffee, or whether this is your eleventh coffee this year and therefore needs justifying.
Healthcare splits apart what retail fuses together. The person who receives the service, the person who authorises it, and the person who pays for it are three parties with three different interests. Everything in this article exists to reconcile them.
So a claim is not a bill. It is a piece of evidence. It asserts that a licensed clinician, under a valid order, treated a covered diagnosis, inside an approved plan, at a recognised place, for a contracted price. Each of those clauses is a field, and each field is something a payer can argue with.
In one line: six parties, and one of them — the "payer" — is secretly five things at once.
The patient receives care and usually owes part of the money. The practice is the terminal: it employs clinicians, holds contracts, keeps the books.
Inside the practice, two roles must stay separate on a claim. The billing provider is the organisation being paid (Type 2 NPI). The rendering provider is the individual who actually treated (Type 1 NPI). They are tracked apart because credentials and money are audited apart. A referring provider is a fourth party who may never set foot in the building, but whose name and number the claim legally requires.
The clearinghouse is the freight forwarder — it checks your paperwork against the destination's rules and routes it. The payer is the customs house, and here precision matters. A payer is not one thing. It is an organisation (Aetna), selling a plan (a specific PPO product), under which the patient holds a policy, governed by a contract with your practice, with a fee schedule attached. Five things. Beginners collapse them into one; working billing systems never do.
In one line: "insurance" is not one thing — a practice deals with eight or nine kinds of payer, and each one changes the price, the paperwork, and who chases the money.
The single most useful question about any visit is who is paying for this? — because the answer changes almost everything downstream: what you are paid, whether you needed permission first, what the patient owes, how you submit, and how long you wait.
The mix of those payers across a practice is called its payer mix, and it drives the economics of a clinic more than any clinical decision does. The same forty minutes of treatment can be worth twice as much on one rail as on another, with none of the difference visible to the patient on the table.
A word on the word "payer". It is never one thing. An organisation (say, Aetna) sells a plan (a specific product), under which a patient holds a policy, governed by a contract with your practice, with a fee schedule attached. "We take Aetna" is therefore not a meaningful statement — you take some Aetna plans, on terms that vary.
In one line: between "patient" and "visit" sits the case — one injury, one referral, one plan, one payer — and almost every rule in this article is scoped to it.
This is the entity people leave out, and leaving it out quietly breaks everything downstream.
A visit is too small to hold any rule worth holding. A patient is too big: the same person can have two open cases at once — a workers' comp case for a shoulder and a commercial-insurance case for an unrelated knee — with different referrers, different diagnoses, different permissions, and different payers. Put charges from both on one claim and you have made a mess that no amount of appealing will fix.
So the case is the box in between. It owns the referral, the plan of care, the certification window, the authorization, the diagnosis list, the goals, and the discharge. It is also the join between the clinical world and the money world: visits hang off a case, claims are built from visits, and the case is what says whether any of it was allowed.
Different counters reset on different clocks, and mixing them up is one of the most expensive bugs in this domain. Opening a new case resets the authorization and the certification window. It does not reset the patient's annual KX threshold spend — that runs per person, per calendar year, no matter how many cases they open. The panel below sorts them out.
In one line: a question asked before treatment costs cents; the same question asked after treatment costs a denial.
The front desk sends an eligibility inquiry — transaction 270 — and gets a 271 back. Is the coverage active? Where is the deductible? What is the copay? How many visits are left? Is authorization required? The exchange costs pennies and prevents denials that cost tens of dollars each to rework.
Authorization deserves its own paragraph, because it is the thing most often modelled badly. An authorization is not a flag on a patient. It is an object with a life. It is requested (transaction 278), approved with four attributes — a number, a unit budget, a date range, and a scope — then consumed visit by visit, and finally it expires.
All four can kill a claim independently. And the classic failure is not running out of visits. It is the calendar quietly passing the end date while visits remain.
An authorization removes one reason to deny you. It does not promise payment, does not establish medical necessity, and does not survive its own expiry date. Payers say so explicitly in the approval letter. Practices misread it constantly.
Under Medicare, two separate windows must both be open on the day you treat: the authorization period (if a plan requires one) and the certification period covering the plan of care. They are different dates from different people for different reasons. Check them as two independent tests, not one.
Different payers put permission in different places. Traditional Medicare needs no prior authorization for outpatient PT at all — its gates are certification and an annual spending threshold. Workers' comp puts permission in an adjuster's hands, scoped to one body part. Self-pay swaps authorization for informed consent and a written price.
