Medical billing denials are rising: Can AI help health care practices get paid faster?

The complexity of medical payment processing continues to challenge health care practices across the country. From navigating constantly changing payer rules and prior authorization requirements to managing increasing claim denials and fragmented systems, providers face mounting frustration in getting paid for services they've already rendered. The problem is compounded by the need for specialized billing staff who are in short supply and prone to burnout. While technology and artificial intelligence promise solutions, determining which tools actually deliver results remains difficult.

Medical Economics spoke with, Roshan Patel, founder and CEO of health care payments company Arrow, to discuss the current state of medical payments, why these challenges persist, and how emerging technologies might help—without completely replacing the human expertise that remains essential to navigating this complex landscape.

Medical Economics: What are the biggest pain points right now in payment processing for medical practices?

Roshan Patel: I think it's always a challenge to get paid by insurance, and especially nowadays, it's becoming more and more complicated. The payer rules are constantly changing. There are more prior authorization requirements. There are more denials happening. Claims are getting rejected. Systems are more fragmented, so it's just harder to get paid by insurance for the services that you've already rendered. And there's a lot of reasons why that is, but I think we're just seeing mass frustration across the board from most physicians and providers in the industry.

Medical Economics: Why do these pain points exist? Why haven't they been fixed? Isn't it in everybody's best interest to have a smooth payment processing system?

Patel: I think it's a complicated question, because I think it's hard to align incentives in the way health care is paid for in our country. Fundamentally, one side wants to get paid and one side doesn't want to pay. So it's almost like a zero-sum game. I think the way health care payments has come about from when insurance as a model originated, systems are so fragmented, it's just hard to even get the correct data. Even knowing if a patient's insurance plan requires prior authorization for a certain procedure, it's hard to just get that information. Oftentimes it's not easily accessible in your EMR or clearinghouse. You have to call the insurance company. Oftentimes they don't know—they're so large and their systems are so fragmented—so it's just kind of a mess.

And this problem is compounded because there's so many different payers, so many different rules. Plans are becoming more complicated, and there's just so much manual work that has to get done. And when there's that amount of manual work, you need specialized, trained staff, and that staff is hard to find. They're constantly getting burnt out and leaving, and it's hard to replace them. So the whole system is kind of a mess right now.

Medical Economics: So what's the solution? How do we fix this?

Patel: I think if there was an easy answer, we wouldn't be here today—this would have been solved. But typically, I see providers go one of three routes. One is they just hire more people in-house. Just throw more people at the problem, because it is a ton of work, and that usually can help. Second is try and outsource it. There's obviously a lot of medical billing companies, both in the US and abroad, that are specialized to handle this, and a lot of practices just don't have the ability to spend time hiring, or maybe they're growing and so they don't want to deal with hiring staff—that's a great option. And then lastly is use technology. I think technology has really accelerated in the past decade, and especially in the past few years with AI. So there's a ton of tools out there which practices can take advantage of, but it's also hard to know which tools to use, which ones actually work. As with any new technology that comes about, there's going to be a ton of noise, so it's really hard to make sense of what actually works.

Medical Economics: How will AI and automation solve these pain points? How will it make it better for physicians?

Patel: I think in a lot of ways. There's sort of two models that I see right now. One is using AI to automate work. And so you're sort of taking a workflow or a job or a function and getting AI to do that completely, end to end. So one great example we've seen is scribing. That used to be something that doctors had to manually type, and now that whole workflow is pretty much automated, and that's been a huge time saver. The other approach that I see is using AI more as a copilot or collaborator, where maybe for more complex workflows or higher-stakes workflows, you don't really want to outsource it completely to AI—you want to have more control and visibility over things. And that's a lot of what I see in revenue cycle and billing, just because it's so important. The stakes are very high. You're dealing with money, you're dealing with the financial health of a practice.

It's not like ChatGPT. It's more of a back-and-forth conversation. You're not just outsourcing an entire task to ChatGPT.

Medical Economics: What would this look like to the end user? Is this something where the AI system will flag a certain insurer and say, 'Hey, insurer X always wants to see this data. Make sure you include that,' or what does it look like in the end?

Patel: It's hard to say, because revenue cycle and billing encompasses so many different tasks, right? Insurance verification, coding, preparing claims, submitting them, denials, payment posting, patient billing. But you bring up a great example. Especially before claims go out, so much can be done with technology and AI to ensure those claims are clean, we're adhering to payer rules, we're flagging any abnormalities or things that would result in a denial. A lot of those tend to be more rules-based, so they don't really even need AI. You can just make sense of the data you already have.

