Healthcare organizations are losing revenue they have already earned. Not through failed contract negotiations or inadequate reimbursement rates, but through the slow, structural bleed of coding errors, missed charges, and denied claims that accumulates across every department, every payer, and every billing cycle. Revenue leakage healthcare research consistently finds that hospitals and practices lose between 4% and 5% of their total revenue to these preventable failures.

The solution the industry has long relied on, such as larger coding teams, additional audits, and more rigorous documentation training, has proven insufficient against a denial environment that is growing more complex by the quarter. Payer AI systems are approving and denying claims with increasing speed and specificity, and human-only coding workflows cannot adapt at the same rate. The AI medical coder has emerged as the most credible response to that asymmetry: a technology that closes the revenue gaps that scale and complexity have made impossible for human coders alone to close consistently.

What Revenue Leakage Actually Costs Healthcare Organizations

The Numbers Behind Healthcare's Hidden Revenue Problem

The scale of medical coding revenue leakage in the United States is difficult to fully quantify because most of it never surfaces as a discrete loss event. Undercoded claims result in reimbursement that is lower than warranted for the services rendered. Missed charges generate no claim at all. Both disappear into billing reports as accepted revenue rather than as identified losses, making them invisible to the standard financial reporting that revenue cycle leaders review.

What the data does capture is substantial. The American Medical Association estimates that up to 12% of medical claims are submitted with inaccurate codes. Across billing operations, research finds that undercoding occurs in more than 19% of office visit charges and that 14% of hospital services may be coded at the wrong level. For a health system processing tens of thousands of claims per month, those percentages represent a consistent, compounding revenue shortfall that grows larger with each billing cycle it remains unaddressed.

How Hospitals Are Losing Billions in Denied and Uncompensated Claims

Coding-related denials represent the most visible dimension of revenue leakage because they generate an explicit loss event, a claim that was submitted and rejected. Denials and uncompensated care represented more than $48 billion in revenue losses for the 2,300 hospitals analyzed in 2025, a 25% increase from $38.6 billion the prior year. Initial denial rates climbed to nearly 12% across the industry in 2024, and 41% of providers now report that more than 10% of their claims are denied on initial submission.

Coding-related denials have increased by 126% over the past three years, driven in significant part by payer AI systems that identify documentation mismatches, specificity gaps, and modifier errors faster than traditional appeals workflows can address.

The Cost of Undercoding and Missed Charge Capture

Undercoding is often an unintentional conservatism rather than deliberate fraud avoidance. Coders uncertain about the specificity level a note supports will default to a lower code rather than risk an upcoding allegation. That conservatism costs practices revenue they earned and documented but never fully claimed. Missed charge capture, in which billable services performed during an encounter are never coded, compounds the undercoding problem. Complex visits involving multiple procedures, diagnoses, or care coordination activities pose the highest risk of incomplete charge capture, particularly in high-volume settings where coders are working through large queues under time pressure, which discourages the thorough documentation review required for complete charge capture.

The Root Causes of Revenue Leakage in Medical Coding

Documentation Insufficiency as the Primary Coding Failure Point

The single largest driver of coding revenue leakage is documentation that fails to support the specificity required for accurate code assignment. When a physician's note describes a condition in general terms, the coder cannot assign the specific ICD-10 code the payer requires. The claim is either coded with lower specificity or denied upon submission.

Of all Medicaid improper payments in fiscal year 2024, 79.11% were attributable to insufficient documentation. That figure reflects a systemic disconnect: physicians document encounters in the language of clinical reasoning, while payers evaluate claims in the language of coding specificity. The gap between those two languages is precisely where revenue leaks most.

Modifier Errors, Bundling Issues, and Payer-Specific Rule Complexity

Beyond documentation, the technical complexity of modifier assignment and bundling rules generates substantial revenue risk in its own right. Understanding each of them is the foundation for evaluating which AI tools will produce the most meaningful revenue recovery:

• Undercoding and Charge Capture Gaps. When coders assign lower-acuity codes than the clinical documentation supports, or when billable services are not captured in the claim at all, the resulting revenue shortfall is invisible in billing reports and cumulative across every affected encounter. Practices generating $3 million annually may lose $150,000 per year to these gaps alone, with no single claim event flagging the problem.

• Modifier and Bundling Errors. Incorrect or missing CPT modifiers account for 15% to 20% of all claim denials, and bundling violations generate rejections that require appeals processes to recover. Both error types depend on knowledge of payer-specific rules that update frequently and are difficult for human coders to track comprehensively across all contracted payer relationships.

