When AI Makes the Problem Worse

The PHTI Finding

In April 2026, the Peterson Health Technology Institute (PHTI) released Administrative AI: Current Use and Potential Impact — the most credible third-party assessment to date of what is actually happening as payers and providers race to deploy AI across the revenue cycle. The report’s headline finding should stop every rural hospital CFO and every Medicaid Managed Care Organization (MCO) operations leader cold:

PHTI documents an “AI arms race” between payers and providers that is doubling activity volume — more prior authorization submissions, more denials, more appeals, more rebills, more back-and-forth — without solving any of the underlying causes of friction. The report is unambiguous about why: “When applied on top of flawed administrative workflows, data complexity, and incentive structures, AI exacerbates the underlying issues.”

This white paper makes four arguments built on PHTI’s findings:

• Generic AI in revenue cycle is making the system worse, not better. Horizontal AI RCM vendors are accelerating churn — not reducing it — and rural providers and Medicaid populations are paying the highest price.
• Rural health is the population most exposed to this dynamic, and the least able to absorb it. Uncompensated care is the single most consequential lever in rural-hospital financial survival, and the AI arms race is pulling that lever in the wrong direction.
• MCOs are not winning either. The labor-intensive churn of rebills and appeals — accelerated by AI — is now one of the largest manual operating costs inside Medicaid managed care plans. As long as the system runs on imprecise data and reactive workflows, that cost compounds.
• AI alone will continue to exacerbate the problem. AI deployed by operators — designed around data intelligence, clinical reality, and the specific economics of Medicaid and rural care — is the only path to a system where providers, payers, and patients all win. That is the model HRN is built on, and it is the opposite of the horizontal “drop-in” AI RCM products PHTI critiques.

Three Findings That Reshape the Conversation

PHTI convened a January 2026 workshop with senior leaders from health systems, health plans, technology developers, investment firms, and federal agencies. The April report distills what they collectively concluded. Three findings are most relevant to rural health and Medicaid.

AI is accelerating volume, not reducing structural cost

When submitting prior authorizations, generating appeals, or reworking claims gets cheap, both sides of the payer-provider relationship do more of it. PHTI explicitly states there is no existing evidence that expedited prior authorizations translate to “lower average cost per claim factoring in the cost of the AI solution.” The cost moves; it doesn’t disappear. And in many cases the per-claim cost rises because the AI subscription, integration overhead, and exception-handling burden exceed whatever throughput gain the AI produced.

The “bot war” creates new losers

PHTI’s most pointed observation: as AI scribes and coding tools drive higher-complexity coding on the provider side, payers respond with algorithmic downcoding, modifier-level denial logic, and peer-comparison reimbursement reductions. The providers who haven’t deployed AI scribes — disproportionately small, rural, and Medicaid-heavy — get hit with the payer countermeasures without having captured the upcoding gains. They are, in PHTI’s framing, the third loser.

AI applied to broken workflows multiplies the brokenness

The report’s structural diagnosis is what every operator already knows but rarely says out loud: revenue cycle inefficiency is not primarily a speed problem. It is a data-quality, documentation-completeness, and incentive-alignment problem. Putting a faster engine on top of incomplete or imprecise data does not produce better outcomes. It produces more outputs of equally low quality, faster — and now both sides are doing it to each other.

PHTI’s recommendation is that reimbursement policy redesign, not technology, is the primary lever for system-wide savings, with disclosure and oversight frameworks as near-term asks.

Why Rural Health Pays the Highest Price

Rural hospitals have been in financial distress for over a decade. More than 130 rural hospitals have closed since 2010, and a majority of those still open operate on negative margins. The single largest driver, year over year, is uncompensated care — services delivered for which the hospital is never paid, whether through self-pay write-offs, charity care, or denied claims that are never successfully recovered.

For rural hospitals, denied Medicaid claims are not a back-office annoyance. They are existential. A 200-bed community hospital in a high-Medicaid market may have 15–25% of its revenue exposed to MCO denial patterns. When denials are mishandled — under-appealed, late-filed, missing documentation, or written off because the hospital lacks the staff to fight them — that revenue becomes uncompensated care. Multiply across the country and the number is in the tens of billions of dollars annually.

PHTI’s finding lands directly on this population. The “third loser” PHTI describes — providers without AI scribes getting hit by payer downcoding — is, in Medicaid markets, even more acute. Small rural hospitals don’t have AI scribes. They don’t have AI prior-auth tools. They don’t have AI appeals automation. They have one or two billers stretched across multiple service lines, working denials by hand, against MCOs that increasingly use AI to issue and defend those denials. The information asymmetry is not closing. It is widening every quarter.

This is the math that should drive every rural health CFO’s 2026 strategy:

• Generic AI tools deployed by larger, better-resourced systems will continue to inflate coding intensity and trigger payer countermeasures that small rural hospitals get caught in.
• Without their own AI capability, rural hospitals will lose ground in both clean-claim submission and denial recovery — every quarter, compounding.
• Uncompensated care will rise, not fall, even as overall AI deployment in healthcare accelerates.

PHTI’s report does not say this in those words. But it is the unavoidable conclusion when you read PHTI’s findings against rural-hospital balance-sheet realities.

The MCO Side of the Same Coin

A few things have become clear from recent direct conversations with MCO operations leadership. The picture inside Medicaid managed care looks worse than the public narrative.

The labor cost of churn — rebills, reworks, appeals, second-level review, manual documentation chasing — is now one of the single largest manual labor categories inside an MCO’s operations footprint. It is not a side function. It is one of the most expensive things an MCO does. And it is growing, not shrinking, as both sides deploy AI that generates more of this work without solving the upstream causes.

