AI Native RCM: Why the Hype Outpaces the Reality (And How to Compete and Win)

The fear is real. Well-funded AI native RCM companies are flooding the revenue cycle management and claims processing space, and private equity groups are watching nervously. With companies like Athelas raising $400 million and Angle Health pulling in a jaw-dropping $750 million, it’s easy to feel like resistance is futile — to borrow a phrase from Star Trek’s Borg.

But is it really? Should RCM companies, TPAs, and healthcare claims administrators abandon their strategies and scramble to become an AI native RCM operation overnight? The short answer: no. The longer answer is what this post is about.

Across three deep-dive episodes, we’ve unpacked the AI native RCM myth — what it actually means, why it’s considered superior, where it repeatedly fails in practice, and what a smarter, more sustainable strategy looks like. This post synthesizes all three parts into one comprehensive guide.

What Does “AI Native RCM” Actually Mean?

Before evaluating whether AI native RCM is a threat or an overhyped buzzword, it’s important to define the term clearly.

AI native RCM refers to a software platform built from the ground up with the explicit intent of being fully AI-enabled — not retrofitted onto existing operations, not patched together from legacy systems, but designed from day one to be run entirely or predominantly by artificial intelligence.

In the RCM and claims processing context, an AI native RCM company typically looks like this:

• A venture-backed startup with no existing clients
• A team of engineers building a software platform first
• The pitch: zero (or minimal) human involvement in claims processing, billing, eligibility verification, and adjudication
• The goal: outcompete traditional technology-enabled service companies by being faster, cheaper, and more scalable

On paper, this sounds compelling. Ask any AI search engine whether AI native is better, and it will almost certainly say yes. (The irony of an AI advocating for AI is not lost on us.)

But the gap between the investor pitch and operational reality is wide — and it’s widening.

The 100% AI Myth

Here’s what the pitch decks don’t say clearly enough: virtually no AI native RCM company is actually 100% AI.

Exception handling alone almost always requires human judgment. Beyond that, real-world AI native operations typically involve:

• Significant human management and oversight
• Human-led problem-solving and quality control
• Manual diagnosis and process redesign
• Human-driven implementations and client onboarding

Anyone claiming their platform is 100% AI-operated is either overstating the truth or running a system riddled with problems. This isn’t a criticism — it’s the current state of the technology. The question is whether companies acknowledge this reality or hide behind the marketing narrative.

Why AI Native RCM Is Harder Than It Looks

The appeal of AI native RCM is understandable, especially to Silicon Valley-style engineering teams who look at healthcare administration and see an industry ripe for disruption. The logic goes: healthcare billing and claims processing are inefficient and backwards — let’s engineer our way out of the problem.

This reductionist view consistently underestimates the complexity involved.

The Operational Complexity Problem

Revenue cycle management and insurance claims administration — whether you’re an RCM company billing for providers or a TPA processing employer health claims — are not purely mathematical problems. They involve:

• Massive numbers of edge cases and exception scenarios
• Irrational human behavior that defies clean algorithmic logic
• Institutional knowledge that lives in people’s heads, not in documentation
• Constantly shifting payer rules, regulatory requirements, and contract terms
• Real-time coordination across multiple stakeholders

Pure technology teams building an AI native RCM platform without ever having run a medical billing operation or a TPA tend to learn this lesson expensively. Building in a vacuum is very different from running operations where every claim matters and every denial has a downstream financial consequence.

According to a 2024 McKinsey-cited report published by the American Hospital Association, AI holds tremendous promise in RCM — but realizing that promise requires deeply understanding existing workflows before replacing them. Technology that ignores operational nuance tends to break in the field, and AI native RCM is no exception. For a practical view of where generative tools fit, see our overview of AI in medical billing.

The “Dog Fooding” Distinction

There’s an important strategic difference between:

1. Building AI for yourself — you’re a TPA or RCM company that builds AI tools to improve your own operations, then sells services in the marketplace
2. Building AI to sell — you build a software platform and try to sell it into an RCM or TPA market

The first model is far more forgiving. When you’re your own first customer, real-world operational feedback is immediate and continuous. The software must work or the business fails — and that tension produces better, more grounded technology.

The second model, which is how most AI native RCM startups operate, requires building something that works across dozens or hundreds of different client environments, workflows, and edge cases — before you’ve fully encountered any of them. That requires exponentially more development time, customization capability, and cost.

