Artificial intelligence in hospital auditing: what changes

Hospital auditing has always been a race against the clock. A team sized for a smaller volume than today’s, thousands of claims a month, different rules for each payer and a medical record that no one wrote with the billing specialist in mind.

In this scenario, the way out was sampling. You audit what there is time for, and the rest goes to the payers with the errors it had before it got there.

For decades, this was the only possible model. Volume grew faster than review capacity, and the errors that slip through have two known destinations: they become denials, when the payer finds them, or they become underbilling, when both sides miss them.

The use of artificial intelligence is already a reality in Brazilian hospitals, notably full claim coverage, detection of unbilled items, automatically generated denial appeals and a shorter payment cycle.

This article explains how this transformation works in practice, what AI already does in hospital auditing today, what remains human work and how to generate results from the first months of using AI.

What changed in auditing: from sampling to full coverage

Auditing by sampling was a reasonable solution to a scale problem. The hospital picks the highest-value or highest-risk claims, reviews them carefully and sends the rest. It works when volume is manageable and contracts with payers have some predictability.

Both conditions deteriorated together. Patient volume grew, the number of payers with their own rules multiplied, and the denial rate followed: the average initial denial rate at ANAHP (National Association of Private Hospitals) hospitals went from a historical range of 3% to 5% to 15.89% in 2024, according to the Anahp Observatory 2025, and reached 17% in the first quarter of 2025.

What escapes review has two destinations. The first is the denial: the payer finds the error, rejects the item and the hospital enters the appeal cycle. The second is underbilling: items delivered during care that simply do not appear on the claim. That money never comes back, because it was never billed.

Artificial intelligence solves the problem at the root. The model reads medical records, prescriptions, nursing progress notes and procedure records, compares them with the medical claim and identifies discrepancies before submission, across every encounter, with no sampling.

What artificial intelligence does in hospital auditing

AI in auditing works on four main fronts, each with a direct impact on the revenue cycle.

Reading medical records and clinical documents

Electronic medical records are written in natural language, with variations in terminology among physicians, services and shifts. Using natural language processing (NLP), AI does this reading at scale: it identifies which procedures were performed, which materials were used and which medications were administered. Cross-checking against the medical claim reveals the items delivered clinically that do not appear in billing, the main driver of hospital underbilling.

Validating claims before submission

Before the XML file is sent to payers, as regulated by ANS (Brazil’s National Supplementary Health Agency) Normative Resolution No. 305/2012 (TISS standard), AI checks the consistency of each item: TUSS code compatibility, consistency between diagnosis and the procedure billed, compliance with each payer’s contract rules and completeness of the required documentation.

Concurrent audit

Concurrent auditing takes place during the hospital stay, before the claim is closed. The earlier an inconsistency is identified, the lower the cost of correction and the lower the risk of denial. AI makes this coverage possible at scale by processing data in real time. If a procedure was authorized for a given code and the clinical record points to a different course of action, the alert reaches the team during the stay.

Automatic generation of denial appeals

When a denial arrives, AI identifies the reason, locates the clinical evidence in the medical record and generates the appeal with technical grounds. Hospitals that have adopted this automation report appeals completed in under ten days, compared with weeks in the manual model, according to data from the Futuro da Saúde portal.

The table below summarizes the before and after in the main stages of auditing:

The impact on the revenue cycle: what the numbers show

The gains from AI in auditing spread across the entire hospital revenue cycle. Time to payment, cash predictability and the institution’s capacity to invest respond directly to the quality of auditing.

Hospitals that implemented automation in the revenue cycle recorded a 24% reduction in billing times and a 32% reduction in payment times, with denial appeals completed in under ten days, according to data from the Futuro da Saúde portal. In an industry where the average time to payment exceeds 80 days, every week recovered represents extra working capital.

Underbilling has a different financial effect from denials. An appealed denial returns to the cash flow after weeks or months of rework. An item billed correctly from the start enters the cycle with no dispute, no appeal deadline and no operational cost of recovery. For hospitals operating on thin margins, that difference determines the capacity to invest in patient care.

