With every new release from OpenAI, Anthropic or Google, a question comes back to hospital board meetings: if artificial intelligence has become so good, won't it solve billing, denials and cash on its own? The short answer is no. And the reason isn't a lack of model capability, but a lack of access to what really matters.
An AI model knows what it has read
Large language models are trained on huge volumes of publicly available text from the internet, complemented by licensed datasets. That is why they know medicine, regulation, the published TUSS table, ANS rules and the TISS standard. They write, summarize and reason very well about all of it.
But a model only learns what it has seen. And the data that determines whether a bill will be paid, denied or recovered has never been on the internet.
The data that drives revenue is private
The hospital revenue cycle runs on a layer of information that belongs to each hospital and to each commercial relationship with payers. It is private data, protected by contract, trade secrecy and the LGPD:
Contracts between hospital and payer
Each contract has its own rules on coverage, bundles, daily rates, fees and exceptions, plus amendments that change over time. None of them is public.
Tables and prices actually negotiated
The TUSS table standardizes codes, but the amount each payer pays for each code, for each hospital, is restricted commercial information.
Brasíndice and Simpro applied to each contract
These are licensed price references for drugs and supplies. And what defines revenue isn't the table itself, but how each contract applies it: percentages, deflators and exceptions.
History of denials, rejections and appeals
Which payer denies what, for what reason, with which auditor, and which arguments reverse the denial. That is the knowledge that separates an accepted appeal from a rejected one.
Bills, procedures and charges
What was prescribed, performed, charged, billed and actually received, bill by bill, over the years.
Each payer's payment behavior
Actual payment times, delay patterns and seasonality. The basis of any cash forecast.
No generic model learns the rule of a contract it has never read, or predicts the denials of a payer whose history it has never seen.
Why this blocks a generic solution
Without this data, a general-purpose model can explain what a denial is, but it can't tell which bill, from this hospital, with this payer, is most likely to be denied tomorrow. It can draft a well-written appeal, but it doesn't know which argument this payer usually accepts. It can calculate a price, but it doesn't know the deflator the contract applies to Brasíndice.
And it isn't enough to send this data to a general-purpose assistant. Health information is sensitive data, contracts have confidentiality clauses and operations require traceability of every decision. Solving the revenue cycle calls for models trained in this context, in a restricted environment and with accountability for results.
Where Accrua comes in
Accrua doesn't compete with general-purpose models. We use the best of them as a starting point, and build our own restricted models on that foundation, trained on operational data from the hospitals where we run the revenue cycle: contracts, bills, denials, appeals and payments.
The result is agents that know the rules of each contract, the behavior of each payer and the history of each type of appeal. In this territory, a specialized, well-fed model is more effective than any generic model on the market. And, as across all of Accrua's operations, specialists validate critical decisions.
A network effect that really works
Every hospital that connects to Accrua makes the models better for all the others. The cycle reinforces itself with every new bill processed:
- 01 →
More hospitals connected
More contracts, bills and payers represented.
- 02 →
More patterns identified
Each payer's behavior becomes clearer with every denial and every appeal.
- 03 →
More accurate models
Predictions of denials, appeals and payment times become more accurate.
- 04 ↺
More revenue recovered
Every hospital in the network, including the newest, captures more results from the first month.
This only works with a clear rule: each hospital sees only its own data. What the network shares is aggregated, anonymized learning about patterns, never contracts, negotiated prices or patient information from another institution.
Good for the whole industry
When bills arrive correct the first time, denials drop, appeals decrease and the relationship between hospitals and payers stops being a permanent dispute. It is the convergence agenda the industry has been discussing, as at Conahp 2026, brought into day-to-day operations.
The disruption of the private healthcare revenue cycle won't come from a generic model trained on the internet. It will come from those with legitimate access to the right data, who know how to protect it and turn that knowledge into results for hospitals, payers and patients.