The account went out. The payer paid less. And the hospital only finds out how much it lost when the remittance arrives — sometimes 60 days after the care, when nobody remembers why that line was on the account in the first place.
The agent audits the account against the patient chart, flags the item the documentation does not support and assembles the appeal with the evidence. Signing the correction is still the hospital's job.
Anyone who works in hospital billing knows this by heart. We spell it out anyway, because the rest of the page depends on us naming the same thing: the account goes out whole, the payment comes back smaller, and the difference has a name, a cause and a deadline. In Brazil the denied line is called a glosa, and the payer is the health plan that holds the contract.
The item exists and the care happened, but the paperwork does not close: a claim form with no authorization, an out-of-date member number, a mismatched date, a mandatory field left empty, a claim sent after the window shut. It is the denial that stings most, because it was almost always avoidable — and almost always recoverable once the evidence shows up.
The payer questions whether the item was warranted: supplies above what the procedure calls for, an item with no justification in the patient chart, a code that does not match the room rate billed, a quantity outside the usual range. Here the appeal is not paperwork — it needs someone on the clinical side to stand behind what was done.
A cut to the amount with no line-by-line criticism: the payer pays a percentage of what was submitted and keeps the rest without pointing at a single line. It is the hardest one to work, because there is no denied line — there is an account paid short, and you have to rebuild the difference before you can dispute anything.
It is the payment return: which accounts were processed, how much was paid on each one, which items were denied and under which reason code. It arrives per closed billing period, in the payer's own layout, one file per payer.
Every contract sets how many days the hospital has to appeal a denial once the remittance lands. After that, the line stops being an argument: it becomes a loss. No appeal, no negotiation, no later invoice brings it back.
The problem is almost never disagreeing with the cut. It is finding the cut too late, in a file that sat in a shared folder waiting for somebody to have a free week.
A denial that expires is not a financial loss. It is an administrative loss wearing a financial loss's clothes.
A Squad is an orchestrating agent: no channel, no waiting for a question. It has a schedule, a set of tools and a job. This one's job is to match, line by line, what the account asked for against what the payment brought — and to separate what was paid, what was denied and what was paid short.
Reconciliation is simple arithmetic that nobody has time to do at the volume it shows up in. The Squad does the whole calculation, every day, and only calls a person when a decision is left over.
api_get_payer_remittance brings the billing period, the accounts paid and the items carrying a denial reason, one payer at a time.api_get_patient_account returns what was submitted: procedures, supplies, drugs, room rates and fees, line by line.api_post_denial_appeal files the dispute with the item, the reason, the attached evidence and the deadline that specific contract allows.It is the same mechanism as the Squads already running in other operations: a schedule, tools declared one by one, a logged run and a stop button. What changes here is the vocabulary.
The payer publishes the remittance for the period. The Squad picks it up on the next run — it does not wait for somebody to announce that the file arrived.
Every paid line finds the line that was submitted. Whatever is left unmatched becomes a declared exception, with the account number and the item that did not line up.
The reason the payer stated is translated into the cause that produced the item back upstream — authorization, price table, charting, deadline or coding.
It finds the evidence, assembles the dispute and marks the contract deadline. If hospital policy requires human approval, it stops here and says so.
Line-by-line matching is the part nobody can do at real volume. Below, the shape of the comparison — with no amounts, because what matters here is the outcome of each line, not the figure.
No column on this screen carries a patient name, a diagnosis or any clinical data.
| Account item | What the remittance brought | Stated reason | Root cause | Outcome |
|---|---|---|---|---|
| Main procedure | Paid in full | — | — | Reconciled |
| High-cost supply | Denied | no justification for the item | Clinical charting | Appeal assembled |
| Closed-unit room rate | Denied | incompatible code | Coding | Appeal assembled |
| Operating-room fee | Paid short | no reason given | Payer processing rule | Difference rebuilt |
| Formulary drug | Denied | amount above the contract | Price table mismatch | Appeal with contract |
| Supporting exam | Denied | no prior authorization | Authorization | No evidence · returned |
| The three piles of reconciliation: paid in full + denied with a stated reason + paid short with no line flagged. No item stays outside one of them. | ||||
Reconciliation stops being a month-end scramble and becomes a routine that runs overnight.
