The Agentic Revenue Cycle Arrives
Clicks joined Y Combinator’s Fall 2025 batch with a pitch centered on using AI agents to automate healthcare back-office work across EHRs, payer portals, faxes and phone lines.
Across healthcare, AI agents are beginning to automate end-to-end revenue-cycle work across fragmented legacy software. Incumbents are launching autonomous platforms, providers and payers are automating against each other, and the labor behind healthcare back offices is shifting from processing volume to supervising exceptions.
The San Francisco company was founded in 2025 by Dominik Helmreich and Oliver Knapp.
Clicks Health now sells computer-use agents for prior authorization, claims, denials and payer follow-up, as well as eligibility and benefits verification, payment posting, patient credit refunds and form filling. Its agents work inside electronic health records, practice management systems, payer portals, spreadsheets and other internal tools. They pull codes and supporting records, clean claims before submission, check status, draft an appeal after rejection, and post payments and reconcile adjustments. Beyond the screen, they call payers for status and authorizations, read incoming faxes and send outgoing faxes.
Clicks does not require providers to replace their EHR or revenue cycle management system. Providers keep their existing system while the agent moves through the steps around it. That gives providers and revenue cycle outsourcers a way to shadow existing staff, compare agent-prepared packets and then limit live use to clean requests while staff see the exceptions.
Clicks Health said:
“Nothing here is complicated on its own. The cost is that every stage asks a person to move the same information between systems that do not talk to each other, hundreds of times a week.”
Unlike an assistant that can only act through a clean interface, a computer-use agent can work through a provider’s existing screens, spot an exception and route the case rather than leave the next step in a staff member’s head.
Deloitte’s February 2026 analysis of agentic AI in health care describes agentic systems as able to plan and sequence tasks, adapt to conditions and coordinate with people and platforms. Its survey found that more than four in five health systems prioritized agentic AI for clinical operations and revenue cycle management, while seven in ten health plans prioritized it for utilization management, prior authorization and claims management.
Administrative denials give agents a large queue to inspect. A February 2026 Massachusetts Health Policy Commission report, based on fully adjudicated claims from 2022 through 2024, found an average overall denial rate of 20.4%. Administrative reasons accounted for 16.6% of all claims in 2024, while clinical reasons such as medical necessity or experimental and investigational denials made up no more than 1% of denials for any insurer.
Those Massachusetts figures do not describe the whole country, but they show why agents have so much to inspect: duplicate claims, missing information, coding errors and insurer-specific procedures.
Prior authorization has its own burden. A 2023 survey of 353 hip and knee surgeons cited in Clicks Health’s materials found that most reported a high administrative burden, with many providers dedicating staff to it. Deloitte reports that MUSC Health has used agents to complete four in 10 prior authorizations without human involvement. In that model, software handles submission mechanics while staff retain responsibility for cases that require clinical judgment or missing documentation.
Autonomy does not mean an agent can guess. Clicks says its prior authorization agent identifies missing information, sends the case to the team, never invents documentation and gives the team a chance to check each packet before submission. Diagnosis, treatment selection and medical necessity remain with qualified clinicians. Deloitte also calls for explicit guardrails, human oversight and workforce preparation. In practice, an agent can execute a repeatable process, but it does not get the final say on every judgment inside that process.
Clicks is hiring across commercial and implementation roles. Zero G Talent’s live board has added one role in the past seven days and currently lists:
For incumbents, the competitive pressure comes from where control sits. Clicks can take on execution across systems a provider already owns. If agents can perform work around a system of record, control of the workflow may shift even when the underlying software remains in place. The system of record may stay; the work around it is moving.
Waystar Fires Back with AltitudeAI
AI agents enter the healthcare revenue cycle
Waystar says its platform processes billions of healthcare payment transactions each year, representing trillions of dollars in gross claims. The company now wants software agents to act on those transactions, not merely report on them. In its August 26, 2026 announcement, Waystar introduced autonomous claim resubmission, moving AI from recommending work to carrying out defined steps with minimal human intervention.
Waystar did not name competitors or frame the launch as a response to startups. Providers are left to compare the systems around which each vendor is building autonomy.
Action replaces advice
Waystar’s January 12, 2026 release established the base for the August expansion. The earlier announcement covered agent-assisted prior authorization, denial prevention, claim correction recommendations and clinical appeals. By January, the announced workflows were designed to support prior authorization, denial prevention, corrections and appeals. By August, the new resubmission capability could carry out those tasks with minimal human intervention.