A case that may take more advantage of AI would be writing appeal letters once you've gotten a denial. That can take a biller several hours to do all of the research, compile the information, write a compelling letter, and submit it. But now with AI, AI is very good at producing content, assembling data, and writing. So we see tasks that used to take hours now take minutes.

Medical Economics: What about patient payments? Practices are dealing with insurance companies, but they also have some patients that are paying cash or through a health savings account. Is that a separate system, or does that need to be integrated so that all the payments are in one system within the medical practice?

Patel: I think a huge problem right now is that there really is no one system of record or source of truth. I think the EMR tries to do that, but fundamentally, the EMR is really built as more of a clinical tool, not really focusing on billing and payments. So oftentimes, when I ask providers how much they're collecting or what their denial rate is, they just don't know. The data is not very easily accessible, because there is no one source of truth. In an ideal world, yes, everything is in one place, and it should be all connected. But I think I see most physicians and practices use their EMR's patient billing functionality as the way to collect the patient responsibility, whether that's their balance after insurance is paid or they're just paying private pay.

Medical Economics: If a practice wanted to modernize its payment systems and do everything it could to streamline the process, what steps should they take? What questions should they be asking?

Patel: I think the best place to start is just ask your peers: what are they using? What are they seeing that's working? Because if you just Google 'revenue cycle technology' or 'AI tools in billing,' you're going to see dozens, hundreds of companies, and it's impossible to do your research, because all the websites say the same thing. They all look great. They all promise the world. So I would just get trusted information from peers that have actually tried it. I think depending on the size of the practice, usually the smaller you are, you don't really want too many tools that are separate. You kind of want one or two that do most of the work. But the larger you get, it tends to make more sense to look for point solutions that do one thing really well and really customize that to your practice, because you're much larger and you may have a different way of doing things.

And I also think conferences and events are a great way to learn about tools. You can go to booths and see the vendors right there and get conversations going. You don't have to wait to do your research and book a sales demo.

Medical Economics: Are there any metrics that practices could use to measure the impact that these systems would have, or metrics they should be asking about to a vendor?

Patel: I think it's a great question to ask the vendor: what metrics are you going to measure and improve for me? But I think it kind of comes back to the system-of-record problem where no one really knows their metrics. The metrics are also hard to measure, so there's just a lot of inability to know what works. But I think if you're on top of your numbers, you can come to the vendor and say, 'Here's my collection rate, or here's how long it takes me to get paid by insurance on average. How does that stack up to what you see in the market? And what are you going to do to those metrics to get them to where they need to be?' I think the more informed you are as a physician or practice owner, the more you can discern what's real and what's not, because the vendors are always going to promise that you're not doing well and the vendor is going to make you do well.

Medical Economics: Is there one particular area where these systems really make the biggest difference?

Patel: I would say it's when there's a lot of manual work required. I would say that's probably in most places in revenue cycle and billing, but the one that I see the most is when you're dealing with denials or claims that have not been paid. That's typically the most labor-intensive part of revenue cycle, because it's not as simple as just clicking a button to verify a patient's insurance. You have to do a lot of digging to understand why something was denied. Oftentimes, the reason that comes back from the insurance company is often not the right reason, or the reason is very vague, and you have to call them and figure out what's going on, and you have to write appeal letters. And even when you submit that appeal, you have to follow up on it and make sure it gets paid out. So it just creates a lot of work, and that's a great area where I see use of technology.

Medical Economics: Looking out five to 10 years, where do you see these systems? What pain points do you think they will solve in the long run?

Patel: I think it'll mostly be the same pain points. I think there's an argument whether AI is going to replace medical billers or not. I think a lot of people in the industry think that it will, just like most types of knowledge work could be completely automated. But I don't think that's the case. I think there's a lot of specialized knowledge within a practice—billers have decades of experience of, 'Oh, this payer always requires this certain little quirk when we submit the claim'—that it's just hard for AI to pick up on because it's not going to have all of the context and data that comes with years of billing experience. So I personally think we'll still have medical billers around, but they'll be strongly empowered and more efficient and more productive by the tools that we have.

I don't think health care payments are ever really going to be solved the way the current system is set up. There's always going to be denials, there's always going to be friction in the process. We're probably not going to have some sort of real-time payments in health care for a very long time. So I think this job is going to be required, but just made a lot less frustrating.