How Is AI Changing Revenue Cycle Management in 2026?

Healthcare AI in revenue cycle has reached the inflection point where buyers need to distinguish between technologies that work in production and technologies that work in slide decks. The lines are easier to draw than they were a year ago, but only if you know what to look for.

Where is AI actually deployed in revenue cycle today?

The categories that have crossed the production threshold at scale fall into two groups. The first is the older category of structured AI and rule-based automation that has been deployed for years: eligibility verification, claim scrubbing and clearinghouse pre-submission edits, computer-assisted coding (CAC) suggestion engines, and clinical documentation improvement (CDI) tooling. Vendors in this category include Solventum (formerly 3M Health Information Systems), Optum, Microsoft Dragon, Dolbey, Streamline Health, Waystar, Experian Health, R1 RCM, and FinThrive. The maturity is real and the integration is deep; most U.S. hospitals are using one or more of these tools today. What distinguishes these deployments from the newer category is that they augment human work rather than replace it: a coder reviews and accepts CAC suggestions, a denials specialist works the queue that the workflow tool builds.

The second is the newer category of autonomous AI that produces final outputs without per-case human review. The 2025 KLAS report Autonomous Coding 2025 formally recognized autonomous medical coding as a distinct healthcare technology segment for the first time. Current production deployments are concentrated in specific encounter types: outpatient professional coding, radiology, and emergency department coding, where documentation is structured and the code distribution is narrow. Inpatient autonomous coding for complex specialties, prior authorization automation at full scope, and clinical quality abstraction are all reaching production at a smaller number of named vendors with longitudinal audit evidence. The dividing line between the two categories is whether the system's accuracy is high enough that routine per-case human review does not improve the output. Where that line falls determines whether a deployment is autonomous in practice or only in name.

The two categories also differ in how they are bought. Mature production-deployed RCM AI is purchased as a feature within a larger platform: the CAC engine is part of the coding suite, the eligibility tool is part of the patient access suite, the denial workflow is part of the claims management platform. The newer autonomous category is purchased as a discrete service that integrates with the EHR and revenue cycle stack but is not embedded in either. The pattern reflects both vendor strategy and the operational reality that autonomous outputs require independent audit infrastructure (chart-by-chart citation traceability, longitudinal accuracy reporting, specialty-stratified validation) that mature platforms have not yet built into their core RCM modules.

Where is AI still mostly marketing?

The clearest example is agentic AI in revenue cycle. The label has reached peak hype in vendor marketing materials and trade press coverage. The underlying claim is that AI systems can autonomously plan and execute multi-step revenue cycle workflows: pulling charts, querying providers, posting payments, working denials, communicating with payers. KLAS Research's Healthcare AI Update 2025 surveyed more than 3,000 healthcare respondents and found that only 17 specifically mentioned agentic AI, and only one organization reported actually using it in production. The gap between vendor claims and customer reality is, in this category, larger than any other comparable AI category in healthcare. Buyers should treat "agentic" as a marketing label that requires the same level of audit scrutiny as any other accuracy claim.

A second category is the "end-to-end AI revenue cycle platform". Most large RCM vendors now claim some version of this: a single integrated AI-enabled platform that handles eligibility through payment posting with embedded automation at every step. The underlying components are real (CAC, claim scrubbing, denial workflow tooling, prior authorization automation), but the "AI-native end-to-end" framing usually overstates the integration depth and the actual automation rate within each step. Per the Workday/Bain 2026 enterprise AI maturity research summarized in recent TechTarget coverage, roughly 70% of healthcare leaders report early to mid-stage AI maturity rather than embedded enterprise deployment. The platform vs point-solution debate is real, but most "platforms" today are aggregated point solutions with shared infrastructure rather than genuinely unified AI workflows.

Three further categories deserve scrutiny because the marketing materials substantially outpace what is actually deployed. Inpatient autonomous coding for complex specialties (cardiac surgery, neurosurgery, complex multi-procedure inpatient cases) is still maturing; production deployments exist but are narrower than vendor claims suggest, and KLAS specifically documented the current concentration in outpatient/radiology/ED settings. Real-time DRG assignment at full inpatient case complexity has been promised for years; the actual deployments are mostly DRG validation and audit workflows that surface candidate DRGs for human review, which is a different category of work. AI-driven payer negotiation is a frequent vendor claim with no published audit evidence of actual deployment. Generative AI patient experience automation at scale (intake, scheduling, billing inquiries) exists in pilots; the deployment depth across health systems varies enormously and most production deployments are narrower than the marketing implies.