The “Who Designs the Process?” Problem

When an existing TPA or RCM company builds technology, they have an enormous advantage: the entire operational team is already in place. They know what needs to happen. They can map processes, identify exceptions, write SOPs, and provide continuous feedback to engineering.

When you start from scratch as an AI native RCM company with no employees, no clients, and no operational history — who designs the process? Do you hire hundreds of operations specialists years before you need them, before any software is built and before any clients exist? The math and the logic simply don’t hold up.

The Real Problems with a Monolithic AI Native RCM Platform

Beyond the foundational challenges of building AI native RCM from scratch, the architecture itself carries significant operational risks — for the companies building these platforms and the clients relying on them.

Inflexibility and the Modularity Advantage

Technology evolves rapidly. A competitor may build a better claims scrubbing tool tomorrow. A new vendor may offer superior eligibility verification at a lower cost. If you’re locked into one monolithic AI native RCM platform, you cannot easily swap out components.

A modular approach — where AI and automation are layered onto specific workflows iteratively — offers a decisive competitive advantage over a monolithic AI native RCM build:

Factor	AI Native Monolithic	Modular/Iterative Approach
Implementation risk	High	Low
Technology flexibility	Low	High
Time to first go-live	Long	Short
Failure impact	Entire system	Isolated module
Cost of change	Very high	Manageable
Client switching cost	High (pro and con)	Moderate
Scalability	Difficult	Strong

Higher Risk for Clients and Operators

The implementation complexity of an all-in-one AI native platform is significantly greater than a modular rollout. For companies selling this software into the marketplace — say, competing against TPA administration platforms — the sales hurdle is enormous. Clients are asked to trust a system they can’t partially adopt or gradually test.

For operators running these platforms internally, if one part of the system fails, everything can fail. There is no fallback module. There is no isolated component to roll back. The risk profile is fundamentally different — and not in a good way.

A modular, iterative approach solves this. Teams can:

• Go live with individual modules as they’re ready
• Run modules in parallel during transitions
• Roll to a functioning module if another is delayed
• Continuously improve specific components without systemic disruption

AI’s Blind Spots Are a Real Operational Risk

One underappreciated challenge in AI native RCM is AI’s troubling tendency to be confidently wrong. AI systems are frequently blind to their own errors, can misalign with intended goals, and in more sophisticated implementations, have even been shown to conceal problems from oversight layers.

This is not science fiction alarmism — it’s a practical, day-to-day operational concern. In claims processing, a confident but wrong AI making consistent errors across thousands of claims represents catastrophic financial and compliance risk.

This is one more reason why a modular approach — which confines AI to specific, well-monitored workflows — is meaningfully safer than a system where AI is responsible for end-to-end claim adjudication with minimal human checkpoints.

For a deeper look at these risks in the healthcare AI context, see our analysis of the hidden risks of AI in RCM, which apply directly to any AI native RCM rollout.

What True AI Native RCM Would Require: Multi-Layered Metacognitive Architecture

For an AI native RCM platform to actually work as advertised — meaning without significant ongoing human intervention — it would need something that currently does not exist in any known healthcare claims processing company: a multi-layered metacognitive architecture.

This is a complex concept, but broken down it means a system that can:

Layer 1 — Worker Layer

Individual agents and modules executing specific tasks. In an RCM or TPA context, this includes processing claims, routing to appropriate departments, triggering eligibility checks, applying payer-specific rules, and generating remittance. These are the building blocks behind today’s medical billing AI agents.

Layer 2 — Evaluator Layer

A system using machine learning and large language models (LLMs) to continuously monitor the worker layer — identifying anomalies, potential errors, hallucinations, and logical inconsistencies in real time.

Layer 3 — Meta/Diagnostic Layer

An overarching system that pulls data from the evaluator layer, identifies systemic failures, develops new processes and SOPs, dynamically updates workflows, and self-corrects without human initiation.

This architecture would represent a genuinely self-diagnosing and self-improving claims processing system. It is theoretically achievable. But as of today, no known healthcare claims processing company — whether on the RCM side or the TPA/insurance administration side — has built this. Not Athelas, not Angle Health, not a major health plan.

The absence of this architecture is precisely why so many AI native RCM companies quietly reverse course and add humans back into the process.

Case Study 1: Athelas — $400 Million and a Pivot Away from RCM

Athelas is one of the most cited examples of an AI native RCM company, and its trajectory offers important lessons.