Appealing well recovers money once. Finding out why that item was denied stops it from being denied again next month, and the month after. So the agent does not stop at the dispute: it groups denials by the cause that produced them — and the cause almost never sits in the department that issued the account.
| Root cause | Where it starts | Who can eliminate it | Appealable |
|---|---|---|---|
| Missing prior authorization | Front desk and authorization, before the visit | Authorization, alongside the schedule | Depends on the contract |
| Price table mismatch | Price registry and the version of the reference table | Compliance and contracting | Yes, with the contract attached |
| Item with no justification in the chart | Clinical charting during care | Clinical staff and internal audit | Yes, if the justification exists |
| Claim sent past the window | The billing submission window | Transmission and batch closing | Rarely |
| Code incompatible with the room rate | Coding, at account closing | Compliance between care and billing | Yes |
| Paid short with no line flagged | The payer's processing rule | Contracting, with the difference rebuilt | Only with the account rebuilt |
A price table mismatch and an incompatible code never have to become denials: they are compliance deviations the agent's audit catches while the batch is still open, by cross-checking what was delivered against what was billed. That is a different agent on a different front, and it is what actually pulls the denial rate down instead of only improving recovery — preventing is worth more than appealing.
A claim past the window rarely comes back on appeal, because the argument is not about the item — it is about the date. That one belongs to the transmission front: the batch validated before it goes out, technical rejections caught on the spot, and the protocol receipt imported with no manual checking.
Grouping by cause is what makes the denial stop happening. Grouping by payer only tells you who to argue with.
At the end of the run, the Squad publishes a document: what was reconciled, how much of the denied value came back, which causes concentrate the loss and what needs a human decision. It goes out as HTML to read on screen and as a PDF to attach to the revenue committee's email.
Denial reconciliation and recovery
Document published at the end of the 06:00 run, covering the remittances received in this billing period. Mid-to-large hospital network, 28 outpatient specialties, 14 contracted health plans.
Every remittance received was reconciled on the day it arrived. The initial denial rate — the share of billed value the payer cuts before any appeal — came in at 4.1%, against 9.8% at the start of the cycle. Of the denied value actually appealed, 71% came back, against 38% before the routine existed.
Figure 1 — 71% returned + 29% denied or under review = 100% of the value appealed this cycle.
| Root cause | Payers with the highest incidence | Weight this cycle | Trend |
|---|---|---|---|
| Missing prior authorization | Payer A · Payer D | highest item volume | falling |
| Paid short with no line flagged | Payer K | highest disputed value | needs the contract |
| Item with no justification in the chart | Payer A · Payer F | mid volume, high value | depends on clinical staff |
| Price table mismatch | Payer D | residual since the review | fixed at the source |
| Code incompatible with the room rate | Payer F · Payer K | residual | blocked before submission |
| Claim sent past the window | no concentration | nearly extinct | window shortened |
Payer K holds the highest disputed value of the cycle, all of it paid short with no line flagged. The agent rebuilt the difference line by line and left the appeal assembled, but disputing a processing rule is a contract conversation — and a contract is not appealed by routine.
A document, not a dashboard: numbered sections, written analysis, one figure with the proportion of what came back and a table with the cycle's root causes. This is the piece the revenue committee reads in the weekly meeting, with nobody having to build a deck.
The table and the figure are not typed by the agent. It declares which API call the data comes from and the server builds the block from what returns. When a figure is an aggregate, the parts are printed with it — the 71% and the 29% sit right next to the total.
No patient name, no diagnosis, no clinical data. The aggregate dashboard works with item, cause, payer and value. Identifiable data only appears inside a specific appeal, for someone with permission to open that appeal.
The appeal that comes back is the one that arrives with the proof attached. The agent audits the item against the patient chart: it finds the justification that supports it, or records that there is none, attaches what the payer asked for and files the dispute inside the contract deadline. The limits of that autonomy sit right beside it — and matter just as much.
api_post_denial_appeal records the appeal already tied to that payer's contractual deadline, and from then on the clock is visible.When the evidence does not exist, it says where it is missing. An item with no justification recorded in the chart goes back to internal audit named as a charting gap — with the item, what the progress note failed to record and the deadline — instead of becoming a weak appeal filed to tick a box.
The agent's autonomy runs up to the edge of the decision: the work of finding what does not hold up arrives done, and the signature stays with the hospital.
Missing the contract deadline is the one kind of denial with no appeal. So the deadline is not a hidden column in a monthly report: it is an alert that surfaces on screen and in the inbox of whoever answers for the appeal, with the days remaining and what is still missing for the case to go out.
No appeal leaves this queue without someone approving it, when hospital policy requires approval.
