Revenue cycle management is the operating system of healthcare. It runs end-to-end through every part of the patient encounter without owning any single piece of it — which makes AI the upgrade healthcare has been waiting for.

That's not a knock on the many digital innovations that have transformed the field in the years following HITECH. EHR adoption, clearinghouse automation, eligibility APIs, patient portals — these technologies matter, and revenue cycle wouldn't function at modern scale without them. But they're largely rules-based systems in an environment that doesn't follow stable rules.

Payer policies change constantly. Coding guidelines shift quarterly. Patient financial circumstances evolve in real time, driven by external factors like job loss, deductible resets, and new diagnoses. And the conversations themselves are often emotionally charged in ways that no static tool can anticipate, whether it's a patient questioning a $4,000 balance or a billing rep navigating their twelfth denial appeal of the day.

Where traditional automation breaks down, AI shines. Not by replacing the systems providers have built, but by adding a layer of adaptive intelligence that can handle the messy, dynamic, human parts of the revenue cycle that everything else has to work around.

‍

How AI is reshaping the revenue cycle

So how can AI improve revenue cycle management in healthcare, specifically? The answer isn’t one big technology shift. It’s three smaller ones converging at once:

1. Large language models (LLMs) and natural language processing (NLP) can now process clinical documentation and large datasets at scale. That unblocks coding, billing, and prior authorization workflows that used to require human judgment to interpret a chart, match it to a policy, or assign the right code.
   
   
2. Predictive analytics trained on claim and patient records can score risk at the transaction or account level. Every claim, patient balance, and authorization gets evaluated against historical outcomes, transforming downstream workflows like denial management and patient billing from reactive to preventive.
   
   
3. Conversational AI has crossed the quality threshold where patients can get answers to their questions and resolve bills without the immediate urge to escalate to a human. That completely changes the economics of the call center, the largest cost center in most revenue cycle operations.

But the real unlock is orchestrating multiple agents across multi-step workflows, with humans stepping in at the points where judgment actually matters. That orchestration layer — not any single model — is where the real operating leverage lives.

‍

Five places AI is already at work in revenue cycle

Across five applications, AI is targeting the workflows that account for the vast majority of revenue cycle's labor cost: coding, denials, call centers, prior auth, and patient billing. That's not a coincidence. Those are also the workflows where the rules-based systems of the last decade hit their ceiling.
‍

Generative AI for autonomous coding and charge capture

Coding has been on a long evolutionary arc, and generative AI is accelerating it dramatically. Each generation of coding technology has pushed humans further from the keystroke and closer to the exception cases.

The first generation was computer-assisted coding (CAC) — NLP-driven systems that suggest codes for human coders to validate. Useful, but the labor model didn’t really change. The second generation, autonomous coding, uses LLMs to read the full encounter and assign ICD-10, CPT, and HCC codes, routing only edge cases to humans. In high-volume specialties like radiology and emergency medicine, autonomous coding can handle the vast majority of charts without human review, at accuracy levels that match or exceed experienced coders.

The newest generation pushes further upstream: ambient AI captures coding signals at the point of care, listening in on the clinician-patient encounter and grounding every code in the dialogue itself. HCC capture improves because diagnoses get documented during the visit rather than reconstructed afterward, audit defensibility improves because every code traces back to a specific moment in the conversation, and the back-and-forth between coders and providers is greatly reduced.

What ties all this together: AI isn’t eliminating coding roles, it’s redirecting human talent toward higher-judgment work. That's the real operating model shift, and it's why generative AI for coding has become the entry point for most health systems' broader revenue cycle AI strategy.
‍

‍Predictive analytics for denial risk and prevention

Claim denials remain the single largest source of preventable revenue leakage in healthcare, and they’re getting worse. Increasingly, payers are deploying AI to fight claims, and hospitals spent nearly $18 billion on overturning denials in 2025 alone.