In one line: documentation posts nothing to the ledger, yet decides whether anything on the ledger is allowed to stay there.
A case opens with an evaluation. The therapist takes a history, examines, judges how stable the presentation is, and picks one of three complexity tiers — 97161, 97162, 97163. The rule is: default to the lowest tier your weakest qualifying element supports. Only a physical therapist may perform or bill an evaluation. An assistant may not. And under Medicare all three tiers pay the same, so the tier changes your audit exposure rather than your revenue.
The evaluation becomes a plan of care: diagnosis, measurable goals, frequency, duration, planned interventions. This is the spine of everything after it, because medical necessity is judged against it. Medicare additionally wants it certified by a physician or NPP within thirty days, recertified every ninety, and backed by a progress note at least every tenth treatment day.
A progress note is required and unbillable. A re-evaluation (97164) is billable but conditional — it needs an actual clinical trigger, like a significant change or a failure to progress as expected. Billing 97164 every tenth visit because the progress note came due is one of the most reliable ways to attract an auditor.
Now the point that matters for anyone building software: none of these documents post to the ledger. Not the plan of care, not daily notes, not progress notes, not certifications. They are clinical records that govern money without ever being money. Treat documentation as a side-effect of billing and you will get this backwards. Treat it as an independent record that claims must point at and you will get it right.
In one line: there are two different rules for turning minutes into billable units, they disagree, and which one applies depends on who is paying.
Most treatment codes are timed: they bill in nominal fifteen-minute units. But sessions do not divide into neat quarter-hours, so a rule has to decide what happens to the remainder. There are two such rules. That is the whole problem.
Medicare's rule — everyone calls it the 8-minute rule — pools every timed minute in the visit, divides the total by fifteen, and grants one extra unit if at least eight minutes are left over. Remainders from different codes get combined.
The AMA's rule — the Rule of Eights — applies the eight-minute test to each code on its own, and never combines remainders. Many commercial payers use this one.
They can disagree, and the gap is real money. Three codes of nine minutes each — 27 minutes of work — gives you three units under the AMA rule and two under Medicare's, because Medicare pools the 27 minutes into one bucket while the AMA lets each nine-minute block clear the bar on its own. Repeat that across every visit of every day and it is not a rounding quibble. Which is why unit calculation has to be a per-payer setting, never a global constant.
In one line: read a claim as a sentence, and every field turns out to be a clause someone can dispute.
The visit is now a set of coded lines. Turning them into a claim means answering the entitlement questions from §1, field by field. As electronic data it is an 837P; printed, it is the CMS-1500. Same assertion, two containers.
The sentence runs: this organisation (billing provider, Type 2 NPI) bills for this clinician (rendering, Type 1 NPI) who treated this person on this date, at this place (place of service) for this problem (diagnosis codes, with pointers tying each line to one) using these procedures (CPT, modifiers, units) under this order (referring provider) with this permission (authorization number) at these charges.
Modifiers carry meaning the codes cannot. GP says the service was under a PT plan of care. KX attests medical necessity past the annual threshold. CQ discloses that an assistant did the work — and costs you 15% of the payment. The 59 and X-series modifiers claim that two normally-bundled services were genuinely separate.
A modifier is a legal assertion, not formatting. Each one is a promise that your documentation can back it up if someone asks.
In one line: a rejection and a denial feel identical and are completely different — one never reached the payer, and only one can be appealed.
The claim leaves, but not straight to the payer. It goes to a clearinghouse, which validates the file and forwards it. Two acknowledgements come back. A 999 says the file parsed. A 277CA says whether the individual claim was accepted for adjudication.
That second one produces the most misunderstood event in the revenue cycle. A claim turned away at the 277CA has been rejected. It was never adjudicated. No reason code in any meaningful sense, no appeal rights, no decision to overturn — the payer genuinely never saw it. You fix the defect and resend. Do that before the timely-filing deadline and it costs you nothing but days.
A denial is different. It happens later and means the payer looked and refused. A rejection is a lorry turned away at the gate for a bad manifest. A denial is a shipment seized inside customs.
Same money, different remedies, different metrics. That is why clean claim rate and first-pass resolution rate are two numbers instead of one — and why a system that stores both as "failed" cannot tell you which half of your operation is broken.
In one line: pricing is a sequence of reductions, and doing the same reductions in a different order gives a different answer.
The payer starts from its allowed amount. For a commercial payer that comes from a private contract. For Medicare it is computed — and it is worth seeing how, because the whole thing rests on three numbers per code.
