With hospital operating margins often sitting in the single digits, the complex maze of medical billing is silently costing health systems millions in uncollected revenue. In this episode of HealthTech Remedy, we sit down with Benjamin Beadle-Ryby, co-founder of AKASA, to explore how generative AI is fundamentally transforming revenue cycle management (RCM). You will discover how shifting from manual coding to an autonomous revenue cycle can drastically reduce claim denials, uncover missed charges, and ultimately protect a hospital's financial mission.
The conversation unpacks the structural complexity of hospital reimbursement, where a single inpatient encounter can generate 50,000 words of documentation that human coders must manually distill into precise ICD-10 and DRG codes. Benjamin shares the framework behind AKASA’s custom large language models, revealing how fine-tuning healthcare AI on a health system's specific historical data can seamlessly surface missed quality indicators in up to 10% of claims. We also debate the delicate balance of AI automation and human oversight, questioning what happens to the highly specialized knowledge of medical billers as these tools evolve. You’ll have to listen to find out why traditional rules-based systems are failing and exactly how generative AI could drive the cost to collect below one percent by 2030.
If you enjoyed this deep dive into the intersection of health tech and healthcare finance, please subscribe and leave a review for HealthTech Remedy.
AI Chatbot Preferences and Workflows
Dr. Tim Showalter:
Morning.
Dr. Trevor Royce:
Good morning, guys. I have to tell you, I'm starting to feel old-fashioned. I'm still using ChatGPT over here.
Dr. Tim Showalter:
Paul, you're a Claude fan, right?
Dr. Paul Gerrard:
I am. I'll be honest with you, I thought all the LLMs were junk until I tried Claude.
Dr. Tim Showalter:
And this is clearly—there's no political angle here, we should clarify to our listeners.
Dr. Paul Gerrard:
Absolutely not. I've thought this for months on end now.
Dr. Tim Showalter:
The only thing is my ChatGPT knows me now. I've got a few ongoing discussions where I've loaded up a bunch of information. That part is pretty nice.
Dr. Trevor Royce:
You better switch over to Claude before it knows you too well, Tim.
Dr. Tim Showalter:
That is true. Actually, I can start fresh. I could be a serial monogamous for my generative AI. Well, speaking of AI, should we hop in?
Dr. Trevor Royce:
Let's do it.
Dr. Paul Gerrard:
Let's do it.
Introduction to Health Tech Remedy
Dr. Trevor Royce:
Welcome to Health Tech Remedy, the podcast where three physician leaders discuss health tech innovations shaping the future of care. I'm Trevor Royce, a radiation oncologist with a background in AI diagnostics and real-world evidence.
Dr. Paul Gerrard:
I'm Paul Gerrard, a former physical medicine and rehabilitation physician focused on reimbursement policy, molecular diagnostics, and market access for AI-enabled technologies.
Dr. Tim Showalter:
And I'm Tim Showalter, radiation oncologist and former med tech entrepreneur, now working on scaling AI technologies that improve care delivery.
AKASA and Autonomous Revenue Cycle Management
Dr. Trevor Royce:
This week, we're diving into AKASA, a company applying generative AI to one of healthcare's most sticky and often esoteric domains: revenue cycle management.
Dr. Paul Gerrard:
Then we'll speak with Benjamin Beadle-Ryby, co-founder and SVP of Commercial at AKASA, about what it really means to build an autonomous revenue cycle.
Dr. Tim Showalter:
I think this is one of the most important topics in healthcare because obviously: no margin, no mission. It's such a challenging area. We've talked before on this podcast about how the underlying medical record and documentation pressures on physicians tie in with reimbursement overall. It's a good one for us to dive into.
If you're not familiar with AKASA, it is a San Francisco-based healthcare AI company founded back in 2018. The focus for the company is on automating revenue cycle management, specifically on workflows for hospitals and health systems.
The idea is that revenue complexity ultimately requires higher-level technology with adaptive AI systems rather than static automation. This branded approach is referred to as autonomous revenue cycle. By enabling end-to-end automation of financial workflows, you can limit the need for human oversight and help health systems scale.
The Structural Complexity of Hospital Billing
Dr. Paul Gerrard:
Getting back to "no margin, no mission," the margin in healthcare comes from reimbursement. Medicare is obviously the biggest payer, and a lot of commercial payers benchmark on Medicare. However, hospital reimbursement is something very few people understand.