Predictive analytics give providers a fighting chance. Models trained on large historical claims data — segmented by payer, service line, code combination, authorization status, and more — score every claim before submission and surface the ones most likely to be denied, along with the specific reason codes driving the prediction.

Instead of working denials after the fact at an average cost of $57.23 per claim (up from $43.84 the year prior), billing teams fix the upstream issue before the claim leaves the door. The business case is real but adoption is lagging. Only 14% of providers currently use AI to reduce denials, even though 69% of those who do report fewer denials and more successful resubmissions.

The AI arms race between payers and providers will only intensify, and provider-side predictive analytics is now a necessary defense — but it's not entirely within providers' control. That makes the patient, the fastest-growing, most unpredictable payer, the segment with the greatest financial upside for providers, and the place where AI can most directly influence outcomes.
‍

Conversational AI for patient billing and collections

The call center has long been the revenue cycle’s most expensive and least scalable function. Conversational AI, which can be deployed across voice and chat channels, is rewriting that equation.

On the patient side, AI voice agents can now handle routine and high-volume billing inquiries, including balance questions, insurance updates, and why a patient received multiple bills for one encounter. Patients can speak naturally about their issues and get answers faster, with the AI passing along full context to a human agent when needed. For providers, that means containment rates that meaningfully reduce average handle time and scale support without increasing cost to collect.

Inbound customer service is only one use case. Next-generation systems use voice AI for outbound collection calls, extending the impact of in-house self-pay teams and early-out vendors. Unlike robocalls of the past, these conversational agents act as a patient concierge that can proactively contact patients to confirm insurance, establish extended payment plans, and resolve affordability issues — all without the patient feeling like they're being hounded.

Done well, conversational AI does two things at once: it lowers cost to collect, and it improves yield from harder-to-reach patient cohorts where every basis point of collection performance now matters more.

‍

AI-driven prior authorization and clinical evidence matching

Prior authorization might be the most universally despised workflow in healthcare. The American Medical Association estimates that prior auth has caused a serious adverse event for a patient in nearly 1 in 4 physicians' practices, and 60% of medical groups report needing at least three employees to complete a single PA request. 

The underlying problem is structural. Every payer has its own medical policies, those policies change often, and matching the right clinical evidence to the right policy in the right format has historically required dedicated staff combing through charts and faxing PDFs.

AI inverts the workflow. Models read the EHR, extract the relevant clinical evidence, match it against the specific payer's current medical policy, and assemble the authorization request — or appeal letter — in the format that payer expects. 

The impact shows up in three places: faster time-to-treatment for patients, lower abandonment rates for scheduled procedures, and meaningful FTE redeployment in utilization management. And the timing is forcing the issue — the CMS Interoperability and Prior Authorization Final Rule requires payers to expose electronic PA APIs starting in plan year 2028, which will turn what is still partly a screen-scraping exercise into a true machine-to-machine workflow.
‍

Patient payment forecasting and personalized financial journeys

Many revenue cycle teams still segment patient accounts the same way they did a decade ago: commercial, government, self-pay, with a propensity-to-pay score layered on top to guide collection activity. The problem is that those signals were built for a financially stable population with stable coverage. These patients are increasingly the minority.

Cedar's 2026 Healthcare Financial Experience Study shows just how broken the existing signals have become. About three-quarters of patient out-of-pocket dollars now sit in difficult-to-collect cohorts. Thirty percent of patients say the payment options in front of them are unaffordable, and four in ten of them earn $100,000 or more. And in Cedar's analysis of 10 million bills with third-party propensity-to-pay scores, 54% over $50,000 were scored "high propensity" — even though those are objectively the balances most at risk of going unpaid. 

AI changes the inputs. Instead of relying on a static snapshot of historical payment data and public records, modern patient financial experience platforms ingest signals such as visit complexity, digital literacy, and past engagement. The output isn’t a single score; it’s a personalized billing journey. Patients get matched to the right options, through the right channel, at the right time, for improved collection and experience outcomes.

