What’s still open in healthcare AI?
In our healthcare AI market deck, you will find everything you need to understand the market
SUMMARY
What’s still open in healthcare AI? Specialty operating systems, prior authorization, nursing workflow automation, end-to-end patient access, payer operations and AI governance still leave meaningful room for new companies.
The market is no longer early in the broad sense. Domain-specific healthcare AI adoption has already reached 22% in Menlo Ventures’ survey, but healthcare still spends roughly $740 billion a year on administration, so the bigger opportunity is replacing expensive work rather than inventing another AI interface.
Ambient scribing proved that clinicians will pay for AI, but it also showed how quickly a breakout feature can become a platform feature. The remaining value is moving downstream from the note into coding, orders, authorization, follow-up and specialty workflows.
Clinical search has a similar problem. OpenEvidence already has strong physician habit and distribution, while frontier-model companies are improving medical reasoning and retrieval. A new entrant needs to own a specialty decision or workflow, not simply answer clinical questions better.
Prior authorization remains unusually attractive because digitization is improving the plumbing without removing the hard work. Finding the right policy, assembling evidence, fixing missing information, tracking responses and handling appeals can still consume entire teams.
Patient access is moving beyond voice. Simple appointment-booking agents will become cheaper and easier to reproduce; the stronger companies will own the full path from referral or inbound request to eligibility, scheduling, authorization, outreach and completed care.
Nursing may be earlier than physician documentation. Hospitals are starting to prove that AI can cut charting time sharply, but much of the shift-level work around flowsheets, handoffs, care plans, discharge and task coordination is still barely automated.
Payer AI has plenty of whitespace, but the economics come with a nasty sales-cycle problem. Low penetration makes insurers attractive, yet an average buying cycle measured in many months means the market favors well-capitalized teams that can survive slow procurement.
Specialty healthcare AI operating systems look especially strong because they combine clinical rules, payer behavior, scheduling, medications, documents and local edge cases. That creates a better chance to build proprietary workflow data and real switching costs than a generic assistant ever will.
Imaging and governance show the same broader pattern: the model itself is becoming less scarce, while deployment, orchestration, monitoring and control remain messy. The startup opportunity increasingly sits around making AI work reliably inside the institution.
The most crowded ideas are now easy to name: another standalone ambient scribe, another generic clinical search tool, another consumer health chatbot and another isolated imaging algorithm. Those markets can still produce large companies, but a new entrant starts with a distribution disadvantage.
If we were starting a healthcare AI company today, we would pick one expensive workflow inside one specialty, automate the job end to end and tie the product to a measurable outcome. Once the software owns enough of the workflow, better models help rather than commoditize it.
This market map, featured in our healthcare AI market deck, highlights top companies and startups in the healthcare AI market
Is healthcare AI already too crowded for a new startup?
Healthcare AI is crowded today, but a new startup still has plenty of room if it attacks work that hospitals, clinics and insurers still pay people to do manually.
The market has clearly moved past the experimental phase. Menlo Ventures surveyed more than 700 healthcare executives and found that 22% of healthcare organizations had already deployed domain-specific commercial AI, up from roughly 3% two years earlier. Healthcare-specific AI spending had reached about $1.4 billion. Health systems alone accounted for roughly $1 billion of it.
Investors have followed. Rock Health's latest quarterly review counted $4.0 billion invested in U.S. digital health across 110 deals, compared with $3.0 billion across 122 deals a year earlier. The more revealing number is concentration: just 12 mega-rounds absorbed 59% of all the money. These days, getting funded in healthcare AI and finding an open market are two very different things.
The room comes from the mismatch between software spending and labor spending. Menlo estimates that healthcare still spends roughly $740 billion a year on administration. Entire workflows around patient access, prior authorization, billing, referrals, nursing documentation and payer operations remain heavily dependent on people moving information between EHRs, portals, faxes, phones and spreadsheets.
So when we call a healthcare AI market "open," we are looking for three things at once: a large amount of work that still costs real money, a workflow that today's products have not solved well enough, and a path for a startup to become harder to replace over time. Another chatbot that happens to understand medicine does not clear that bar.
| What we see today | Evidence | What it tells us |
|---|---|---|
| Domain-specific healthcare AI adoption | 22% | AI buying is already real |
| Healthcare-specific AI spending | ~$1.4B | Customers are paying, not just piloting |
| Latest quarterly U.S. digital-health funding | $4.0B across 110 deals | Capital is available |
| Share of that funding captured by 12 mega-rounds | 59% | Money is concentrating around perceived winners |
| Annual healthcare administrative spending | ~$740B | Current AI revenue is still tiny beside the work being attacked |
Is ambient AI scribing still open?
Standalone ambient AI scribing is largely claimed now, and a new company needs a much stronger wedge than slightly better note-taking.
Ambient documentation became healthcare AI's first obvious breakout category. Menlo estimated roughly $600 million in annual spending, with Microsoft Nuance DAX Copilot, Abridge and Ambience together accounting for about three-quarters of the market in its survey. That is already meaningful concentration for a category that barely existed a few years ago.
