5 areas of healthcare that are hot now because of AI

Before we get started on the analysis, I want to call out that on September 8 at 1PM ET, I’ll be teaming up with the brilliant Blake Madden from Hospitalogy to host a private webinar AMA on the topic of how the media landscape is changing, how to pitch independent media, and what newsletters specifically are looking to report on - and why. If you’re a paid subscriber, click this form to register

And now on to our regular programming...

Healthcare will be one of the sectors most profoundly transformed by AI. Delivering care remains highly manual, labor-intensive, and fraught with inefficiencies; AI has the potential to significantly reduce costs, expand access, and improve quality. But we’re in the early innings when it comes to AI being used in real clinical settings. Still, the promise of it is already inspiring investors and entrepreneurs to think deeply about where to place their bets, so much so that AI bets attracted more than two-thirds of all venture capital dollars for the past few quarters.

Some would say we’re in a hype cycle, while others argue that AI is even undervalued, given its potential. Personally, I tend to flip flop between the two points of view, sometimes feeling that it’s the most incredible technological innovation since the printing press and on other days feeling frustrated and fatigued with all the AI-generated content that LinkedIn and other sites are now awash with.

I’m also concerned about the computing power required for all this and the impact on the environment, as well as the fabrications and the hallucinations that are already driving major medical errors. There’s so much “garbage in, garbage out” potential with this technology. Lastly, there’s the potential for massive job loss and dislocation if this technology meets its potential. No country in the world is prepared for its citizens to enter into an age of abundance, rather I fear increasing inequities. This era is already upon us. As one CEO friend put it to me recently after finishing a big layoff of engineering and customer service teams: “Yes, I feel bad. But it’s also my job to make my company more efficient.”

I make it a point to talk to people across the industry, including AI technologists, doctors, nurses, administrators, call center workers, venture capitalists, and more. I also meet with companies that are building AI for the sector, and I’m actively investing.

So here are the five biggest, near-term opportunities I see for AI in health care, inclusive of a few perspectives from operators in the field, as well as potential risks and downsides to consider.

This piece was made free through the sponsorship of our friends at Ambience. Ambience Healthcare is the leading AI platform for clinical documentation and coding – trusted by top health systems like Cleveland Clinic, UCSF Health, St. Luke’s Health System, and Houston Methodist. Read more about Ambience Healthcare’s $243M Series C round, co-led by Oak HC/FT and Andreessen Horowitz (a16z), with participation from existing investors including the OpenAI Startup Fund, Kleiner Perkins, and Optum Ventures.

1) Revenue cycle management

I put this category first for a reason. Revenue cycle is piping hot right now, which includes solutions that target payers as well as those that are provider-facing. Because it relates to the efficiency of getting paid, which clearly matters when margins are constrained, health systems are making big investments in AI to improve their revenue cycle processes. Per a recent crop of surveys, they anticipate that the technology will be mainstream in the next five years. Payers are also getting in on the action, and are already using AI to detect fraud and ensure policy compliance.

Revenue cycle is a $150 billion market that includes everything from coding, patient payment estimates, preventing dials from payers, streamlining prior authorization, and more.

Something to watch: The media is keeping close tabs on payers that are using AI to deny care, particularly if it involves leveraging technology that is rife with bias. There’s real potential for harm without a human in the loop, and an American Medical Association study indicated that more than three in five physicians are concerned that AI is increasing prior authorization denials.

Sign in or join our free newsletter mailing list to keep reading more details about how AI is going to change healthcare.

An operator’s POV: Warris Bokhari moved from a large payer into a startup that helps patients appeal denied claims. He’s running one of the few companies in the space focused on protecting the interests of patients, rather than payers or providers. As he put it: “Revenue cycle is the tax we pay for a fractured payer landscape. AI is the first real lever to shrink that tax.” Hospitals have teams of people resubmitting claims, and health plans have teams of people examining those claims looking for ways to reduce outlays. Overall, this arms race between providers and payers means that the system is spending billions of dollars to manage how care gets approved, and AI has the potential to make all this less labor intensive.

An existential risk: I’ve long wondered about whether we’re looking ahead to a possible stalemate here in the long-run. Providers are already working with revenue cycle solutions, while payers are investing in tools for prior authorization and utilization management. With humans on both sides, there’s usually some amount of just “giving up” that happens. But imagine AI agents duking it out. They could literally argue their case in perpetuity. What happens if there’s no resolution?

Companies to track: Tennr, which creates payor-ready documentation to cut denials; SuperDial, which automates payer calls; fast-growing SmarterDx in the diagnostic coding space; AI-powered payer management company Anomaly Health and Charta (automation of pre-bill reviews). The OGs of the space include companies like R1, Change Healthcare and AthenaHealth. These companies may have teams of people now doing RCM the old fashioned way, but that’s changing as we speak.