The pattern across these categories is the same: vendor positioning treats AI as a feature label that can be applied to any module, while the buyer evaluation question is whether the system can actually do the work it claims to do at production accuracy on the buyer's actual case mix. The evaluation framework that surfaces the distinction is the same one used for autonomous medical coding: independent credentialed-panel audit, specialty-stratified accuracy, longitudinal monitoring, and methodology transparency. The framework is treated at length in the companion research on evaluating AI medical coding vendors. The same questions apply across every RCM AI category.

What does the next eighteen months look like?

Four developments are predictable enough to anchor an eighteen-month buyer view. The first is the CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F), which is now in active enforcement. As of January 1, 2026, Medicare Advantage, Medicaid, CHIP, and Qualified Health Plan payers must accept electronic prior authorization requests, respond within shortened timeframes (72 hours urgent, 7 calendar days standard), and provide specific denial reasons. The first public PA metrics reporting was due March 31, 2026. By January 1, 2027, all impacted payers must have live FHIR-based Patient Access, Provider Access, Payer-to-Payer, and Prior Authorization APIs. CMS estimates approximately $15 billion in 10-year savings from the move to digital prior authorization. The rule effectively forces standardization across the prior authorization workflow and makes AI-enabled PA automation a structural necessity rather than an optional efficiency play.

The second is the expansion of autonomous medical coding from its current concentration in outpatient, radiology, and ED settings into inpatient and complex specialty coding. The KLAS 2025 segment report on autonomous coding identifies the trajectory: vendors with longitudinal audit evidence in narrower specialty footprints are extending into broader scope, while buyers who have completed initial autonomous coding deployments in their easier specialties are looking for the next phase. The buyer-side discipline that will determine which vendors make the transition is the same audit methodology that distinguishes credible accuracy claims from marketing in the first place: representative sample, panel adjudication against source documentation, specialty stratification, and longitudinal monitoring. Vendors whose accuracy on their easier specialties was earned through audit will scale; vendors whose accuracy was earned through marketing will not.

The third is the arrival of reasoning-based clinical quality abstraction at production scale. The category is roughly two to three years behind autonomous medical coding in market maturity, but the underlying architecture is similar (reasoning against current measure specifications rather than pattern-matching against historical human-abstracted examples, with full source citation for every returned data element). A small number of vendors have published longitudinal audit evidence at production accuracy levels against credentialed human abstractors. The companion research on how automated clinical quality abstraction works and how it reduces the quality reporting burden on hospitals treats the category in operational detail. The CMS eCQM ramp (mandatory submissions set to roughly quadruple between 2024 and 2028) creates the structural pressure that will push the category from emerging production to broader deployment within the eighteen-month horizon.

The fourth is vendor consolidation. KLAS validated 657 vendors with AI solutions in healthcare and noted that over half were mentioned only once by surveyed organizations. The fragmentation is unsustainable from both a buyer evaluation perspective and a vendor unit-economics perspective. The eighteen-month consolidation pattern will likely follow the pattern of prior healthcare technology categories: a small number of established RCM platforms acquire or partner with AI-native specialists (R1 RCM's acquisition of Acclara is a recent example), pure-play AI vendors with strong vertical-specific audit evidence remain independent for longer, and the long tail of single-mention vendors exits the category. Buyers evaluating RCM AI today should weight not just current product capability but the vendor's likely position eighteen to thirty-six months from now: which vendors will be acquiring, which will be acquired, and which will not be operating as standalone businesses. The auditable accuracy framework, the operational depth criteria, and the architectural transparency questions treated in the vendor evaluation research apply directly to that judgment.

About the author

Nathan R. Myers

President & CEO, AccuCode AI

Nathan is the President and CEO of AccuCode AI, a healthcare AI company building autonomous medical coding and clinical quality abstraction systems for U.S. health systems, physician groups, specialty practices, and third-party billing companies. Prior to founding AccuCode AI, Nathan spent twenty-five years building and scaling high-tech manufacturing companies across consumer, professional, and industrial products, shipping hundreds of products into more than forty countries to customers including over twenty Fortune 500 firms such as NASA, BMW North America, and Ford. AccuCode AI grew out of Professional Consulting Services (PCS), an established revenue cycle services company; the lineage gives the production-versus-marketing distinction above its grounding, since the difference between what actually ships and what gets pitched is operational territory rather than analyst speculation.

View full bio →