That framing came from an MCO operations leader directly. The cost is too big to ignore. It is also the part of the conversation that horizontal AI RCM vendors miss entirely. They sell the provider a tool to fight denials faster. They sell the payer a tool to issue denials faster. Neither side gets the cost reduction PHTI says doesn’t exist, because the system — the relationship between provider data quality and payer adjudication — never improves.

HRN’s thesis is that the only way to break this cycle is to upgrade the data and documentation layer at the source. When a Medicaid claim leaves a rural hospital with complete, precise, payer-aligned documentation, three things happen at once:

• The provider gets paid faster, with fewer denials and lower recovery cost.
• The MCO’s operations cost drops because the manual rebill and appeal workload it spawns drops.
• The patient avoids the downstream consequences of friction — surprise bills, care delays, charity-care write-offs that affect how aggressively their next encounter gets scheduled.

This is not a marketing claim. It is the only system-design conclusion compatible with PHTI’s findings.

Why Generic AI RCM Cannot Solve This

The category of “AI RCM” being funded and marketed in 2026 is dominated by horizontal platforms that share a similar shape: a portfolio of AI agents that automate denial prevention, denial response, coding optimization, prior auth submission, and similar discrete tasks. The pitch is throughput. The pitch is “more, faster, cheaper per task.”

PHTI’s finding eviscerates this category’s central marketing claim. There is no evidence the per-claim cost falls. There is strong evidence that doing more of the same broken activity, faster, makes the system worse.

Three structural reasons why generic AI RCM cannot solve uncompensated care in rural Medicaid markets:

It optimizes the wrong layer

Generic AI RCM operates on the claim after it has been generated from imprecise documentation. It is downstream of the actual problem.

It is built for high-volume commercial environments

The ROI math on most horizontal AI RCM platforms requires scale that rural hospitals don’t have. The platforms that do reach rural customers typically do so as a cost the hospital cannot actually afford to absorb against the marginal revenue gain.

It is adversarial by design

Generic AI RCM is sold as “fight the payer faster.” That posture is what created the bot war PHTI is documenting. It cannot, by construction, produce the system-level cooperation that would actually drive cost out.

The distinction that matters is not whether AI is involved. It is how AI is deployed and who is accountable for the outcome. AI tools that are sold as self-serve subscriptions and pointed at the back end of a broken claim workflow will keep producing the dynamic PHTI describes. AI tools that are embedded inside an operator service — built for the specific data, documentation, and payer realities of Medicaid and rural care, and accountable to the recovery outcome rather than the throughput metric — operate on a different curve entirely.

That distinction is the entire argument.

The Path Forward — and What to Do Now

PHTI is right that AI alone will continue to exacerbate the problem. The question is what not-alone looks like. HRN’s answer has four components, each of which directly addresses a PHTI finding.

Fix the upstream data layer first

Before any AI agent generates an appeal, an HRN engagement begins by improving the documentation, coding, and eligibility-verification practices that determine whether a claim gets paid in the first place. This is unglamorous work. It is also the only place in the cycle where you can actually remove cost rather than shift it.

Deploy AI inside an operator service, not as a self-service product

HRN’s AI tooling is operated by HRN. Hospitals do not buy a license and try to make it work. They engage HRN’s recovery service, in which AI is one of several leverage points used by trained operators. This is the difference between selling someone a scalpel and providing them with a surgeon. PHTI’s critique applies to scalpel vendors. It does not apply to operator-led service models.

Build for the Medicaid- and rural-specific economics

Rural hospitals have different volumes, different payer mixes, different staffing realities, and different regulatory environments than the academic medical centers that anchor most horizontal AI RCM customer lists. HRN’s service model is built around those realities — not retrofitted to them.

Solve the MCO’s problem at the same time as the provider’s

When a rural hospital submits cleaner data, the MCO’s manual rebill and appeals workload drops. That is the only structural way to reduce the system-wide cost PHTI says is not currently going down. HRN is one of the few operators in this space whose model creates value on both sides of the table simultaneously.

For rural hospital CFOs

• Treat uncompensated care as a strategic, not operational, line item. It is the single highest-leverage variable on your survival curve.
• Resist the pitch that “AI RCM” is a drop-in solution. PHTI now provides the most credible third-party validation that it is not.
• Ask any RCM vendor to show you the system-level cost-per-claim impact of their work. If they cannot, you have just learned something important.
• Engage operators who deploy AI inside a service model and who can demonstrate impact specifically in Medicaid, specifically in rural environments.

For MCO operations leaders

• The largest manual cost line in your operations footprint is being inflated by both sides’ AI deployments. That cost line is not going down on its own.
• Provider data quality is the upstream lever. Partnerships with operators that fix data quality at the source — not vendors who add more AI on top of the existing claim flow — are the only way to compress that cost line.
• The CMS-0057-F PA API requirement (effective January 1, 2027) for Medicaid managed care creates a hard deadline to standardize this layer. The operators who arrive at that deadline FHIR-PAS-fluent and Medicaid-fluent will have a structural advantage in the years that follow.

The bottom line

PHTI’s report is the most important piece of healthcare-AI analysis published in 2026. Its message is uncomfortable for the category of vendors that have raised hundreds of millions of dollars on the promise that AI alone solves revenue cycle. It does not. PHTI now says so, with names attached, on the record.

Rural hospitals and Medicaid populations are paying the highest price for the bot war. Uncompensated care is rising in markets where it can least be absorbed. MCO operations costs are rising on the other side of the same coin.

That is what HRN does. And it is the only model the PHTI report’s findings are compatible with.

If you would like to start that conversation, reach out here. We will respond within one business day.