Waystar said early use of its AltitudeAI denial-prevention and recovery tools had prevented $15.5 billion in claims from denial, achieved 95% time savings and increased denial-overturn rates by double digits. Those results applied to the earlier generation of AltitudeAI tools, not the new autonomous-resubmission agents, and Waystar did not publish an independent benchmark.
“Agentic intelligence is a critical layer for the autonomous revenue cycle,” said Waystar Chief Executive Officer Matt Hawkins. “By combining AI with the breadth of our data, payer intelligence, and connected workflows, Waystar increasingly identifies what needs attention and deploys specialized agents to pursue resolution.”
Autonomous resubmission targets denial rework
Waystar says payers ultimately pay about seven in ten initially denied claims, but providers often spend more staff time and money reworking them before payment. Its autonomous-resubmission agent interprets the payer response, applies payer-specific rules, chooses the next action and refiles an eligible claim. The stated goal is to shorten the path from rejection to correction.
The capability applies to eligible claims and aims for minimal human intervention; Waystar’s clinical-documentation agents separately organize supporting evidence for specialist review.
The announcement leaves several procurement questions unanswered. The August release did not provide an error rate, a full definition of eligible claims or an independent evaluation of the results. Providers still need to know what triggers human review and what happens when the agent’s interpretation conflicts with a payer’s response. The launch sets a direction, not a finished standard for unattended payment work.
Scale is the incumbent’s counterweight
Waystar says its payment network touches roughly three in five U.S. patients and one in three hospital discharges. The company serves over 30,000 clients and more than one million providers, including 16 of the 20 institutions on the U.S. News Best Hospitals list.
The network gives Waystar a base for payer-specific intelligence and a way to learn from prior outcomes across payment workflows. The company says its agents can analyze about 30,000 data points in a medical record within seconds, synthesize the relevant clinical information and assemble evidence for review. Early adopters of its conversational analytics reported up to a 75% reduction in time spent on data analysis. Waystar expects its clinical-documentation agents to cut review time by about one-quarter.
Waystar says its reconciliation solution has reduced reconciliation time by 80%; for one early-adopter health system, it matched $32 million in revenue risk, work equivalent to about 13 full-time employees. The company says results vary by organization and use case.
For health-system operators, routine claim resubmission can leave staff to handle exceptions, evidence review and cases with financial or clinical risk. That shift is more credible than the disappearance of every billing job.
At the moment a payer rejects a claim, the two products illustrate two routes to automation. Waystar embeds autonomy in its payment platform; Clicks packages agents as a service that operates around existing software. Providers adopting these tools will have to decide where the execution should live.
The Payer-Provider Bot Arms Race
In a 2022 report, the Department of Health and Human Services’ Office of Inspector General found that one in eight prior-authorization requests denied by Medicare Advantage plans met Medicare coverage rules and likely would have been approved under original Medicare. The finding did not attribute those denials to artificial intelligence.
The payer advantage
A May 2026 analysis from BAM.ai said payers had historically audited claims by reviewing samples and that payer AI now reviews full claim datasets in real time. Staff capacity limits manual review. Software can apply a payer’s rules across an entire book of claims and surface payment obstacles without waiting for a person to inspect each file.
Health systems brought a different burden to the fight. HFMA reported that insurance companies had invested heavily in automated claims review while many hospitals spent the previous decade implementing electronic health records and remained cautious about AI. The result is an asymmetry: payers can process claims at machine speed, while many providers still assemble the evidence needed to contest them by hand.
Initial denials accounted for a growing share of claim value before provider-side agents arrived.
Kodiak Solutions’ December 2023 revenue-cycle benchmarking analysis, cited by HFMA, reported:
| Denial measure | Reported level |
|---|---|
| Initial denials as a share of claim value | 10.15% in 2020; 11.99% by the end of Q3 2023 |
| Inpatient initial denials | 14.07% through Q3 2023 |
The March 2024 HFMA report framed the rise in the context of payers’ increasing use of AI, but the denial figures do not isolate how much came from algorithms rather than changing payer policy, claim mix or billing practices. They do show that providers were absorbing more denials before providers had comparable tools for answering them.