And a mention: Second Opinion is teaming up with Tennr and Superdial for a dinner on the topic of RCM and AI on September 19 in NYC. We are calling it the “Getting Paid” dinner. Reach out if you’d like to attend.

2) Clinical notetaking

I view the provider documentation burden as the “hair on fire” problem to solve with AI, given the growing burnout crisis. As such, there’s been a lot of activity in this area, with companies integrating with the leading electronic health record companies to make it as seamless as possible for providers, so they don’t need to pop open an additional window. As I’ve been exposed to some of the demos for these services, it truly does seem like magic, and I’ve been surprised by the low error rates. The clinician can focus on the interaction with the patient, and the notes are summarized within seconds. The best products in the space have all the right checks and balances built in, so that clinicians aren’t approving documentation that is inaccurate (there’s a big role for voice AI here as it’s already possible for the clinician to return to a moment during the consult and play back the recording).

Something to watch: The big question of the moment, particularly as the behemoths in the space have raised hundreds of millions of dollars in venture capital or been acquired by publicly-traded companies, is where they all go from here. In other words, how to avoid being commoditized and stuck in a downward spiral of competing on price. Doximity recently entered into the market with a free option. The next step is likely to involve billing and the broader revenue cycle management space, and the winners here are already well on their way.

An operator’s POV: Oliver Kharraz, the CEO of Zocdoc, a company that helps patients with finding doctors’ appointments online, notes that the commodity problem in AI is very real. “The company that has built-in distribution to drive adoption plus an underlying core competency advantage will win.”

An existential risk: Epic. The electronic health record giant Epic is currently playing the role of partner (both marketing and integration). But there are questions circling in the industry about whether that might change if Epic spots a bigger opportunity, and the company famously doesn’t partner or do deals. Epic has an AI team, so I wouldn’t discount its internal technology capabilities.

Companies to track: The leading players include Suki, Abridge, Nuance, and Ambiance, plus entrants started in other geographies like Nabla (founded in Europe) and Tali.ai (Canada). Commure also has a product in the space after acquiring Augmedix. Doximity recently entered the market with its free product with its CEO Jeff Tangney telling STAT News that “budgets are tight right now.”

3) Drug discovery

We’ve all been talking about the potential to use AI for drug discovery for years. It makes sense, given that more than 90% of drug targets fail before they hit the market and they’re incredibly expensive to develop. I recently read a fascinating article in Wired about why AI is so promising, given human biology is so complex. Drug discovery experts don’t believe that AI will make the role of chemists obsolete, or remove all the challenges associated with making new drugs. But there does seem to be some promise in using AI to do things like predict how proteins fold up into their final form (Google’s DeepMind is at the forefront of that). For a podcast, I recently sat down with Siddhartha Muckherjee, the Pulitzer Prize winning author and oncologist, who is starting a new company in the space with technologist Reid Hoffman called Manas AI. He described how AI is making it possible to map out the rules of how drug molecules influence their biological targets, with a goal of making new drugs in just a few years. We talked about a future where a patient could have a drug developed just for them.

Something to watch: Could AI be set loose on a database of untried or untested molecules? The software could take in all the information about biology, generated by humankind, and make new suggestions. Would these be nothingburgers that waste time and resources, or lead to promising targets? Given the stakes, scientists seem to think it’s worth figuring it out, but no one knows quite yet what the outcome will be. That said, these drugs are still going to go through the same development process as any other, so risk is limited. The big question is whether it’ll happen at a lower cost and in less time.

An operator’s POV: Shwen Gwee, formerly of Novartis and Bristol Myers Squibb, believes that R&D is both where the biggest investments have been made thus far by biopharma companies, and where the progress is. “You’ve got AI drug discovery startups that have all become TechBio companies with their own pipelines after partnering with pharma for a few years,” he says. “You’ve also got native Gen AI-driven TechBio companies that are bringing assets through clinical development.” He would put Recursion and Benevolent in the first camp, and both Generate Biomedicines and Insilico in the latter.

“Most pharma companies are investing a ton of money into building, partnering and even buying AI for discovery and design,” said Gwee, who shared that Insilico already has a strong pipeline of drug candidates in various phases of development.

Companies to track: Recursion (the TechBio OG), Vilya (cofounded by a team of renowned scientists), Xaira (CEO is a former chief scientific officer at Genentech), Leash Bio (using machine learning for medicinal chemistry), Atomwise (another one of the OGs) and Insilico (deep learning for drug discovery).