Every CPT code carries three relative value units: work (the clinician's effort), practice expense (room, equipment, staff), and malpractice (risk). Each is scaled by a local cost index called a GPCI, then the total is multiplied by a single national conversion factor — $33.4009 in 2026. That last number is set politically and moves every January, which is why nothing in a billing system should ever hard-code it.
Billed minus allowed is the contractual adjustment. It shows up as CO-45, the practice absorbs it, and it is never billable to an in-network patient.
Then the reductions, in sequence. Any unmet deductible comes off. MPPR cuts the practice-expense share of every therapy unit after the highest-valued one — the logic being that the second unit of an hour reuses the same room and the same equipment. Coinsurance moves to the patient. The assistant differential takes 15% if a PTA did the work. And sequestration takes a final 2%.
Sources genuinely disagree about where the deductible sits relative to MPPR. Pick an order, make it configurable, and then check it against real remittances from your own MAC before you trust any expected-payment number your system produces. Every projection in the practice depends on getting this sequence right.
In one line: posting a payment is not recording cash — it is splitting one receivable three ways and proving nothing went missing.
The payer's decision arrives as an 835, carrying — per claim and per line — what was paid and what was adjusted. Every adjustment has two codes: a group code and a reason code. The group code is the one your software must act on.
CO is contractual: the practice absorbs it and may not bill the patient. PR is patient responsibility: it moves to the patient's account and becomes a statement. OA is other, usually meaning another payer is up next. PI is payer-initiated, and Medicare never uses it.
Reading CO as PR is how a practice commits a balance-billing violation without ever intending to. It is a two-character difference with legal consequences.
So posting a remittance is a three-way split of a receivable: cash for what was paid, an adjustment for what the contract disallows, and a transfer for what the patient now owes. The receivable closes only when all three are accounted for. That is precisely the discipline double-entry bookkeeping was invented to enforce.
An 835 can also carry a PLB segment — a provider-level adjustment. That is a recoupment: the payer clawing back an old overpayment by shrinking today's cheque. It belongs to no claim on the remittance, so systems that only post claim lines will never reconcile the deposit to the remittance and will never know why.
In one line: errors cost minutes, rejections cost days, denials cost weeks and cash — and lumping them together hides which one you actually have.
An error is caught by your own scrubber before submission. A missing modifier, a bundled pair, an invalid place of service. It costs staff minutes and posts nothing, because as far as the books are concerned it never happened.
A rejection is caught at the gate — after submission, before adjudication. It posts nothing either, but it costs days. And days are the raw material of days in A/R.
A denial is a decision, so it posts. The zero-payment is recorded with its reason code, the receivable stays open and flagged, and an appeal clock starts running.
The remedy depends on what went wrong. If the defect is data, you send a corrected claim — the same claim resubmitted with frequency code 7, replacing the original. If the dispute is judgement, you file a formal appeal. Both take weeks, which is why a denial's real cost sits in aged receivable rather than in the eventual outcome.
In one line: a billed charge has exactly five possible destinations, and a system that cannot name which one is missing a transaction.
Underneath everything sits the ledger, the only narrator that cannot lie. A practice keeps at minimum: cash, insurance receivable, patient receivable, revenue, contractual adjustments, and discounts. Every event writes at least two entries, and debits equal credits or something is wrong.
The invariant worth memorising: the billed charge is always fully explained. It resolves into cash collected, receivable still open, adjustment absorbed, discount granted, or bad debt written off. There is no sixth destination and nothing evaporates.
When a practice cannot say where a dollar went, the problem is rarely accounting. It is that somewhere, an event got recorded as a status change instead of as a transaction.
Switch payers below and watch identical clinical work — four visits, same notes — settle into four different shapes.
In one line: half the revenue cycle is built once and reused forever; the other half runs every visit — and most failures live in the half nobody looks at.
Everything in this article belongs to one of two timelines, and separating them is the most useful thing you can do with the model.
Some things are built once and then reused by every claim forever: enrolment with each payer, credentialing each clinician, the contracts and their fee schedules, the charge master, the code and edit tables, the clearinghouse connection. This is the slow, unglamorous foundation. Nobody photographs it, and nothing gets paid without it.
The rest is a loop that runs per visit: verify, treat, document, code, scrub, submit, acknowledge, adjudicate, post, resolve. The loop is short. It runs fast when the foundation is sound and is agony when it is not — which is why so many practices misdiagnose a contracting problem as a billing problem.
In one line: almost every figure in this article is a dated parameter, not a constant — and a system that hard-codes them breaks silently every new year.