The deadline is not a global platform setting. Every contract has its own, and that is where the count comes from: the remittance date, the payer's contractual deadline and the business days left before it expires.
The alert email goes to a closed recipient list. No number in the warning exists outside the document published on the same run.
Recovering more is half the way. The other half is no longer losing for administrative reasons: loss to expired contract deadlines falls 94% — not because anyone started working faster, but because the deadline stopped being discovered in the rear-view mirror.
Orders of magnitude for a hospital of the size described across this section — a mid-to-large network, 28 outpatient specialties, 14 contracted health plans. Each indicator points back to the list that produced it; none of them is an aggregate without an origin.
The share of billed value the payer cuts before any argument. It is the indicator that tells you whether the problem is being solved at the source: it falls from 9.8% to 4.1% of billed value, and that drop comes from compliance before submission, not from appeals afterwards.
High recovery with a high initial rate means the hospital got good at arguing. A low initial rate means it got good at billing.
How much of the disputed value the payer gives back. It goes from 38% to 71% once the appeal stops being boilerplate and starts carrying the item's evidence with it — the authorization, the table version or the recorded justification.
The figure in the dashboard prints the parts: 71% returned and 29% denied or under review add up to the value appealed this cycle.
Days between the remittance arriving and the appeal being filed. From 11 days to 2, because reconciliation stopped depending on a free window in somebody's calendar and started happening overnight, every day.
The time dropped; human approval stays in the path, when the hospital requires it.
The value that never got disputed because the contract deadline passed. It falls 94% — the harshest indicator of the set, because every cent here is a loss that never had a technical argument at all.
Whatever still expires is logged with its cause, so it becomes a review agenda item instead of disappearing.
The daily dashboard answers what was planned. When the question is something else — "which denial causes grew at this payer since the price-table review?" — the one who answers is the Hal-AI Copilot, in read-only mode, with a table and a chart on screen and a PDF export. The same audit trail applies there: what was queried stays on the record.
Manual reconciliation is typing plus checking: open the remittance, find the account, compare line by line, note it in a spreadsheet, repeat fourteen times. It eats the whole week and leaves nobody better at the subject.
Worth saying what does not change: approving an appeal is still a person's job, arguing a contract is still a person's job, and correcting the chart is still the job of whoever records it. What arrives finished is the audit — the item, the evidence that is missing and the likely cause.
These are the five questions that come up in every revenue-cycle conversation. The answers describe the standard design; what changes from hospital to hospital is the approval policy and the access scope.
Whoever the hospital decides. Approval is a policy configured per payer and per denial type: you can allow automatic submission on an administrative denial with objective evidence — an authorization that exists and was attached — and require human approval for anything that touches clinical warrant or contract terms.
When approval is required, the appeal sits in the queue with the evidence already attached and the deadline visible. The agent does not submit, does not nag and does not route around it: it raises the warning again as the deadline gets closer.
Reading it is not a rigid parser pinned to column positions. The agent interprets the return and validates what it understood against a declared schema: if an expected field is missing, or turns up under another name, it marks the run as partial and names what it could not read — instead of reconciling with data missing.
A layout change becomes an exception on that day's run, with the payer and the field identified. The fix is tool configuration, not a rewritten routine.
Only what is needed to support the disputed item, and only when the hospital grants that scope. Pulling the justification for a supply out of the patient chart is not the same as reading the chart: access goes through a parameterized query, restricted to that account and that item, with the purpose on the record.
On the aggregate dashboard, patient data never appears under any circumstance — the columns are item, cause, payer and value. Access scope, legal basis and purpose are defined during the project, before any integration.
Integrations, LGPD — Brazil's data protection law — and what the agent can do on its own
Every run is on the record: which API calls were made, what each one returned, how the items were matched, which evidence was attached, what was classified as an exception and who approved the submission. The trail covers the whole agent, not only the result.
It is the same control that applies to any agent on the platform — curated queries instead of open access, tools declared one by one, and runs with a history and a stop button.
No. Appealing everything is an expensive way to lose credibility with the payer and to tie up your own team with cases that never come back. When the evidence does not support the item, the agent closes it as an accepted denial and hands the cause back to whoever can treat it at the source — authorization, clinical charting or coding.
That is why the root-cause table exists. An accepted denial is not a failed appeal: it is the most useful information in the cycle, because it shows where the process has to change so it does not repeat.
Tell us how your revenue cycle works: how many payers, which system holds the account, how long each contract gives you to appeal. We design the reconciliation Squad with you, using the tools and the access scope your operation allows.