What is the study?
It's a capstone project for the Master of Health Administration at Marshall University, in the United States, published in May 2026 in the university's digital repository [1]. The author, K'reesa Webster, reviewed the literature to answer one question: do hospitals that use AI in the revenue cycle have better financial results?
The method followed the steps of the PRISMA 2020 protocol, used in systematic reviews, but the work doesn't present itself as a systematic review, as required by the course [1]. A second reviewer, the advisor, checked the selection and data extraction. The work also includes an interview with a revenue cycle leader.
How the 27 studies were selected
The search covered PubMed, CINAHL, Scopus and Google Scholar, in English, for studies from 2015 to 2026 that measured at least one of three financial KPIs [1].
| Stage | Number of studies |
|---|---|
| Records found in the databases | 642 |
| After removing duplicates | 428 |
| Excluded after reading title and abstract | 316 |
| Read in full | 112 |
| Excluded after full reading (no numerical result, clinical focus or insufficient method) | 85 |
| Included in the analysis | 27 |
The included studies are retrospective, observational or quasi-experimental: they compare before and after AI adoption, without randomizing groups [1].
What the numbers show
Every KPI moved in the expected direction, with wide variation between studies [1].
| Indicator | Studies that measured | Before AI | After AI | Average effect (range) |
|---|---|---|---|---|
| Claim denial rate | 19 | 8,7% a 14,2% | 5,1% a 9,8% | Relative drop of about 31% (12% to 35%) |
| Days in accounts receivable | 14 | 42 to 67 days | 28 to 52 days | Drop of 14.3 days (8 to 21 days) |
| Operating margin | 9 | −2,1% a 4,3% | 0,3% a 7,1% | Increase of 2.4 percentage points (1.2 to 4.1) |
Source: Webster, Marshall University, 2026 [1]. US studies published between 2015 and 2026.
The largest effects on denials appeared in studies using predictive models and pre-submission claim validation, with drops above 25%. Automated coding validation had a smaller effect, 10% to 18% [1]. On payment time, automated tracking of open claims cut 10 to 17 days and AI prioritization 12 to 20 days [1].
As operational effects, the review records an 18% to 25% gain in coding accuracy, a 25% to 40% reduction in manual work and a 12% to 30% reduction in claim processing time [1].
Why these numbers aren't a promise
The numbers show a consistent direction, but there are five reasons not to use them as targets:
- Sample not reported. The tables summarize five studies per KPI, and none of them gives a sample size [1]. The full list of 27 studies isn't presented.
- Publication bias. Projects that worked are more likely to become papers. The author acknowledges this risk [1].
- Before and after, with no comparison group. Other changes in the same period (new contract, new team, payer changes) may explain part of the gain.
- An interview without AI use. The interview was with a single revenue cycle leader from a free nonprofit clinic that didn't use AI tools in the revenue cycle between 2023 and 2025 [1]. It helps understand the barriers, not measure results.
- A different health system. Payment denials in the US and claim denials in Brazil are measured differently, and the US starting collection period is shorter than Brazil's.
How to compare with Brazilian hospitals
The review's KPIs have Brazilian equivalents, but with different calculation bases. That's why the percentages don't transfer directly.
| KPI in the review (US) | Equivalent in Brazil | Verified Brazilian benchmark |
|---|---|---|
| Claim denial rate (% of claims denied) | Initial denials | 15.93% of amounts receivable from payers, Anahp hospitals, 2025 [2] |
| Days in accounts receivable | Average collection period | 77.35 days, Anahp hospitals, 2025; 73.84 days in Q1 2026 [2][3] |
| Operating margin | Operating margin (Anahp publishes EBITDA margin, a different KPI) | EBITDA margin of 11.05%, Anahp hospitals, 2025 [2] |
One possible reading: if AI shortens the collection period by days, not by a percentage, the cash effect is larger where the period is longer. A Brazilian hospital with a 77-day period that cuts 14 days frees up the equivalent of 14 days of payer billing. The calculation for your hospital is in the Impact Calculator.
What the interview adds
The interview confirms barriers any billing manager knows [1]:
- Team hesitation. Even with an automated tool available, the team preferred the familiar manual process, and many errors blamed on the system were user errors.
- Manual denial analysis. The interviewee reported having to download and analyze thousands of rows of denial data to find patterns a tool would find faster.
- Return under scrutiny. The tool has to generate gains that pay for its cost, and management should hold it to that.
- Data and privacy. A partner organization banned AI meeting note-takers over cloud storage concerns; and poor source data produces poor results.
In Brazil, the last point falls under the General Data Protection Law, which treats health data as sensitive data [4].
How to measure AI's effect at your hospital
The review's most applicable lesson is its design: measure the effect on three KPIs at once, because an operational gain only matters if it shows up in denials, payment time and margin.
| Step | What to do |
|---|---|
| 1. Baseline | Measure initial denials, final denials, average collection period and margin for at least 6 to 12 months before the project, with a fixed formula |
| 2. Scope | Start with one step and one group of payers, leaving another group without the tool as a comparison |
| 3. Seasonality | Compare the same period of the year, because denials and payment time vary with payers' calendars |
| 4. Breakdown | Break results down by payer and denial reason, to know what the tool solved |
| 5. Total cost | Add up license, integration, training and team time to calculate the return |
Where to start
- Measure the baseline of the revenue cycle KPIs before any project.
- Ask the vendor where its numbers come from: sample, period, hospital profile and comparison group.
- Run a pilot with a comparison: one step, one group of payers, a period of at least 6 months.
- Calculate the return with the total cost, including integration and team time.
- Prepare the team from the start, presenting the tool as support for their work.
Access the full paper
Artificial intelligence–enabled revenue cycle management and financial performance in healthcare organizations
K'reesa Webster · Research paper, M.S. in Healthcare Administration, Marshall University, 2026 · Advisor: Alberto Coustasse · PDF in English, 25 pages
The paper is freely available in the Marshall Digital Scholar repository. The analysis above is Accrua's and was not reviewed by the author.