The pitch: A fully AI-enabled revenue cycle management company — no humans required, built from scratch, and ready to disrupt the entire medical billing industry.

The funding: Approximately $132 million raised in 2022, followed by additional rounds totaling roughly $400 million, backed by top-tier investors including Sequoia Capital and General Catalyst.

The reality:

• The all-AI model failed to hold up against real-world RCM complexity
• Athelas acquired an offshore billing company with approximately 400 employees based in Delhi, India — directly contradicting the “no humans” premise
• The company ultimately pivoted away from revenue cycle management entirely, shifting its focus to AI-powered medical scribing and EHR systems
• Sequoia notably did not participate in later funding rounds, signaling loss of confidence

The lesson here is not that Athelas failed as a company — pivoting is a legitimate business strategy. The lesson is that with $400 million in capital, world-class engineers, and elite investors, the AI native RCM vision still could not be executed as described. That should tell you something important about the model itself.

Case Study 2: Angle Health — $750 Million and a “Hybrid” Reality

Angle Health is the more current and arguably more instructive case study of the AI native RCM model for TPAs and self-funded employer health plan administrators.

Background: Angle Health (spelled A-N-G-L-E) is a third-party administrator focused on the self-funded employer health insurance space. It was founded by engineers from Palantir — a company known for elite data capabilities, mission-driven culture, and notably: messianic arrogance, brittle custom logic, and staggering implementation costs.

The funding: Approximately $750 million raised, including a single round of $600 million — a genuinely staggering sum even by venture capital standards.

The original pitch: Pure AI and machine learning infrastructure to power insurance claims administration — no humans, pure data-driven automation.

The reality:

• The all-AI approach proved unworkable in TPA operations
• Angle acknowledged a shift to a “hybrid model” — which is a more polished way of saying they added substantial human operations
• Today, the platform functions more as an AI acceleration layer on top of human-operated processes than a truly autonomous claims administration system
• Their claimed advantage — the ability to serve employer groups as small as two employees — is real but targets a market segment most traditional TPAs actively avoid due to margin and complexity concerns

The Palantir DNA Problem

The Palantir engineering culture that Angle Health inherited brings specific structural challenges that map directly onto the problems we’ve discussed:

• Brittle custom logic — highly specific rule-based systems that don’t generalize well across diverse client environments
• Non-modular architecture — deep integration that makes components difficult to replace or upgrade
• Extreme implementation costs — which partially explains why you need three quarters of a billion dollars to operate

Larger plan sponsors with complex benefit designs, stop-loss arrangements, and meaningful employee populations may find that Angle’s platform lacks the operational flexibility that a service-first TPA with layered technology can provide.

For a broader look at how AI ROI actually works in practice versus the pitch, see How to Get Real RCM AI ROI: Beyond the Hype.

Why AI Native RCM Companies Keep Getting Funded Despite These Challenges

If the AI native RCM model is so flawed, why do companies like Athelas and Angle Health keep raising hundreds of millions of dollars? Several forces are at work.

The VC Portfolio Model

Venture capital doesn’t require every investment to succeed. The standard model is:

• Make 10 investments
• Expect a few to fail entirely
• Expect several to return invested capital approximately
• Expect one or two to return 20X or more, carrying the entire portfolio

This philosophy incentivizes swinging for the fences — low probability, high reward bets. A modular, iterative AI strategy is fundamentally incompatible with this model. You don’t raise $100 million to go modular. You raise $100 million to build a unified platform that could theoretically dominate a market.

The AI Bubble Dynamic

The current investment environment has significant parallels to the late 1990s dot-com bubble. Enormous capital is flowing into AI companies — native or otherwise — with valuations that are difficult to justify on current fundamentals. OpenAI and Anthropic have both delayed IPOs, suggesting investor appetite for public market AI exposure continues to grow. This environment creates market distortions where well-funded competitors can undercut on price, over-invest in client acquisition, and operate at a loss for years.

This will not last indefinitely. Investment requirements for return will tighten. And historically, AI native RCM companies in complex operational service industries have not demonstrated strong long-term performance. For a wider view of where the industry is heading, see our perspective on the future of medical billing.

The Better Path: Modular, Iterative AI Deployment in RCM and TPA Operations

The alternative to the all-or-nothing AI native RCM approach is not “do nothing.” It is a deliberate, modular, and iterative strategy for deploying AI and automation across your revenue cycle or claims processing operations.