We built this interactive tool to help revenue cycle leaders visualize the journey — adjust patient attributes and see how AI adapts the experience for different scenarios.

‍

What this looks like in practice: three Cedar partners

The five applications above aren't slide-deck promises. Cedar's partners are putting them to work today — across very different patient populations, operational setups, and stages of revenue cycle management maturity. Three examples of what that looks like in practice:

Scaling patient billing support at Gastro Health

As one of the country’s largest gastroenterology practices expanded to more than 120 locations across seven states, Gastro Health found its phone lines becoming a constraint on growth and margin. “You have to be able to scale without increasing your cost to collect,” says Leo Vela, Director of Patient Financial Services at Gastro Health.

Gastro Health deployed Cedar’s AI voice agent, Kora, to autonomously answer billing questions — including in Spanish, critical for the practice's South Florida footprint. “It’s very much like speaking to a person,” Vela says. “A patient will interrupt and Kora stops its conversation to hear the patient.” Beyond conversational ability, Kora continues to learn the nuances of Gastro Health’s business, from specialty-specific details to how the practice operates. 

But the real surprise for Gastro Health wasn't the technology; it was how quickly they saw results. In nine months, live agent handle time has dropped 24% and Cedar-supported call center staffing has been reduced by 22%. Read the full Gastro Health case study →

‍
‍

Personalizing self-pay discounts at ApolloMD

ApolloMD, a leading clinician services provider specializing in emergency medicine, saw a 42% lift in patient collections in the first year of its Cedar patient financial experience partnership. But the more interesting story is what they’ve done since -- continuously adding new AI capabilities to optimize collections and reduce patient friction.

‍

One of the highest-impact additions: a machine learning (ML) model that ApolloMD uses to personalize self-pay discounts. For patients with balances between $400 and $5,000 — where affordability is typically the main barrier to payment — Cedar's ML model identifies the optimal discount to convert an otherwise unpaid balance into payment. 

The discount model drove a 9% lift in collections among the targeted self-pay cohort, contributing to four straight years of cash growth and a record $48.2 million collected in 2024. Read the full ApolloMD case study →

Optimizing patient outreach at U.S. Anesthesia Partners

U.S. Anesthesia Partners (USAP) came to Cedar in 2018 already operating with sophisticated internal analytics across its revenue cycle. The goal was to find a partner that could take their strong patient billing foundation to the next level. As Trish Donohue, Vice President of Shared Services at USAP recalls: “We talked to more than 25 vendors. Cedar was one of the few thinking about ML and AI, and how it could actually improve the patient journey.”

That approach now shapes how they communicate with patients. ML models help USAP determine the right outreach strategy for different populations, optimizing what “right” means over time. “Other companies either over-communicate, overwhelming patients, or under-communicate, leaving them confused,” Donohue says. “Cedar has found that sweet spot.”

Eight years into the partnership, USAP collects at a rate more than 20% above the industry average, with 90% patient satisfaction. Read the full USAP case study →

‍

The future of revenue cycle management is AI + human

Every previous wave of healthcare technology asked providers to bend their workflows around the tool. EHRs are the most familiar example — providers spent years redesigning clinical and financial operations to fit the constraints of the system, often at significant cost to productivity, morale, and patient experience.

AI is different. Partly because it's more flexible than rules-based systems. Agents can adapt to existing workflows rather than forcing teams to adapt to them. But more importantly, its arrival creates a moment to step back and ask a harder question: how should this work actually get done? For the first time in decades, revenue cycle teams have both the permission and the tools to redesign operations from the ground up — around the outcomes they want, not the constraints of the systems they inherited.

Most providers will spend the next two years evaluating AI tools, running pilots, and adding to their vendor stack. That's necessary work. But it's not enough. The bigger opportunity is to rethink workflows around what AI does well, redeploy staff toward the work that requires people, and build the governance to keep models honest as payer rules and patient circumstances continue to shift.

‍