An American Hospital Association report cited by HFMA found that 35% of hospitals and health systems reported losing at least $50 million because of denied claims.
MD Revenue Group says providers appeal fewer than one in 100 denied claims, even though more than half of first-level appeals succeed when filed. Human teams are leaving potential wins unclaimed, while an appeal agent can work through that neglected queue.
The provider counterattack
Providers are starting to answer machine denials with machine appeals. Clicks, for example, drafts appeals inside providers’ existing systems. Across the market, vendors including MD Revenue Group and BAM.ai describe tools that match a denial’s stated reason against clinical documentation and payer policy, assemble supporting records and track outcomes by payer. BAM.ai also describes formatting responses around the patterns a machine reviewer checks and tracking recovery by denial reason and billing code. The appeal becomes a sequence of checks rather than a generic letter.
The counterattack also moves upstream. Provider-side coding tools review clinical documentation before submission, flag under-coded encounters and identify missing specificity that could give a payer AI reviewer grounds to reject the claim. Coding staff now prepare claims to survive machine review, not just human review.
HFMA reported that Care New England cut authorization-related denials by 55% by using bots to assist with payer notifications when patients were admitted to the hospital, a payer-specific process. The system sent secure notifications within a 24-hour window, and staff reviewed more complex cases that required a phone call.
The escalation loop
Provider deployment remains uneven. BAM.ai’s analysis says many organizations using AI have confined it to one or two functions, leaving the rest of the revenue cycle exposed. Healthcare Dive also reported that payers retain the advantage in claims review, though experts expect providers to narrow the gap.
BAM.ai estimates that only between one in five and two in five healthcare organizations have reached enterprise-wide deployment of revenue-cycle AI.
Smaller practices have the least room to maneuver. They face automated review across their entire claim portfolio but may lack the staff, data and policy expertise required to answer every denial. An appeal agent can increase filing volume, but it cannot compensate for a provider that cannot identify which payer rules changed, which exceptions matter or which appeal outcome should alter the next response.
“There has been a substantial increase in activity on the payer side to create obstacles for payment, and the pain is being felt on the provider side through the ‘triple-D effect’: downgrades, delays and denials,” said Amy Assenmacher, senior vice president of revenue cycle at Corewell Health, in HFMA’s 2024 report.
BAM.ai’s analysis describes the loop: A provider agent adapts to one denial pattern. Payer AI then changes how it identifies that pattern, and the provider must adapt again. Each side can spend more on software and staff while the underlying claim remains unresolved longer.
HFMA reported a forecast that private payers could save:
| Forecast measure | Annual savings within five years |
|---|---|
| Share of total costs | 7%–9% |
| Dollars | $80 billion–$110 billion |
America’s Health Insurance Plans also told the White House Office of Science and Technology Policy that health plans were finding value in faster claims processing, consumer experience and fraud detection through AI. The forecast does not prove that automation’s purpose is denial. It does help explain why investment on both sides of the claims market will continue.
A February 2025 American Medical Association survey found that three in four physicians said prior-authorization denials had increased over the previous five years. Six in ten said they were specifically concerned that AI was driving the increase. That documents physician experience, not proof that algorithms caused the rise.
CNBC reported in May 2025 that UnitedHealthcare said it approves and pays about nine in ten claims upon submission and reaches a 98% approval rate after additional review. Even those rates leave room for denials while showing that most claims pass without one.
Regulators are trying to put a brake on the loop without banning it. The 2018 Office of Inspector General report had already documented widespread and persistent problems with Medicare Advantage denials of care and payment. A November 2023 House oversight letter said CMS’s 2024 rule required medical-necessity determinations to account for each beneficiary’s circumstances and called for plans to understand and disclose the evidence behind their tools. The lawmakers also urged CMS to assess whether algorithms incorporate overturned denials. Oversight cannot reconcile two autonomous systems without a record of what each system decided.
For providers, the next bottleneck will not be generating another appeal letter. It will be producing a defensible audit trail: the payer policy version the agent read, the evidence it attached, the exceptions it escalated and the result returned for the next run. Providers that build that discipline can make the bot war work for them. Providers that do not will simply generate more evidence for the other side’s machine.