Every real recovered from underbilling is new money in the bank. An appealed denial comes back after months of work. An item billed correctly enters the cycle with no rework.

What remains human work

Audit automation redistributes the teams’ work. High-volume, low-cognitive-complexity tasks go to AI. Auditors focus where expert judgment is decisive.

Denial appeals with legal implications, clinical cases with multiple possible interpretations, disputes that require direct negotiation with payers: these situations call for contextual reasoning, experience and argumentation skills that automation does not reproduce.

The model that delivers results is AI with human supervision. Technology covers the volume, generates alerts and produces appeals. The specialist reviews the critical cases, decides in the gray areas and maintains the relationship with payers. This arrangement makes it possible to audit 100% of claims without demanding more time and work from the team.

There is also a governance dimension that depends on human decision. When an AI model misclassifies a claim or generates an appeal with insufficient grounds, someone needs to identify the error, correct it and make sure it does not happen again. This ongoing supervision requires technical seniority and knowledge of the industry.

Governance and LGPD: what needs to be in place before implementation

AI in auditing works on patients’ clinical data, classified as sensitive data under Brazil’s General Data Protection Law (Law No. 13,709/2018). The institution must ensure that the platform’s access is restricted to what is necessary, that all data processing activities are documented and that every decision made by the model is traceable.

Traceability is especially critical in auditing. When an automated decision is questioned by the payer or the patient, the hospital needs to explain the reasoning. Platforms that operate without auditable logs create a regulatory and legal liability that can outweigh the operational gains.

The recommended minimum governance structure involves a committee with clinical, IT and legal participation, with defined responsibilities for each stage of the model’s life cycle: from implementation to maintenance and periodic revalidation.

A new generation of hospital auditing

Efficient hospitals save more lives. This sentence carries clinical weight: every real recovered from underbilling or saved on denial rework can become a bed, equipment, a nurse’s salary, a working ICU.

Hospital auditing with artificial intelligence represents a paradigm shift in the revenue cycle. The process becomes preventive: errors identified before submission, items captured before they turn into losses, appeals generated before the deadline expires. The revenue the hospital is entitled to arrives on time, without the appeal cycle that consumes cash and staff.

Rivio builds this infrastructure with specialized AI agents that work at every stage of the revenue cycle: from reading the medical record to sending the XML file, from detecting underbilling to automatically generating denial appeals.

The model is built on a straightforward principle: the vendor only earns when the hospital gets paid. It is the guarantee that aligns both sides around the same goal.

FAQ: frequently asked questions about AI in hospital auditing

Does AI replace the hospital auditor?

No. What happens is that the auditor’s role changes. AI takes on high-volume, low-cognitive-complexity tasks: reading medical records, validating fields, cross-checking contract rules. The auditor works on cases that require clinical judgment, negotiation with payers and decisions with legal implications. Team productivity grows because time goes where it creates the most value.

Which audit stages can be automated?

The main ones: reading and interpreting medical records, validating claims before the XML file is sent, detecting underbilling, cross-checking against each payer’s contract rules and generating denial appeals. Clinical cases with multiple interpretations and appeals with complex legal implications still depend on human judgment.

Does AI in auditing comply with the LGPD?

It depends on the implementation. The platform must respect the LGPD principles for sensitive health data: access restricted to what is necessary, documentation of processing activities, traceability of decisions and auditable logs of every action taken by the model.

What return can be expected from audit automation?

The most frequently reported indicators: a 24% reduction in billing times, a 32% reduction in payment times, denial appeals completed in under ten days and recovery of 5% to 8% of billing from previously invisible underbilling.

How does AI-powered auditing fit into the hospital revenue cycle?

Auditing is the first control point of the revenue cycle. When it works well, errors are corrected before claims are submitted, denials decrease, timelines shorten and cash predictability improves. Covering 100% of encounters is precisely what the industry’s current volume has made unfeasible in the manual model.