This is the single most important architectural fact in the domain. The conversion factor, the therapy threshold, the appeal thresholds, telehealth eligibility, supervision rules, the edit tables — all of them moved between 2025 and 2026.
Worse, they must be applied by date of service, not by today's date. Reprocessing a 2024 claim in 2026 has to use 2024's rules. That single requirement forces effective-dating through the entire data model, and it is very painful to retrofit.
In one line: everything above, run once on a single visit, with the ledger keeping score.
A Medicare patient, four weeks into a case, treated for thirty-eight timed minutes. Every object you have met appears in order, with what it concretely is here.
Every object in the terminal, and its one job. click any card for more
Share of claims needing no rework before submission. Grades the front end and the scrubber. Healthy: above 95%.
Share paid on first submission with no rejection or denial. Grades the whole pipeline. Healthy: above 90%.
Average age of the receivable. Grades everything at once, slowly. Healthy: under 40 days.
Collected divided by what was collectable after contractual adjustments. Measures leakage, not pricing. Healthy: above 95%.
Share adjudicated and refused. Separate this from rejections or you will chase the wrong fix.
What the terminal costs to run, per dollar landed. The number that makes the case for every automation.
In one line: the shape is complete; the corners are not.
Deliberately missing: coordination of benefits when a patient has two payers and the secondary claim must carry the primary's decision. Medicare Secondary Payer rules and crossover. Credit balances and refunds. Underpayment detection — comparing what a payer actually paid against what the contract owed, which is a large and badly neglected pile of recoverable money.
Also absent: audits and the documentation requests that arrive years later. Telehealth's shifting eligibility. Group therapy, remote therapeutic monitoring, dry needling, and the other codes that live in the margins. Deposits, payment batches and bank reconciliation. The whole of state-by-state variation — Medicaid rules, workers' comp fee schedules, direct access scope — which is genuinely large and needs its own dimension in any real system.
But the shape holds. Prove entitlement, ship the claim, clear customs, split the receivable three ways, and never let a dollar go unexplained. Everything else is detail — and now you have somewhere to put it.
In one line: one patient, one injury, fourteen visits, one hundred and ninety days — and every object in this article appearing exactly once, in order, with the money moving underneath.
Everything so far has been a part. This is the whole thing — and it is the same clinical work run down three different rails, so you can see what changes and what does not.
The timeline is organised by who is acting. Seven parties get their own lane: the patient, the referring physician, the practice, the clearinghouse, the payer, the ledger, and the bank. Every event sits on the lane of whoever performs it, and an arrow shows where the artifact goes next. Watch the ping-pong: practice → clearinghouse → payer → back. Nothing moves on its own.
These are two different records of the same money, and they get their own lanes because they routinely disagree.
The ledger is the practice's internal account of claims on value — who owes what to whom. Its "cash" balance is an assertion: money we believe we have received. The bank is an external custodian holding actual funds, and its statement is somebody else's assertion about the same thing.
They part company for structural reasons, all of which appear below. A remittance posts on one date and the EFT settles on another. One deposit covers many claims across many patients, so the relationship is one-to-many, not one-to-one. Card payments settle net of processor fees. And a payer recoupment shrinks today's deposit for a claim from six months ago.
Reconciliation is the work of proving the gap is explained rather than absent — and on the commercial rail below, the correct closing variance is not zero.
Each rail tells a different story:
Run all three and five things stand out.
The bank never quite agrees with the ledger, and that is normal. Medicare closes at zero variance. Commercial closes $88.50 apart, permanently, because a recoupment took money for someone else's claim. Self-pay closes $30.60 apart because a processor took its fee. None of those is an error, and none should be "corrected" by adjusting the case. A system that cannot hold an explained variance will either corrupt the books or accumulate a mystery.
The clinical case and the financial case have different lifespans. On Medicare, treatment ends on day 90 and the last dollar resolves on day 190. A system that closes the case at discharge loses track of a third of the money.
The expensive failures were never clinical. On Medicare it was an accumulator nobody wired into the scrubber. On the commercial rail it was a date. In both cases the treatment was correct, well documented, and medically necessary — and the money was still lost.
An authorization can fail four ways, and only one of them is obvious. The commercial rail runs out of calendar while it still has visits left. Because the failure is the practice's, not the patient's, the denied amount is a CO write-off and cannot be billed to the patient at all. The practice simply eats it.
Every rail ends the same way. Different parties, different documents, different reductions — and in all three, the billed charge is fully explained and the receivables reach zero. That invariant is the one thing that never varies.