4) AI-native medical clinics

This one may be less obvious, but I am a big believer that AI will make provider businesses far more efficient and scalable to more patients —and therefore more attractive to investors. Already, I’m meeting with companies that are able to do far more with less. Clinic businesses are becoming far less capital intensive than they used to be, as no longer is it a requirement to have teams of people dealing with billing and prior authorization or scheduling. AI oriented practices are also lowering the cost of care and expanding access, because they don’t require a 30m video or in-person visit for every interaction, whether it’s warranted or not. They also don’t have large teams of people doing highly manual operational work. All of that in theory makes the margin profile of these clinics more attractive over time.

Something to watch: There is such a thing as too much automation, and too few humans in the loop. I’ve had a few of these experiences recently where I had to deal with so many steps of automation, and just wanted to reach a human to have a conversation. Sometimes the AI is simply missing context. One example? I received an automated-looking note recently from a clinic suggesting a visit to LabCorp on a Sunday for bloodwork, when the vast majority of locations are closed (and in theory, any human would know that). Agentic AI, though, would be a game changer, particularly if patients can use it to find clinics in network, take care of scheduling, and more. But near-term, I’m more bullish on the prospect of using AI for backend functions, versus anything that significantly touches the patient including AI-aided diagnosis.

An operator’s POV: Given that so many companies now claim to be using AI, Kevin Wang, a growth leader at Sword Health, believes it can be hard to tell whether it’s having a real impact on the business. He asks: “Is it having a meaningful impact on the business’s unit economics?” For instance, can the company scale a single expensive resource like a clinician to significantly more lives and people. Secondly, is the AI being used in a way that’s patient facing to improve engagement and outcomes. If a virtual care provider is using a few tools, like a Nabla, isn’t super interesting to him. But if a company has AI decision support or messaging built in so a 5-minute task for a human only takes 5 seconds, then it starts looking interesting.

NOCD’s Stephen Smith said it’s common for clinic businesses to use AI for provider-specific tasks, like note taking and chart auditing. “We’re in the early innings,” he said. “In the near future, AI will also effectively optimize costly backend processes that sit outside of care delivery.” In his view, that will reduce the amount of fixed cost needed to manage a clinic business, and unlock resources that could be reinvested back into patient care or in R&D.

Companies to track: There are countless companies to name, and I’m sure I’ll offend someone if I list some but not others. So I’ll reference those I’ve spoken to most recently about this topic. LunaJoy in the women’s health space, NOCD, in behavioral health, Sword Health in MSK. Slingshot, which is building an AI therapist. Most virtual clinic businesses that I’m aware of are using AI, so it’s really a question of getting into the finer details (I agree with Kevin here). How much, what kind, and what impacts are really worth the investment?

5) Radiology / medical device

I probably could have listed this one first. Radiology and pathology have long been considered the two specialties that will be most impacted by AI by physicians. So it’s entirely unsurprising that AI is already making a meaningful difference in the field. I’ve spoken with specialists and radiologists who are already using it to help with imaging and triaging because of AI’s pattern matching skills. There are also studies suggesting AI could be useful in patient education, related to imaging. It could help answer questions and provide context in medical charts. The big question I’m grappling with is related to incentives. The companies I’m most excited about in the space are thinking about how to partner with radiologists, versus replace them. Benchmark-backed New Lantern, for instance, is focused on reducing busywork so radiologists can read more scans (that’s how they make money). Bottom line: It has got to be mission critical, or it won’t be adopted.

Something to watch: Regulations are very much still evolving to regulate AI in radiology as a medical device, either through the 510(k) or PMA pathway based on risk and novelty. It’ll be crucial in the future to see radiologists, data scientists and regulators collaborate to determine the future of this space. There’s still potential for fabrications and inaccuracies, so development needs to happen in a way that builds trust and transparency on all sides.

An operator’s POV: Dr. Panagis Galiatsatos, a pulmonologist and critical care medicine physician with the John’s Hopkins School of Medicine, said AI is already being used in his practice with dedicated breast imaging to better assess abnormal findings as benign versus pathological. That saves patients from both anxiety and stress, as well as tests and procedures they don’t need. He said it’s still learning and developing its pattern recognition but he already views AI as a “great complementary tool to radiology reading in real time.”

“With all of AI’s potential, that is likely the closest to finding its way into medical practice in the near future,” he said.

Companies to track: New Lantern, maker of the AI radiology resident; Rad AI, which is focused on radiologist workflow; Subtle Medical, which improves the quality of medical imaging; and AI-powered care coordination company Viz AI.

Want to support Second Opinion?

• 🌟 Leave a review for the Second Opinion Podcast

• 📧 Share this email with other friends in the healthcare space!

• 💵 Become a paid subscriber!

• 📢 Become a sponsor