For years, the healthcare industry has been promised a "digital utopia" where the friction of the revenue cycle is vanished by the cold, efficient logic of algorithms. The narrative presented by a burgeoning cohort of health-tech startups is seductive: because claims follow rigid rules and payers operate within defined guidelines, the billing process should be fully automatable. If a task is repetitive and rule-based, the logic goes, an AI should handle it.
However, a more nuanced reality is emerging from the front lines of clinical operations. Todd Manion, revenue cycle chair at the Mayo Clinic, offers a pragmatic counter-narrative to the hype of total automation. Speaking at the Healthcare Financial Management Association (HFMA) annual conference in National Harbor, Maryland, Manion argued that while AI is a powerful tool for efficiency, it hits a hard ceiling when it encounters the chaotic, linguistic complexity of clinical medicine.
The Chasm Between Clinical Reality and Billing Logic
The core tension identified by Manion lies in the fundamental disconnect between the way physicians document care and the way payers adjudicate claims. In the eyes of an automated billing system, data must be structured, binary, and perfectly mapped to billing codes (such as ICD-10 or CPT). In the eyes of a physician, documentation is a narrative of clinical decision-making.
"Clinical complexity doesn’t compress neatly into the structured data that automated systems require," Manion explained during his interview at the HFMA conference.
He highlighted a common scenario that plagues medical coders: the "vocabulary gap." A physician might treat a patient for pneumonia, documenting the treatment protocols, imaging results, and clinical observations perfectly. However, if the clinician uses the descriptive term "pulmonary infiltrate" rather than the specific, billable diagnosis of "pneumonia," the billing process stalls.
"Even though payers can see the evidence in the medical record, they can’t really act on it," Manion noted. "Only an explicit, signed diagnosis from a clinician, entered in a specific part of the medical record, can actually appear on a claim."
This creates a high-stakes bottleneck. If the diagnostic data is not in the precise location required by the payer’s algorithms, the claim is either denied or flagged for review. AI, for all its pattern-recognition prowess, often struggles to bridge this gap because it cannot independently "diagnose" a patient based on clinical notes—a task that carries immense legal and ethical liability. Consequently, the revenue cycle still requires a human to query the provider, seek clarification, and ensure the documentation matches the rigid requirements of the claim.
A Chronology of the Automation Evolution
To understand where we are today, it is helpful to look at the trajectory of revenue cycle management (RCM) over the last two decades.
- The Pre-Digital Era (Early 2000s): Revenue cycles were largely paper-based or relied on rudimentary electronic systems. Staff spent hours manually faxing records and calling insurance companies.
- The EHR Mandate (2010s): With the widespread adoption of Electronic Health Records (EHRs), data became digitized, but it remained largely siloed. RCM departments began using "rules engines" to automate basic billing checks.
- The AI Explosion (2020–Present): Startups began pitching "autonomous RCM," claiming they could replace large swaths of the billing department. These tools utilized machine learning to predict denials and automate payer follow-ups.
- The Current Maturity Phase: As seen in the perspective of leaders like Manion, the industry is now moving into a phase of "augmented intelligence" rather than "autonomous automation." Organizations are learning to use AI as a force multiplier for staff, rather than a total replacement for human judgment.
Supporting Data: Where AI Actually Excels
Despite the limitations Manion highlighted, he is far from a technophobe. At the Mayo Clinic, AI is already proving its worth, but it is being deployed in specific, high-leverage areas that do not require clinical interpretation.
The most successful implementations of AI in RCM today involve "administrative drudgery." These include:
- Claim Status Monitoring: Instead of staff members waiting on hold with insurance companies for hours to check the status of a claim, AI agents can query payer portals in the background and update the internal system automatically.
- Remittance Management: AI is highly effective at flagging outstanding remittances and identifying when payments are lingering beyond contractual timelines.
- Denial Prediction: Machine learning models can analyze historical data to identify which claims are likely to be denied before they are even submitted, allowing staff to proactively correct errors.
"I don’t need people waiting on hold to figure out where a claim’s status is with the payer," Manion remarked. "There are simplistic tasks that are repetitive that we’ve used AI to simplify so that we can elevate our people toward more complex patient issues."
By shifting the human labor force away from mundane, status-checking tasks, healthcare organizations can redeploy their highly skilled coders and billing specialists to address the complex claims that require medical expertise and nuance—the very tasks that an AI would almost certainly misinterpret.
Implications for Healthcare Strategy
The implications of this shift are significant for both hospital administrators and health-tech investors.
1. The Death of "Full Automation"
The dream of a "touchless" revenue cycle—one that requires zero human intervention from the point of care to the point of payment—is likely a mirage. Healthcare is fundamentally rooted in human interpretation. As long as clinical language remains descriptive and evolving, there will be a need for human oversight to act as the bridge between clinical intent and financial reimbursement.
2. The Rise of the "Augmented Professional"
The role of the revenue cycle employee is evolving. Instead of being "data entry" specialists, these professionals are becoming "data auditors." Their value is no longer in their ability to type quickly or follow up on status checks, but in their ability to resolve complex documentation disputes and negotiate with payers. This requires a higher level of training and clinical knowledge.
3. A Focus on Accuracy Over Speed
Manion’s philosophy shifts the goalpost of RCM from "chasing claims" to "reflecting care." He argues that if the clinical documentation is accurate, the revenue cycle becomes a byproduct rather than a battle. "If we can do that and do it accurately, everything else falls into place," he said. This suggests that the next generation of RCM technology should focus more on clinical documentation improvement (CDI) at the point of care, rather than billing optimization at the back end.
The Path Forward
The path forward for healthcare finance is not about choosing between human intelligence and artificial intelligence. It is about understanding the boundaries of both.
Artificial intelligence is an exceptional tool for managing the "predictable chaos" of digital data—the endless lists of claim statuses, the patterns in payment delays, and the repetitive checks that keep the wheels of a hospital turning. However, the "unpredictable chaos" of human health—the subtle ways in which doctors describe complex patient conditions—remains the exclusive domain of human expertise.
As the Mayo Clinic and other leading institutions continue to refine their RCM strategies, the focus is clearly moving toward a symbiotic model. By offloading the repetitive, time-consuming tasks to algorithms, healthcare systems can ensure that their most valuable asset—their human staff—is focused on the most critical objective: accurately representing the care provided to the patient.
In the final analysis, the goal of a robust revenue cycle is not just to get paid; it is to ensure that the patient’s clinical journey is documented with such clarity and precision that the payment naturally follows. As Manion’s comments suggest, the technology is catching up, but it is not—and perhaps should not—take the wheel entirely.