On the inpatient side, ICD-10 codes ladder up to DRGs. On the outpatient side, CPT and level two HCPCS codes level up to ambulatory payment classifications. It's tough to find people within hospitals or payers who can really put all the pieces together on how individual services ladder up to specifically what gets paid.
This results in a situation where it might be clear what diagnoses a patient had, but if you have many diagnoses, you have to figure out which ones to put on the claim. Understanding how they combine to create a billable service is important.
The rules are also changing. Sometimes it takes a team to keep up. You have a lawyer reading the new policy, and the finance and billing people figuring out how to implement what the lawyer tells them. This is an opportunity for something that has been a highly manual process and has not been automated in the past. There are good opportunities for automation with AI.
The Strategic Value of AI in RCM
Dr. Trevor Royce:
I'll double down on that. This has probably been one of the hottest areas in the generative AI era in health technology. There are a ton of companies trying to tackle revenue cycle management from various perspectives.
AKASA is really well positioned. They were founded as early as 2018, so they were working on this problem before a lot of these large language models really came to market. It is also a very data-rich problem.
There's tons of information being exchanged on the revenue cycle management side. The impact on health systems is huge. AKASA's goal is to reduce denials, improve margins, and increase revenue through advanced generative AI. They have raised over $100 million with some very big investors and backers.
Dr. Tim Showalter:
As someone whose familiarity with revenue cycle management is mostly what we're asked to do as clinicians, evaluation and management coding can be tricky. You have to make sure your documentation is correct, and then there's so much more work that happens downstream.
Operating margins on hospitals are very thin, sometimes in the single digits—I've seen one to 3% quoted. When the revenue cycle breaks down and you have lost charges or unresolved pain points, it can be a major problem. If you start to drop claims unnecessarily, it can really shut down a health system.
In this context, how autonomous is autonomous? Obviously, there are still things asked of healthcare providers and staff to make sure the information needed to drive revenue cycle management is there. Some day-to-day actions in clinic will still need to be there, but it's going to be helpful to have a lot of that built into the system.
Dr. Paul Gerrard:
Even if things are not fully autonomous, having AI that can flag to a human reviewer where something is in the record already provides time savings. Health systems and hospitals tend to have a staff on the back end that traditionally had to look through hospital records.
That was usually moderately skilled labor, often someone with at least a nursing degree, trying to make sense of diagnoses. It's very manual. You have to look at a lot of text to find a little bit of information.
Something like an improved search engine that knows what to look for within the hospital record is already moving things in the right direction. It reduces labor and the costs of getting paid on the services that hospitals are legitimately doing but struggling to appropriately identify to payers.
Specific Product Applications for Generative AI
Dr. Trevor Royce:
I thought it might be useful to illustrate some specifics regarding how they position their product. What does it mean to use generative AI for revenue cycle management?
One example is a coding optimizer. This ensures you have the right billing codes to get paid for your services. There is a ton of data with associated codes well-positioned for generative AI.
Related to that is a clinical documentation optimizer. You need documentation to support those codes and ensure it's accurate. AI can flag that. They also handle prior authorization status—checking those statuses automatically to reduce manual follow-up. Finally, they provide automated claim status updates to reduce the burden on humans who typically follow up on claims.
Balancing Human Expertise with AI Accuracy
Dr. Tim Showalter:
Whenever a new AI technology comes out, there's always a discussion about accuracy, trust, and the hallucinations you can get with these tools. But when you look at something as complex as coding and reimbursement, the comparison is interesting.
Even if the AI is not perfect, there are so many challenges to this process. As Paul pointed out, policies are always changing and it can be confusing for clinicians to ensure their notes accurately reflect all revenue-driving services.
It strikes me as an ideal situation for AI. There's still source documentation available so the revenue cycle management staff can go back and check things. There's a real opportunity because I doubt humans do that perfectly. Paul, you've been in this space for a while. What do you think about the current accuracy before technology?
Dr. Paul Gerrard:
I'm going to say highly variable. I'm sure when you guys have been in the hospital, you've received calls from someone in revenue cycle management saying, "I see you're treating the patient for this and giving them this drug; does the patient have this diagnosis? If they do, could you mention it?"