What This Looks Like in Practice

Step 1 — Centralize your data. Regardless of your technology approach, centralized data warehousing is non-negotiable. Every AI and automation module you build should connect to the same centralized data infrastructure. This eliminates data silos without requiring a monolithic platform, and it gives your RCM analytics a single source of truth to work from.

Step 2 — Identify high-value, high-volume workflows. Start with the processes where automation can handle the most volume with the least exception risk. In RCM, this might be eligibility and benefits verification, claim status checks, or payment posting. In TPA operations, it might be member onboarding or standard claims adjudication for clean claims.

Step 3 — Deploy, measure, and iterate. Go live with individual modules. Measure impact. Optimize. Then expand. A workflow that starts at 60% AI-handled can grow to 70%, then 80%, as the technology matures and your team builds confidence.

Step 4 — Run parallel operations during transitions. Rather than hard cutover to a new module, run both the legacy process and the new module in parallel. This dramatically reduces risk and gives you real-world data before full deployment.

Step 5 — Build metacognitive capability incrementally. You don’t need a full multi-layered metacognitive architecture on day one. Start by building evaluation layers into individual modules — monitor for errors, log anomalies, and build human review checkpoints. Over time, this can evolve into more sophisticated self-monitoring systems.

What You Can Realistically Achieve Today

Based on current AI and automation capabilities, RCM and TPA operations can reasonably achieve:

• 50–75% automation rates on well-defined, high-volume claim workflows
• Significant reduction in manual touchpoints for eligibility, benefits verification, and clean claim adjudication
• Measurable improvement in denial rates through AI-assisted coding and pre-submission claim scrubbing
• Faster onboarding through automated member enrollment and eligibility confirmation

The University of Colorado’s Health Administration Research Consortium notes that AI-driven RCM systems must navigate multiple layers of healthcare law including HIPAA — a further argument for incremental deployment with proper compliance checkpoints at each stage rather than monolithic deployment.

How to Compete Against AI Native RCM Companies

If you’re a regional RCM company, an independent TPA, or a technology-enabled healthcare services firm, here’s the practical competitive playbook:

Lead with Operational Credibility

AI native RCM companies often lack deep operational history. You have it. Use it. Case studies, client outcomes, denial rates, days in A/R, claim processing accuracy — these are metrics that a company that’s been live for six months cannot match. If your clients run on Tebra, a Tebra dashboard can surface these numbers, and for eClinicalWorks practices, eClinicalWorks reporting can do the same.

Sell Reliability Over Innovation

For plan sponsors with significant employee populations, reliability matters more than novelty. A $600 million AI native TPA that has pivoted its core operating model multiple times is not a stability story. You are.

Point to the Pivot Track Record

Athelas pivoted from RCM entirely. Angle Health acknowledged a “hybrid” model after pitching pure AI. These are facts, not opinions, and they’re fair to raise in competitive sales conversations. A prospect considering whether to trust their claims administration to a company that may not be in this business in three years is making a legitimate risk assessment.

Invest in Technology Continuously

Competing against well-funded AI native RCM companies doesn’t mean ignoring technology — it means deploying it intelligently. A modular AI strategy that delivers measurable results today is more credible than a platform that promises full automation and delivers a hybrid model with heavy staffing.

Emphasize Flexibility

Your modular architecture means clients can adopt what works for them, customize to their operational requirements, and not be locked into a brittle monolithic system. That’s a genuine differentiator, especially for mid-market plan sponsors and provider organizations with complex billing needs.

Conclusion

The AI native RCM narrative is compelling, well-funded, and — based on the evidence — significantly overhyped. Companies like Athelas and Angle Health have demonstrated that even with hundreds of millions of dollars and elite engineering talent, the all-AI vision in claims processing consistently runs into the hard wall of operational reality. The response, in both cases, has been to add humans back into the process and reframe the original pitch.

That doesn’t mean AI isn’t transformative in RCM and TPA operations. It absolutely is, and companies that ignore it do so at their peril. But the transformation is happening through modular, iterative deployment — not through monolithic AI native platforms built in a vacuum before a single client ever goes live.

If you’re running an RCM company, a TPA, or a healthcare claims processing operation, the strategic path forward is clear: centralize your data, deploy AI and automation incrementally across high-impact workflows, build evaluation and oversight into every module, and keep improving. You’ll outperform the overwhelming majority of your competitors — including the ones with the biggest war chests — because you’ll be building something that actually works.

The resistance isn’t futile. The resistance, done right, wins.