Back-Office Jobs Are Being Redrawn, Not Just Deleted
A June 2026 RCM career-market review, based on RCMJobs.com listings, showed that postings requiring AI oversight or automation-management experience had risen by more than a third since the first quarter of 2025. Over the same period, postings for entry-level claims entry, basic follow-up and payment posting fell. Back-office hiring is moving away from processing volume and toward supervising software that processes it.
The entry rung is shrinking
The review found that billing companies using AI employed fewer people in transactional roles such as payment posting and routine eligibility verification. Salaries in those roles remained flat or fell. The work itself has not vanished; it has moved into queues that agents can process continuously and exceptions that trained staff must resolve.
For Clicks, this creates a split inside the billing department: software handles repeatable work, while people focus on cases it cannot confidently finish.
The staffing implications extend beyond hospitals. A 2026 MD Revenue Group analysis says seven in ten hospitals and health systems plan to expand their revenue cycle outsourcing partnerships. Its recommended model keeps experienced employees on complex cases and gives outsourced or virtual teams the high-volume queues. Clicks, which is built for RCM companies and says its agents work inside existing systems, fits that split between software execution and human review. The back office now has tiers, not a simple across-the-board cut.
Judgment moves up the ladder
New job titles are appearing at larger health systems and billing companies, including RCM AI Trainer, Revenue Integrity Analyst and Payer Intelligence Specialist. These positions combine billing expertise with workflow management and often pay above the market median.
The June review said:
“AI is compressing the bottom of the RCM career ladder and expanding the middle and top.”
Denial management, payer contracting and prior authorization oversight are also gaining ground. The work requires context that an agent may not have: whether a generated appeal omitted a clinical detail, whether recurring payer behavior signals a policy change, or whether a claim deserves escalation beyond the standard queue.
The new work centers on judging output from agents such as those offered by Clicks. CPC and CCS certifications remain valuable, while employers increasingly list AI literacy alongside Excel, Power BI and Tableau as hard requirements. That combination is changing who gets interviews and what counts as experienced revenue cycle work.
Experienced coders are not becoming obsolete
The shrinking entry level has not resolved the shortage of experienced medical coders. American Medical Association figures cited in a June 2026 TechTarget report put the national shortfall of skilled certified coders at up to 30%. CarePilot reported that UC Davis Health still could not fill Level 1 coding seats despite offering remote work and top-of-market pay.
Autonomous coding tools have compressed onboarding. The test now is whether a certified coder can trust and correct the output, not whether a model can assign a code.
CarePilot reported that coding leaders reached usable accuracy in 7 to 10 business days with newer tools. The Bureau of Labor Statistics still projects about 14,200 medical-coder openings a year through 2033. AI coding has entered production, but certified coders still review output, investigate anomalies and handle claims outside the model’s comfort zone. Their work is moving from assigning every code to reviewing the difficult cases and the questionable ones.
Supervision becomes part of the job
Reliable deployments start with a narrow service line, parallel testing, review thresholds that reflect confidence and claim value, and more frequent audits during the first stages. The provider remains legally responsible for every submitted claim, whether a person or an AI system produced the code. Providers must document reviews, overrides and audit trails; those controls belong in the operating model.
Agent supervision becomes a distinct job. Supervisors monitor queues, escalate exceptions, record why output was overridden and audit whether automation is producing more speed without adding new denials. Their judgment has to cover both sides of the workflow: the clinical evidence behind a claim and the behavior of the payer processing it.
Employers that automate routine work without redesigning the process will simply produce faster backlogs. Before deploying another agent, write the exception-handling job description, assign its owner and define the audit trail. Automation without named ownership merely relocates delay from routine processing to the exception queue.
Olive’s Collapse and the Clearinghouse Warning
A June 2026 RevCycleAI retrospective put Olive AI’s total capital at $902 million and its 2021 peak valuation at $4 billion, describing it as the most-funded healthcare AI startup at that point. ToolDirectory records that the company shut down on October 31, 2023. RevCycleAI reports that Waystar acquired its revenue cycle management assets, including prior authorization, eligibility and claim automation, while Humata Health acquired its clinical AI assets. The Olive brand and independent company dissolved. Building software that automates a task is easier than supporting that task across hundreds of hospitals at a profit.