I've helped with that at some of the places I've worked. It was a very manual process. You basically had to know what costs the hospital money and what pays. Then you connect the dots.
You take your medical knowledge and know that when you see someone with condition XYZ, there's a high probability they have one of these other costly things. Sometimes it's clearly mentioned in the record and the physician is treating it, but the person extracting the data isn't aware of it or its importance. Other times, it's not even mentioned because the clinician is using systems-based noting instead of problem-based noting.
Dr. Trevor Royce:
The sheer complexity of medical billing is often underappreciated. It's absurd to take a complicated human patient-doctor interaction and distill that into numbers on a page as billing codes.
I want to give a shout out to Michael Lewis's podcast, *Against the Rules*, specifically an episode called "Six Levels Down." He goes deep into the absurdities of medical coding.
The conversation focuses on Athena Health and how they cracked the code of billing to optimize how to bill for things. It mentions a medical coder who had developed this totally arcane and esoteric knowledge on how to optimize billing. Her services were so valuable but unrecognized. This isn't sustainable. Let's hope generative AI can help make this more efficient.
Dr. Paul Gerrard:
What happens when that person retires?
Dr. Tim Showalter:
Exactly. She needs a digital twin who can live on.
Dr. Trevor Royce:
Absolutely.
Dr. Paul Gerrard:
The flip side is you could say we should recognize the value of that skill more and encourage people to go into it. One of the concerns about AI is that some of this expertise is going to be devalued. But here we're talking about a problem where not many people even have the expertise and it's already undervalued.
Interview: Benjamin Beadle-Ryby of AKASA
Dr. Tim Showalter:
I'm looking forward to hearing from Ben firsthand about their plans and how they're using technology. Any closing thoughts, guys?
Dr. Trevor Royce:
I'm excited to chat with a real leader in this space. There's a ton of competition and a lot of startups thinking about AI and RCM for different entities. It'll be great to have him on.
Dr. Paul Gerrard:
The idea sounds like a clearly good one. Going from idea to implementation is always the hard part. It'll be interesting to hear from someone who has bridged that gap.
Dr. Tim Showalter:
All right. Stay tuned. Up next is our interview with Benjamin Beadle-Ryby, co-founder and SVP of Commercial at AKASA. Ben, welcome to Health Tech Remedy.
Benjamin Beadle-Ryby:
Thank you so much for having me, Tim. Delighted to be here. I really appreciate the opportunity.
Dr. Tim Showalter:
We've been looking forward to this. AKASA is taking on some of the most consequential and under-discussed problems in healthcare. The revenue flow through health systems is critical to taking care of patients.
Benjamin Beadle-Ryby:
Me as well. Again, thank you so much.
Dr. Trevor Royce:
To kick things off, we'd love to hear your origin story. Tell us how you went down the path to co-found AKASA.
The Origin Story of AKASA
Benjamin Beadle-Ryby:
Prior to founding AKASA, I spent my career working with hospitals and health systems. I was a partner at the Advisory Board Company. The Advisory Board's focus was identifying best practices to improve healthcare performance. At this stage, I've had the privilege of working with well over 350 hospitals across the country.
It was really interesting from my vantage point. When it came to technology innovation efforts, particularly out of Silicon Valley, a lot of emphasis was placed on how you change or improve clinical care. That has merit, but when I was advising health systems, I kept seeing a recurring pattern.
There is extraordinary clinical work happening, followed by a back office where you have fragmented documentation, manual handoffs, and armies of people trying to reconstruct what happened. Often, organizations and physicians weren't getting the full credit for the care they were providing.
I started focusing on what was happening behind the scenes. It's a little less sexy, but it's critically important. I think of the revenue cycle as sitting downstream from clinical care, but upstream from financial survival and quality performance. It relied heavily on manual efforts, people, and rules-based systems. We felt there was opportunity to drive improvements in performance, accuracy, and efficiency with modern machine learning and advanced AI. That's what started AKASA.
Dr. Trevor Royce:
Can you tell us exactly what revenue cycle management is? Why is it such a big problem?
Benjamin Beadle-Ryby:
The revenue cycle has been around for decades. Unfortunately, if you look back over time, the amount of change that's occurred in that space has been minimal. We've introduced some systems and software that facilitate those workflows.