• Documentation Specificity Failures. When physician notes lack the laterality, etiology, acuity, or complication detail that ICD-10 specificity requirements demand, coders cannot assign the most accurate code available. The claim is coded at a lower specificity level, reimbursed below the clinical complexity of the actual encounter, and accepted by the payer in a way that makes the revenue loss undetectable through standard denial tracking.

• Payer Rule Changes and Evolving Coverage Criteria. Commercial and government payers update their coverage criteria, prior authorization requirements, and documentation standards on a rolling basis throughout the year. Revenue cycle teams that rely on periodic training updates rather than continuous rule monitoring create a lag between payer requirements and coding practices, resulting in preventable denials across high-volume claim categories.

These four factors compound each other. A documentation specificity failure that produces an undercoded claim also eliminates the coding precision needed to navigate bundling scrutiny, and an outdated understanding of payer modifier requirements turns an otherwise accurate claim into a denial.

How Coder Shortages Amplify Every Other Risk Factor

The United States faces a significant shortage of qualified medical coders at precisely the moment that coding complexity has reached its highest point. Health systems responding to coder shortages by increasing individual workloads are not solving the accuracy problem. They are accelerating it. Higher volume per coder correlates directly with higher error rates, which in turn translate into denial rates and revenue leakage that organizations are already struggling to contain. Medical AI adoption in the revenue cycle is being driven, in part, by the recognition that human-only scaling is neither financially sustainable nor operationally viable given the current coder supply.

Why Traditional Medical Coding Falls Short at Scale

The Accuracy Gap in Human-Only Coding Workflows

Medical coding accuracy in human-only workflows is constrained by factors that are structural rather than correctable through training or supervision alone. Coders working through high-volume queues make accuracy trade-offs under time pressure. Complex multi-diagnosis encounters require more time than high-volume environments consistently allow. Payer-specific rule variations require institutional knowledge that is difficult to maintain at scale and impossible to update in real time as payer policies change.

Research on human coding accuracy finds error rates ranging from 7% to 12%, depending on specialty and encounter complexity. In emergency medicine, where coding complexity and volume are both high, error rates can exceed those averages. Medical billing AI systems trained on large datasets of clinical documentation and payer adjudication outcomes operate without the time pressure and knowledge maintenance constraints that drive human error rates, which is the foundational accuracy advantage that makes AI medical coding compelling for health systems managing large claim volumes.

How Claim Denial Rates Have Trended as Payer AI Has Advanced

The denial environment that healthcare organizations face in 2025 is fundamentally different from the one that existed three years ago. Payer AI systems now review claims at speeds and with specificity that human adjudicators cannot match, identifying documentation gaps, coding inconsistencies, and bundling violations before claims are approved. Health systems are actively resisting AI-powered payer systems that have driven denial rates to levels straining revenue cycle operations across the industry.

The Capacity Crisis Facing In-House Revenue Cycle Teams

AI in healthcare revenue cycle adoption is accelerating in part because the alternative, scaling human coding teams to match claim volume and payer complexity, is neither financially sustainable nor practically achievable given coder availability. Health systems that attempt to address denial rates by adding headcount find that the economics do not work. Each additional coder adds cost without resolving the systemic accuracy and rule-tracking problems that drive denials in the first place.

How AI Medical Coders Identify and Close Revenue Gaps

AI medical coding systems read clinical documentation - physician notes, operative reports, discharge summaries, and encounter records - and apply trained models to extract the diagnoses, procedures, and clinical specifics needed to assign accurate codes. Unlike rule-based coding systems that apply fixed logic to structured fields, modern AI medical coders use natural language processing to interpret unstructured clinical text, identifying relevant clinical content even when it is embedded in narrative prose rather than structured data fields.

Automated medical coding through AI eliminates the manual extraction step that drives both time costs and error rates in human coding workflows. The AI reads the note, identifies codeable elements, applies current payer rules, and generates a complete claim with codes, modifiers, and linkages - typically in seconds. Physician review and sign-off remain part of the workflow, preserving clinical oversight without requiring the physician to construct the coding work product manually.

ICD-10 coding AI systems are trained on the full specificity structure of the ICD-10-CM and ICD-10-PCS code sets, including laterality, etiology, acuity, manifestation, and combination code requirements that human coders must look up and verify individually for each encounter. An AI coder assigns the maximum specificity the documentation supports rather than defaulting to a lower code when the specificity is uncertain.

Real-Time Charge Capture and the Elimination of Missed Revenue

Healthcare revenue optimization through AI extends beyond code accuracy to charge capture completeness. AI medical coding systems identify billable services documented in the clinical note that were not included in the original charge entry, flagging missed charges for review before the claim is submitted. That capability addresses the charge capture gap that human coders miss, not because of inaccuracy but because of the volume and speed constraints that govern manual review workflows in high-throughput clinical settings.