Olive first sold a specific product. RevCycleAI’s timeline says the company began in 2012 as CrossChx and built digital biometric identity verification for hospital registration, replacing the “three forms of ID” check-in process. After rebranding as Olive AI in 2019, it promoted a “universal AI operating system” and expanded into revenue cycle management. The June 2026 retrospective described the shipped technology as workflow automation built around robotic process automation, with some machine-learning additions. Narrow use cases could deliver value, but broad “automate any workflow” rollouts stretched implementation from months to years and produced scope creep, contract disputes and churn.
At peak, RevCycleAI estimated an annual burn of roughly $100 million against revenue of $30 million to $60 million.
The model relied on subscriptions and costly implementations by a professional services team. Each new client added implementation work faster than the company could turn the product into a repeatable service.
In fall 2022, Olive laid off about 450 employees, nearly half its workforce, and Sean Lane stepped down as chief executive. RevCycleAI said higher interest rates and compressed technology valuations helped drive the wind-down and characterized the outcome as a managed dissolution rather than a restructuring.
One customer account published by Oyelabs said the promised automated savings never produced measurable ROI, that requests for ROI reports brought vague answers and no hard data, and that the customer ultimately ended its contract.
One account cannot establish a pattern, but it echoes the broader record of delays, disputes and unclear returns. Agent buyers should require measurable outcomes and a tested fallback, not another promise of universal coverage.
The 2024 Change Healthcare cyberattack showed how concentrated the clearinghouse infrastructure had become. A Security.org summary says Change Healthcare processes about 15 billion healthcare transactions a year, handling eligibility checks, prior authorizations, claims submissions and payments. The cyberattack disrupted those transactions and forced many organizations to find other ways to complete them. The episode exposed a single point of failure inside a shared clearinghouse. UnitedHealth Group’s Change Healthcare cyber-response record tracks the operational response.
A 2026 Medisys assessment said healthcare organizations still worried about dependence on single clearinghouse connections. Computer-use agents can work across a payer portal, electronic health record and legacy application without replacing any of them, but they do not give a provider an independent path when a shared clearinghouse fails. Without a tested alternative, automation can generate exceptions faster than people can repair them. That is scale, not resilience.
That distinction matters as agents move into end-to-end work across the revenue cycle. Clicks says it can automate claim submission, denial follow-up and payment posting while routing exceptions to human review. Unlike Olive’s universal layer, this cross-system model can operate around installed software without replacing it. That flexibility can also multiply dependencies: a portal outage or vendor failure can strand a workflow even when each individual component works.
An agent with that authority needs reliability written into the contract, not demonstrated in a polished demo. Buyers should demand:
- Funded runway of at least two years, plus a credible path to profitability that does not assume endless venture capital.
- Three customer references for the exact workflow, not general praise for the vendor.
- Data export and workflow documentation so another operator can take over.
- A precise definition of automation, whether rule-based, machine-learning-assisted or autonomous, and the fallback when it fails.
- Contractual continuity if the vendor is acquired or shuts down.
- A narrow first deployment with measurable impact before the rollout expands.
Those terms address the hard implementation and vendor-shutdown risks documented in the Olive retrospective. They also address what a HealthNext review singled out: multi-year transformation programs lose executive sponsorship before they deliver value. The Olive evidence is more nuanced than the claim that AI failed. Narrow prior authorization and eligibility deployments delivered value, while broader rollouts produced delays and disputes.
The sale of Olive’s assets showed that work can survive a vendor collapse, though continuity varies by product. The transition to Waystar was generally smooth for clients using prior authorization and eligibility automation. Clients who depended on the broader roadmap for products that never shipped suffered disruption. Acquirers could maintain mature workflows, but the universal-platform promise did not carry over intact.
The lesson is to map authority before granting it. No critical step should exist only inside one agent or one vendor, and every handoff needs documented data, a defined owner and a path to human intervention.
Agent adoption can continue. The test for each entrant should move from “Can it complete the workflow?” to “Can we observe it, stop it and take over when a dependency fails?” Before granting an agent authority over claims, revenue cycle leaders should test failure handling: a stopped workflow should stop safely, reach a named person and leave a complete record of what the agent attempted. That also changes back-office supervision. Staff need the authority and training to intervene, not just a dashboard to watch agents run.
The new back-office job is to take control of a stalled claim, explain the failure and preserve its state. Without it, automation is only a faster route to an unclaimed queue.
Working in frontier tech? Zero G Talent tracks the openings: see every open Clicks role, browse frontier tech jobs, the companies hiring, and the people building the field.