The revenue cycle is the process by which we take in patient information, demographics, and insurance information, and couple that with the care provided. We convert that care into documentation in the medical record, which translates into codes.
Those codes have implications for how organizations are reimbursed and the quality for which they get credit based on outcomes. The revenue cycle sits behind the patient care process but facilitates the financial performance of health systems across the country.
Dr. Tim Showalter:
It's so interesting how important it is. As physicians, you really don't learn about all this until late in residency. You're already in practice by the time you start to figure out how to optimize it.
Benjamin Beadle-Ryby:
Tim, it is remarkable. Many clinicians across the board may not fully understand or appreciate aspects of the revenue cycle, or they have an inadequate understanding of what that process means for them and how their care is evaluated.
There are some really revealing insights when we apply AI behind the process. We show that there are actually opportunities missed as a result of insufficient documentation or an inability for the organization to translate existing documentation into the appropriate codes.
Analyzing Current Trends in Claim Denials
Dr. Tim Showalter:
Your company predated the recent obsession with generative AI, yet you've shown a real impact with AI-driven automation. What was the key insight that led to the realization that it was important to take an AI-enabled approach rather than incremental tooling?
Benjamin Beadle-Ryby:
The industry has seen no shortage of efforts to try and improve revenue cycle performance. I applaud all the groups that have attempted to address underlying issues with outsourcing, rules-based systems, and even robotic process automation (RPA).
Ultimately, while there have been some incremental gains, the industry has failed to move the needle in a meaningful way. If you rewind to 2019, initial denials at 9.8%. In 2025, that number has jumped to 12.7%. One in eight claims are hitting a payer roadblock.
During the same time period, write-offs have jumped from 1.3% to 3.6%. The cost to collect—the amount organizations deploy to collect the dollars due—has increased from 2.9% to 3.7%.
By almost every major measure, we've gone backwards through the old approaches of throwing more people at the problem. We saw that AI could help deliver more accuracy and impact even before the emergence of large language models. In the organizations where we apply that AI, it has been driving measurable gains.
Quantifying AI’s Impact on Revenue Capture
Benjamin Beadle-Ryby:
When our AI sits behind human workers to spot-check the work, we're finding that on 8% to 10% of every claim, there are errors—generally errors of omission. For every 10,000 discharges our AI reviews, we surface 4,200 additional quality indicators that should have been on the record. It's also resulting in over $3 million in additional revenue capture for these organizations. Present on admission capture is up by 11.5%, and Elixhauser codes are up by 7.5%.
I think it is important to recognize the complexity. A single inpatient encounter contains on average 59 different clinical documents. There are nearly 50,000 words, which is the equivalent of reading *The Great Gatsby*. We ask back-office teams to review all 59 documents to get the narrative and pull out discrete details to select the right codes from over 150,000 options. It's a Herculean task, but a problem that is ripe for AI.
Dr. Trevor Royce:
That 10% rate of unpaid claims is a very clear value proposition. On the payer side, presumably they also get annoyed by claims that have to be denied. Have they emerged as a customer at all?
Benjamin Beadle-Ryby:
Our work to date has been focused on health systems and ensuring accuracy and integrity in everything they do. A downstream byproduct for the providers is that they are accurately reflecting the care being provided.
The AI is flagging both areas of opportunity and instances where something might not be justified. For payers on the receiving end, they should actually feel more confidence in every piece submitted. A holy grail scenario would be if AI can be an arbiter of truth between these groups to foster trust and reduce administrative waste.
Human Oversight and AI Augmentation
Dr. Trevor Royce:
The generative AI world seems like a perfect application for this. Tell us a little bit about how the human element is still needed. Where does the human come in beyond the AI?
Benjamin Beadle-Ryby:
At the heart of healthcare is humans. Generative AI presents an ability to augment the work humans are doing and drive every worker to operate at the top of their license.
We think it's important that there is human oversight. We're not talking about removing humans from the equation entirely, but fast-forwarding the care process so AI brings insights to the fingertips of providers and staff so they can focus on the patient.
Dr. Tim Showalter:
One of the advantages of AI is that models can continually improve. I'm curious to hear your strategy for balancing local optimization within a health system versus generalizability across all customers.
Benjamin Beadle-Ryby:
Many organizations are concerned because their labor force is retirement eligible. AI provides sustainability and long-term financial viability.