Sully's AI Medical Coder performs this complete charge capture function alongside code assignment, reading the clinical documentation generated during the encounter, and identifying the full set of billable services before the claim leaves the practice. Closing the revenue gap at the point of documentation rather than after a denial has already occurred.

Selecting and Implementing an AI Medical Coding Solution

Selecting an AI medical technology platform for revenue cycle coding requires evaluating vendors against the dimensions that determine whether the solution will close revenue gaps or introduce new compliance and operational risks. The following criteria reflect what revenue cycle leaders and CFOs consistently identify as decisive in AI for medical coding vendor assessments:

• Coding Accuracy Benchmarks by Specialty. AI medical coding accuracy varies considerably across specialties, with tools trained primarily on primary care encounters often performing less accurately in orthopedics, oncology, or emergency medicine. Request validated accuracy benchmarks for the specific specialties your organization bills, and confirm those benchmarks reflect performance on real patient encounters rather than controlled test sets provided by the vendor.

• EHR Integration and Workflow Compatibility. An AI coding solution that requires manual documentation transfer from the EHR adds steps, eliminating efficiency gains and introducing transcription risk. Evaluate whether the platform reads documentation directly from your EHR, returns coded claims to the correct billing workflow, and supports the specific EHR version and configuration your organization uses. Sully's EHR integration infrastructure supports more than 50 platforms, ensuring coding recommendations flow directly from clinical documentation into the revenue cycle without manual handoffs.

• Compliance and Audit Trail Capabilities. AI-generated coding recommendations must be fully auditable. Confirm that the platform generates a complete audit trail for every coding decision, documents the clinical rationale for each code assigned, and supports the oversight requirements of your compliance program and payer contracts. Vendors that cannot demonstrate comprehensive audit trail functionality should be disqualified from further evaluation.

• Denial Tracking and Performance Reporting. The most effective AI medical coding platforms surface denial patterns, accuracy trends, and charge capture rates in reporting dashboards, enabling revenue cycle leaders to identify and address performance gaps before they compound. Confirm that the vendor's reporting provides the granularity your organization needs to manage coding performance at the payer, specialty, and encounter-type level.

Organizations that evaluate AI coding solutions against all four criteria are better positioned to choose a platform that closes revenue gaps rather than creating new compliance or operational risks.

Healthcare revenue optimization through AI-driven medical coding represents the most credible response for health systems seeking to close gaps caused by coding errors, missed charges, and preventable denials. The organizations that deploy these tools with appropriate human oversight, deep EHR integration, and rigorous performance measurement are building the revenue cycle infrastructure that the next decade of payer complexity will require. Sully's AI Medical Coder is designed to be the foundation of that infrastructure, combining clinical documentation accuracy with coding precision to capture the full revenue value of every encounter from the moment the physician completes the visit.

Sources

• American Hospital Association. (2024). 3 ways AI can improve revenue cycle management. Aha3 Ways AI Can Improve Revenue-Cycle Management | AHA

• Experian Health. (2025). State of claims 2025: The denial problem and is AI the answer? ExperianState of Claims 2025: The denial problem (and is AI the answer?) - Healthcare Blog

• Healthcare Finance News. (2025). Hospitals' net revenue leakage increases 25% due to denied claims. HealthcarefinancenewsHospitals' net revenue leakage increases 25% due to denied claims

• Healthcare Financial Management Association. (2025). Health systems start to fight back against AI-powered robots, driving denial rates higher. HfmaBattle of the bots: As payers use AI to drive denials higher, providers fight back

• Joudaki, H., Rashidian, A., Minaei-Bidgoli, B., Mahmoodi, M., Geraili, B., Nasiri, M., & Arab, M. (2015). Using data mining to detect health care fraud and abuse: A review of literature. Global Journal of Health Science, 7(1), 194-202. DoiUsing Data Mining to Detect Health Care Fraud and Abuse: A Review of Literature

• Medical Economics. (2025). The hidden revenue crisis in health care: Why billing accuracy remains an illusion. Medicaleconomicsmedicaleconomics.com/view/the-hidden-revenue-crisis-in-health-care-why-billing-accuracy-remains-an-illusion

• Ndegwa, S. (2024). AI in healthcare RCM: What to expect in 2025. Nym Health. NymAI in Healthcare RCM: What to Expect in 2025

• Tseng, P., Kaplan, R. S., Richman, B. D., Shah, M. A., & Schulman, K. A. (2018). Administrative costs associated with physician billing and insurance-related activities at an academic health care system. JAMA, 319(7), 691-697. Doidoi.org/10.1001/jama.2017.19148