AI for Healthcare: Jon Jaroska on Smarter Claim Denial Fixes

Denied claims don’t just cost money; they drain time, trust, and resources. But what if AI could turn the tide before the damage is done?

In this conversation, Jon Jaroska, CTO, Red Sky Health, breaks down how AI is being used to solve one of healthcare’s most persistent challenges: claim denials. He dives into why traditional approaches fall short, what makes denial remediation so complex, and how Red Sky’s AI-native platform is helping providers recover lost revenue at scale.

What experiences from your earlier roles shaped your approach to building complex, mission-driven systems like Red Sky Health?

Before joining Red Sky Health, I ran a custom development company focused on helping clients overcome high-stakes problems in healthcare, finance, and engineering. We built systems that meet performance and functional requirements, scale under pressure, make it easier for users to do their job, and for the business to be more profitable. I don’t romanticize complexity; rather, I focus on what moves the needle. At Red Sky Health, that means building AI and tech solutions that don’t just sound smart, but actually get the job done by enabling healthcare providers to get claims paid quickly.

Denied claims are a long-standing issue in the healthcare system. From a technical standpoint, why is solving this so complex?

It really comes down to two things: the data is a disaster, and the rules are intentionally opaque. They’re dealing with fragmented systems, inconsistent formats, legacy EDI files, PDFs, scanned faxes, handwritten notes, etc. Every payer has their own playbook, and they’re not exactly incentivized to make denials easy to reverse. And let’s face it, there isn’t an app available that can determine why a claim got rejected. Due to the vast number of factors that can contribute to why a claim can get denied, providers need to apply AI to reason through the data, spot patterns, and make judgment calls like a human expert would. The difference is that AI can do it faster, at scale, and without the typical burnout experienced by humans who attempt to do the same task.

Anyone saying “we’ll just train a model” doesn’t really understand the nuances of the overall challenge. Red Sky Health does, and that’s why we are so successful.

What are some of the overlooked challenges you’ve encountered at Red Sky Health when applying automation in legacy-heavy healthcare environments?

Everyone talks about AI like it’s magic. But in legacy-heavy healthcare, that's the equivalent of duct-taping intelligence to a flip phone.

The biggest and most overlooked challenge with claims denials remediation is that the infrastructure is so fragmented. You can build a slick automation pipeline, but it still has to plug into outdated systems running on COBOL, sit behind a fax queue, or wait for an SFTP drop. Half the time, they’re not integrating with a system; they’re just trying to work around the issue.

Then there’s the human side. Because people have become accustomed to broken workflows, they’ve built manual workarounds, which means that they don’t trust automation until it proves itself repeatedly. This indicates that you're not just replacing a task; you’re undoing years of process debt.

So, while technology may be bleeding-edge, it remains difficult because the environment is stuck in 1998. But that’s where the opportunity is if you’re willing to do the hard work. Red Sky Health is, and that’s why providers are quickly adopting our proprietary AI solution, Daniel, to recover lost revenue by using advanced machine learning and generative AI algorithms to analyze historical claims data, identify and correct errors, and streamline the resubmission process.

What architectural decisions have been most critical in enabling real-time claim remediation at scale?

First, Red Sky Health ditched the idea of building on top of legacy platforms. Too many healthtech startups bolt their solution onto some outdated system and wonder why it can’t scale. Our solutions were built AI-native from day one, which means every architectural decision already took into account massive data ingestion, real-time processing, and explainability.

Second, our solutions are modular. Rules engines, classification models, document extraction, payer logic—they’re all decoupled so we can iterate or swap components fast without dragging down the whole stack. That’s key when the rules change weekly and clients need results yesterday.

Third, we avoided tech religion. It’s not about picking the hottest stack but about choosing what works under pressure. We use message queues, serverless where it makes sense, and persistent storage when it doesn't. If something slows us down, it’s quickly gone.

The goal was never to be clever; it was to be fast, accurate, and adaptable. That’s why it works.

Can you share how your AI model prioritizes which claims to fix first or flags as high-impact?

Absolutely. It’s not just about what’s wrong but what’s worth fixing. Daniel prioritizes claims based on a few things:

• how much revenue is tied to it,
• how likely it is that we can correct it with available data, and
• how fast we can get that fix back into the client’s system.

We’ve trained the AI to spot patterns in claims denials that are cleanly fixable and high-impact, such as missing codes, mismatched eligibility data, or format issues that don’t require human escalation.

This isn’t academic AI. We’re not optimizing for F1 scores but optimizing for money back in the door. And it’s not about being AI-only, but human-augmented AI. If a $50 claim is technically wrong but is going to waste hours of admin and never get paid, the provider can choose to skip it. If a $10,000 claim is fixable with one clean submission, the provider can prioritize what gets addressed—quickly and easily. That’s how we turn AI into actual ROI.

How does Red Sky Health ensure data security and compliance while handling sensitive financial and clinical information?

We treat data security like a product feature. It’s a checkbox, not an afterthought.

From the outset, Red Sky Health designed the infrastructure to meet the demands of healthcare-grade compliance. That means encryption at rest and in transit, strict access controls, full audit trails, and role-based data partitioning. No one touches anything they shouldn’t, and everything’s logged.

We’re also SOC 2 compliant and are actively building toward HITRUST because we know what’s at stake. It’s not just PHI—it’s trust. Our environment is designed to ensure that no one sees what they shouldn’t, and we back that up with continuous monitoring and regular reviews.

But here’s the challenge that most companies won’t admit: security isn’t static. Threats evolve. So, we build with paranoia in mind and bake in continuous monitoring, regular third-party audits, and strict vendor reviews. There’s no room to relax in this space, and we don’t.

Do you envision a shift from reactive denial fixes to pre-claim intelligence? How close are we to that future?

Yes, we do, and we’re already building toward it.

Right now, we’re recognized for being superior in identifying and fixing denials after they happen, but the real game-changer is stopping them before they ever get submitted. That means applying the same AI logic earlier in the lifecycle: catching bad codes, mismatched eligibility, missing documentation, and broken charge structures before they hit the payer.

The shift is from firefighting to prevention. And the good news is, the data is there. Most of the patterns that lead to denials are predictable. What’s been missing is the technology that can interpret that mess in real time and push corrections upstream.

We’re already surfacing insights that hint at root causes, and it will continue to evolve. Pre-claim intelligence is the next natural step, and we’re moving fast in that direction. It’s where this market is headed, and we plan to lead it.