I have spent most of my career building software that helps physicians take care of patients. Too much of what the HCIT industry has built serves billing, compliance, and documentation requirements — all real needs, but not the core need of helping a doctor engage with a patient and help that patient get better or stay healthy.
In 2009, the government had to resort to legislating the use of EMRs through ARRA and Meaningful Use. Those EMRs were modeled on the paper chart. You flip through the paper chart; you flip through the EMR. That was a reasonable first step, but we haven’t yet moved past that. These systems are good at organizing information, but are far less effective at helping people act on it.
The industry calls what comes next Service-as-Software, or SaS — software that does the work, not just the organizing. In healthcare, where physicians are already stretched to their limit, that shift cannot come fast enough.
Why layering AI onto old systems does not work
The risk with much of AI in healthcare today is that it gets bolted on — added as a feature to systems and workflows that were never designed for it. The AI surfaces a recommendation, but the physician has to jump to one screen to verify the source and another to enter the order. The data underneath is incomplete, or hours stale. The workflow the AI lives in doesn’t connect to the workflow where the work actually gets done. So the physician double-checks, finds gaps, and stops trusting the output. It’s not that the models lack capability. It’s that AI is only as useful as the context it can access — and most healthcare systems can’t deliver current, structured, connected data the moment the AI needs it. That’s not a model problem. It’s an architecture problem, and no AI bolted onto the wrong foundation will solve it.
The encounter is the real test
Most EMRs base their encounter workflows on the SOAP note — a documentation method that most medical students are taught. Unfortunately, many EMR systems have become so complex that they require extensive training for physicians to use them to do what they already know how to do. People who spent 15 years learning medicine should not need a week to learn the tool that supports it.
Today’s AI gives us the ability to completely reimagine the physician experience in an EMR. Since the beginning of computer use in medicine, physicians have asked, “Why can’t I have everything I need on a single screen?” Now they can. The computer can know what is going on with a patient, why they are here, how the physician practices, and what the physician needs in that moment. The EMR can read the chart and provide the physician the relevant information.
The note is only half the story
Ambient documentation has been getting a lot of attention in healthcare AI, for good reason. It dramatically reduces the time physicians spend on documentation, which allows them to spend more face time with patients. But the note is only one part of the encounter. After the note comes the administrivia — orders, referrals, e-prescriptions, and billing. That is the work that buries physicians and slows down their revenue cycle. We built our own ambient system because we saw a broader opportunity to revolutionize the entire encounter — from prep all the way through billing.
Interoperability tells the same story. We spent years learning to exchange data between systems and forgot to ask what happens once it arrives. A doctor gets hundreds of pages of discharge summaries from another system, the software says, “Have a look,” yet the encounter is 15 minutes long. Those numbers do not add up. AI can sift through the information, surface what matters, and present it to the physician — much like a really good medical assistant who had the time to read through everything first.
The hidden workflows matter, too
Healthcare remains addicted to faxing, for reasons we do not need to unpack here. Across the athenahealth network, we receive and process roughly 20 million faxed pages every month. Faxes have the annoying feature of taking well-formatted, discrete information and turning it into an image, with historically no way to tell a critical lab result from a roofing coupon without reading it. Roofers, by the way, are some of the biggest fax spammers in healthcare.
We are using LLMs to read those faxes, categorize and label them, pull the data out, make the content searchable, and send the roofing coupons to the spam folder — sorry, roofers. In many cases, nobody needs to look at the actual fax.
Revenue cycle management runs on the same principle: follow these rules, check these boxes, route this claim — in different ways for each of the more than 1,000 payers in the U.S. We submit more than 327 million claims annually on behalf of the physicians and practices we serve. AI is very good at following directives, and in many cases better than humans, because it does not skip steps and does not get tired at 4 p.m. on a Friday.
What SaaSpocalypse gets right
The SaaSpocalypse conversation may prove overblown, at least when it comes to whether AI will collapse per-seat business models. But the underlying question is right: Which companies can make the shift from delivering software that asks humans to do the computer work to software that does more of that work for them?
If you ask physicians about their experience with EMRs, most EMRs receive negative Net Promoter Scores (NPS). Apple and Uber have scores of 50, 60, 70, 80. With AI, I believe physicians will soon view the EMR as the best thing that has happened to medicine, which is about as far from the historical view as you can get.
At athenahealth, that belief shapes everything we build. Unlike the old model — here is your toolset, good luck — we are accountable for how the system performs alongside the people using it every day.
My advice to health system leaders: Push your vendors past the demo. Ask whether the AI can act on its recommendation or only surface it. Ask whether it sees current data or a stale copy. Ask where the physician goes next after the recommendation appears. Ask how the system performs when the workflow breaks from the expected path, and ask who is accountable for outcomes, not just for supplying software. The answers will tell you whether you’re looking at the future or a feature update dressed up as one.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