Regarding complexity, an example is a pulmonary embolism housed in an addendum that human reviewers missed. The AI can simply say, "I found this, let's make sure we capture it."
Change Management and Technical Integration
Benjamin Beadle-Ryby:
In terms of scale, open-source models don't capture the nuances of each health system. For each organization we work with, we take a year’s worth of their historical data and create a fine-tuned, custom large language model specific to that system. Cleveland Clinic’s population is different than Duke’s or Hopkins’. Training an AI model to handle that specificity has resulted in greater improvement.
Dr. Trevor Royce:
How do you measure your success? What measures are you using to say your products are doing what they should do?
Benjamin Beadle-Ryby:
The only way to build trust is by being transparent about the impact we're having. Key quality measures include severity of illness, risk of mortality, present on admission capture, HCCs, and risk adjustment factors (RAF scores).
There are instances where capturing complications changes the DRG. We want to ensure integrity in every code assigned. Finally, we measure efficiency. I was just on the line with a group that saw the productivity of their staff improve by over 20% during a pilot.
Dr. Tim Showalter:
The deployment involves change management and technical integration with EHRs. What's harder? Where do you spend most of your time worrying?
Benjamin Beadle-Ryby:
I think proliferate APIs and standards-based integrations have opened up AI's ability to extract information in ways we haven't been able to do previously. Nearly 70% of the world's data is healthcare data, but it has largely been untapped.
We pull and extract information out of FHIR APIs and EDI files. That's the first step. The later step is the change management. We're talking about changing the way things have been done for decades.
Our first step is to have our AI review 100% of the work humans are doing without changing their workflow. We just flag where they may have missed opportunities. Over time, the AI can start performing those tasks first. We want to be very mindful of each organization’s appetite for change.
Dr. Trevor Royce:
Why hasn't this been solved before? Why is it so hard, and what sets you guys apart from the competition?
Benjamin Beadle-Ryby:
A lot of the industry has attempted to improve this. Things like prior authorizations and denials are known pain points. However, documentation and coding are a bit out of sight, out of mind.
Given the complexity, most organizations weren't aware of the opportunities that existed. When you have powerful AI sitting behind and showing that you're missing things in 10% of cases, it's eye-opening.
Rules-based systems have inherent limitations. You could talk about the board game Go. Google DeepMind team's AlphaGo showed that you can't build enough rules to capture the nuances, but you can have AI teach itself the rules. The same is true in documentation and coding.
The Future of Revenue Operations in 2030
Dr. Tim Showalter:
Looking forward to 2030, what does a best-in-class revenue cycle operation look like?
Benjamin Beadle-Ryby:
The telescoping of milestones is remarkable. Today, you have armies of people doing these functions, and it's rife with error. In five years, these will be streamlined boutique shops with teams of people savvy at leveraging data.
The average health system's cost to collect is 3.7% of net patient revenue. In the future, that should be sub 1%. We should be spending less on administrative pieces and more on the care itself.
Closing Thoughts and Career Advice
Dr. Trevor Royce:
What's something that's surprised you about revenue cycle management that you wouldn't have anticipated when you started out?
Benjamin Beadle-Ryby:
Innovative thinking requires you to question why things are done the way they are. The emergence of AI has been adopted at a faster clip in healthcare than any prior technology. That has been most shocking to me. I've seen nearly 20 years of reluctance and skepticism. Now, organizations aren't waiting; this has become mission critical.
Dr. Tim Showalter:
Final question: what early career experiences set you up to execute in your role today? What advice do you have for the younger generation?
Benjamin Beadle-Ryby:
A lot of credit goes to the Advisory Board, where there was a mentality of mission above commerce. Mission is what drives each of us in healthcare.
I would encourage anyone starting their career to deeply understand what happens today, but then ask critical questions as to what we can do to improve outcomes. We fortunate enough at AKASA to translate that ethos into work using the most powerful technology we've ever seen.
Dr. Trevor Royce:
This has been a fantastic conversation. It's been a really great insight into revenue cycle management. We appreciate you taking the time.
Benjamin Beadle-Ryby:
Thank you so much for having me. I really enjoyed the conversation.
Dr. Trevor Royce:
We wish you guys the best of luck. This has been another wonderful episode of Health Tech Remedy, and we'll see you guys next time.
