For more than a decade, hospitals have invested heavily in revenue cycle automation with the promise of greater efficiency, fewer denials, and improved financial performance. Yet despite these investments, many health systems continue to struggle with persistent inefficiencies, high denial rates, and rising administrative costs.
The problem is not a lack of technology. It is a mismatch between how these technologies are designed and the realities of hospital billing environments.
According to Brian Sathianathan, CTO and co-founder at Iterate.ai, most current approaches to revenue cycle automation are fundamentally misaligned with the complexity of healthcare data, and that misalignment is why so many initiatives fail to deliver meaningful results.
“The tools in place are built for conditions that don’t actually exist in hospitals,” Sathianathan says. “Most revenue cycle automation assumes clean, structured, predictable inputs, but that’s not what hospital billing environments look like.”
The Reality of Hospital Billing: Messy, Fragmented, and Constantly Changing
Hospital revenue cycle operations are anything but uniform. Data flows in from multiple electronic medical record systems, payer requirements shift frequently, and coding practices can vary widely, even within the same organization.
“You’ve got multiple EMR systems that don’t talk to one another, raw EDI files that are semantically complex, payer rules that shift constantly, and coding practices that vary by facility, by department, sometimes by individual coder,” Sathianathan explains.
Traditional automation tools, particularly rules-based systems, depend on structured and predictable data. When those conditions are not met, the systems struggle.
“Rules-based tools require order, and hospital claims data is inherently messy,” he says. “So you end up with high exception rates, constant manual intervention to keep the automation running at all, and revenue integrity teams spending most of their time on rework instead of the pattern analysis that would actually move the needle.”
The financial implications are significant. Industry estimates from HFMA place the cost of reworking a single denied claim between $25 and over $100, depending on complexity. With denial rates hovering around 15 percent, the cumulative impact is substantial, not only in lost revenue but also in administrative burden.
Hospitals are not just losing money on unpaid claims. They are spending heavily just to recover revenue they have already earned.
Solving the Wrong Problem Layer
Even as healthcare organizations adopt newer technologies such as generative AI, many are still failing to address the root causes of revenue cycle inefficiencies.
Sathianathan argues that both traditional automation and generative AI tools are focused on the wrong layer of the problem.
“Rules-based automation operates on the premise that you can define every scenario in advance and write a rule for it,” he says. “The problem is you’re always chasing the last thing that broke.”
As payer requirements evolve and coding standards change, rules must be constantly updated. The system itself does not adapt; it simply executes predefined logic.
Generative AI, while promising, has largely been applied to downstream tasks.
“Generative AI in this part of healthcare has thus far been genuinely useful, but it has mostly helped with the output side of the problem,” Sathianathan says. “Those are all real contributions, yes, but writing a better appeal letter doesn’t tell you why the claim was underpaid in the first place.”
He points out that generating appeal letters or summaries may improve efficiency, but it does not provide the deeper intelligence needed to identify systemic issues.
“Nor does it surface the pattern of a specific payer consistently reimbursing below contract rates for a particular DRG. Nor can it flag a high-dollar account with a filing deadline 72 hours out,” he adds.
The result is a productivity gain without a corresponding improvement in insight. Revenue cycle teams may work faster, but they are not necessarily working smarter.
A Shift in Architecture: Starting with Raw Data
To address these limitations, Sathianathan emphasizes the importance of architectural change, specifically, working directly with raw claims data rather than relying on preprocessed inputs.
Iterate.ai’s Generate for Healthcare platform is designed to operate at the electronic data interchange (EDI) level, bypassing the need for extensive data normalization.
“Because it’s the primary variable that’s always stopped previous approaches from working well at scale,” he says. “When you require structured inputs, you’re building a preprocessing layer before the actual work starts. That layer is expensive, it takes time, and it becomes a point of failure any time the data upstream changes.”
In contrast, working directly with raw EDI data allows systems to function in real-world conditions.
“Operating at the EDI file level means Generate for Healthcare is working with the data as it actually exists, not as hospitals wish it existed,” Sathianathan explains.
This approach enables the platform to ingest and reconcile data across multiple EMR systems and payer formats simultaneously, without requiring a lengthy integration project.
“That means a health system doesn’t need to complete a months-long data integration project before they can start seeing results,” he says.
The impact can be immediate and measurable.
“One of our early hospital deployments uncovered $17.4 million in unpaid claims within the first engagement,” Sathianathan notes. “That’s money the hospital had already earned, it was just trapped in payer complexity they didn’t have the visibility to resolve.”
Enter Agentic AI: From Execution to Reasoning
At the center of this architectural shift is the emergence of agentic AI, a model that goes beyond automation and content generation to enable autonomous reasoning and action.
“The distinction is where the reasoning happens,” Sathianathan says. “Traditional automation executes steps, while Generative AI produces content. Agentic AI, though, determines what needs to happen and then does it, end to end, without requiring a human to define every step in advance.”
In the context of revenue cycle management, this means AI systems can proactively identify issues rather than simply responding to them.
“Systems can compare actual payments against contracted rates, identify coding errors before claims go out the door, track aging accounts against filing deadlines, and detect underpayment patterns across payer relationships without someone writing a rule that says ‘look for this,’” he explains.
Instead of relying on predefined logic, agentic AI learns from data and continuously adapts to changing conditions.
“It’s learning from the data rather than waiting for the data to conform to predetermined logic,” Sathianathan says. “Agentic AI is a genuinely different capability, not just faster execution of the same workflow.”
Importantly, this approach is not limited to revenue cycle functions. Because it is built as a platform rather than a point solution, the same intelligence layer can be applied across multiple operational domains.
“The same underlying capability that works on claims data can be applied across other hospital workflows like supply chain, staff ops, and clinical documentation,” he says. “The intelligence layer doesn’t have to be rebuilt from scratch every time you move into a new department.”
What Healthcare Leaders Are Getting Wrong
Despite growing interest in AI, many healthcare executives are still evaluating solutions based on surface-level demonstrations rather than real-world performance.
“The most common mistake I see is healthcare system leaders evaluating demos instead of evaluating data readiness requirements,” Sathianathan says.
In controlled environments with clean data, many tools appear effective. But those conditions rarely reflect actual hospital operations.
“The question you have to push on is what happens with your data, in your environment, with your payer mix,” he says. “Can it operate on your raw claims files, or does it require a significant data preparation investment first?”
He recommends that leaders focus on a few key questions when evaluating vendors:
- Can the system work on unstructured claims data, or does it require normalization?
- How long does it take to generate meaningful insights?
- How does it adapt to changes in payer rules or EMR systems?
- How are outcomes measured: by activity or by actual dollars recovered?
“If the answer involves a multi-month implementation before you see results, that’s worth noting—and probably avoiding,” Sathianathan adds.
The Critical Question of Data Security
Beyond performance, data security remains a central concern, particularly when dealing with protected health information (PHI).
Sathianathan is unequivocal in his stance: healthcare organizations should prioritize private, on-premises, or edge-based AI deployments for any workflow involving sensitive data.
“The shared infrastructure model for AI carries real risk that enterprise leaders tend to underestimate until something goes wrong,” he says.
Public cloud environments, while convenient, introduce additional layers of complexity and potential vulnerability.
“Multiple tenants on the same hardware layer, orchestration environments that weren’t designed with memory-bound agentic systems in mind, and a track record that already includes high-profile exposure events at major providers all point to the same underlying vulnerability,” Sathianathan explains.
In healthcare, the stakes are particularly high.
“You’re processing PHI at scale, continuously, and a breach in a revenue cycle AI deployment isn’t just an IT problem, it’s also a HIPAA liability with patient trust consequences that healthcare organizations spend decades building,” he says.
By contrast, private deployments keep data within the organization’s control.
“An on-premises or private edge deployment means the data never leaves a controlled environment, and the model runs where the data lives,” Sathianathan says.
Looking Ahead: Architecture Over Tools
As hospitals continue to navigate financial pressures and operational challenges, the decisions they make today about AI will have long-term consequences.
Sathianathan believes the key distinction lies in how organizations approach those decisions.
“The healthcare organizations that are going to look back and feel like they got ahead of this are the ones making architectural decisions now, not procurement decisions,” he says.
Procurement decisions focus on acquiring tools to improve existing workflows. Architectural decisions, on the other hand, involve rethinking the underlying systems that drive those workflows.
“Architectural decisions are about rethinking what the underlying infrastructure looks like, how data flows, where intelligence lives in the workflow, and whether AI is genuinely changing the economics or just making the existing processes incrementally faster,” Sathianathan explains.
With operating margins under pressure and reimbursement challenges intensifying, incremental improvements may no longer be sufficient.
“The financial pressure hospitals are under right now, with operating margins already very thin and the reimbursement environment getting worse, means incremental improvements aren’t sufficient anymore,” he says.
Instead, organizations must be willing to question foundational assumptions.
“The organizations willing to ask whether the fundamental architecture is right, rather than whether the current tools are optimized, are the ones that are going to be in a genuinely different position in five years.”
A Turning Point for Revenue Integrity
Revenue cycle management has long been viewed as a back-office function, an important but often overlooked aspect in broader digital transformation efforts. That is beginning to change.
As hospitals face increasing financial strain, revenue integrity is becoming a strategic priority. And with that shift comes a need for technologies that do more than automate tasks; they must provide actionable intelligence.
Agentic AI, combined with architectures that operate directly on real-world data, offers a potential path forward.
But realizing that potential will require healthcare leaders to move beyond incremental thinking and embrace a more fundamental transformation.
The question is no longer whether to adopt AI in the revenue cycle. It is whether organizations are willing to rethink the systems that underpin it.
For more information, visit iterate.ai.
Image Source: ID 339434034 | Ai ©
Thai Noipho | Dreamstime.com
Daniel Casciato is a seasoned healthcare writer, publisher, and product reviewer with two decades of experience. He founded Healthcare Business Today to deliver timely insights on healthcare trends, technology, and innovation. His bylines have appeared in outlets such as Cleveland Clinic’s Health Essentials, MedEsthetics Magazine, EMS World, Pittsburgh Business Times, Post-Gazette, Providence Journal, Western PA Healthcare News, and he has written for clients like the American Heart Association, Google Earth, and Southwest Airlines. Through Healthcare Business Today, Daniel continues to inform and inspire professionals across the healthcare landscape.