Customer behavior creates another problem. Menlo found unusually high willingness to switch ambient vendors among outpatient buyers. Doctors clearly value ambient documentation, yet attachment to a particular transcription engine appears much weaker. Better foundation models should keep narrowing basic quality differences.
The biggest pressure is now coming from inside the EHR. Oracle Health recently expanded its Clinical AI Agent beyond note generation into professional-fee coding, chart review and dictation. Oracle says its note-generation product has already saved physicians more than 400,000 hours. Epic is also pushing AI deeper into clinical workflows rather than treating documentation as a standalone feature.
Abridge, Ambience and Commure can still become very large companies because they are moving beyond the note itself. Abridge is pushing further into clinical workflows, while Commure has combined ambient AI with coding, revenue cycle and other operational products. That expansion shows where the remaining opportunity sits.
We would be very cautious about launching another pure scribe today. The better opening is to use the clinical conversation as the starting point for coding, orders, prior authorization, follow-up, patient instructions or specialty-specific work that happens after the visit.
| Ambient AI question | Our read now |
|---|---|
| Do doctors want the product? | Clearly yes |
| Is the category still growing? | Yes |
| Is basic transcription hard to reproduce? | Increasingly no |
| Are major EHR vendors moving into it? | Aggressively |
| Is there room for another standalone scribe? | Limited |
| Is there room beyond the note? | Much more |
If you want more recent data on this point, please see our latest healthcare AI market report.
As this chart shows, and as featured in our healthcare AI market deck, search interest in healthcare AI has grown rapidly
Can a new clinical AI search company still beat OpenEvidence?
A generic clinical AI search startup would be entering one of healthcare AI's hardest markets today because OpenEvidence already has physician habit, distribution and scale.
OpenEvidence was valued at $12 billion after raising another $250 million earlier this year. At the time, the company said more than 40% of U.S. physicians used the product on an average day across more than 10,000 hospitals and medical centers. Monthly clinical consultations had gone from roughly three million to 18 million in a year.
Usage has kept moving. OpenEvidence later reported crossing one million clinical consultations in a single day. More recently, an independent Stanford-Harvard clinical AI study allowed practicing physicians to choose outside AI tools while evaluating clinical decisions. According to the study results released by OpenEvidence, physicians voluntarily chose OpenEvidence more often than ChatGPT, Claude, Gemini and all other external AI chatbots combined.
That is a much stronger position than having the highest benchmark score. OpenEvidence is becoming a habit inside a profession where trusted habits are difficult to displace.
Frontier-model companies add another layer of pressure. OpenAI and Anthropic are both pushing further into health, making high-quality medical reasoning, retrieval and summarization easier to access without buying a separate point solution.
There is still room around clinical knowledge, but we would go much narrower. Oncology treatment selection, fertility protocols, complex cardiology, transplant medicine or another specialty can require evidence retrieval plus patient-specific data, payer rules and operational action. In that kind of product, search becomes one piece of the workflow.
Trying to build "better Google for doctors" now looks late. Building software that helps a particular specialist make and execute a difficult decision still looks much more open.
Is healthcare revenue-cycle AI still open?
Healthcare revenue-cycle AI is still open, but the broad horizontal market is getting tougher as Commure, Oracle and other platforms spread across coding, claims and collections.
The money is certainly there. Menlo estimated about $450 million in healthcare AI spending already goes toward coding and billing automation. Buyers can measure the result in dollars through better coding, fewer denials, faster collections and lower administrative headcount, which helps explain why revenue-cycle AI has attracted so many companies.
Commure shows how quickly the leaders are scaling. The company recently raised $70 million at a $7 billion post-money valuation. Commure says it now works with more than 500 healthcare organizations across over 3,000 sites, integrates with more than 60 EHRs, processes over $25 billion in annual claims and handles more than 100 million patient interactions a year.
Commure is also expanding sideways. Its products now cover ambient documentation, coding, call-center agents, referrals, intake and end-to-end revenue cycle. Oracle's recent move into AI-assisted professional-fee coding adds another incumbent directly inside the EHR.
That makes a generic "autonomous RCM platform" harder to start from zero. A new company needs somewhere the broad platforms are still weak.
We see better openings in specialty coding, surgical billing, pre-claim revenue integrity, contract interpretation, underpayment detection and denial prevention. These problems depend on clinical nuance, payer behavior or specialty rules that generic automation does not always handle well.
The strongest wedge is usually a very specific revenue leak. If a startup can find money the provider is currently losing, prove that it can recover it and then expand around the claim, healthcare organizations have a clear reason to buy.
If you want more recent data on this point, please see our latest healthcare AI market report.
This chart, featured in our healthcare AI market deck, shows annual VC investment in healthcare AI startups
Is prior authorization still one of the best healthcare AI opportunities?
Yes. Prior authorization is still one of the best healthcare AI opportunities because the workload is huge, the pain is measurable and the workflow is still far more manual than it should be.
The American Medical Association surveyed 1,000 physicians and found that doctors and their staff handled an average of 40 prior authorizations every week. That work took about 13 hours of physician and staff time. Forty percent of physicians had employees dedicated exclusively to prior authorization, while 32% said requests were often or always denied.
The problem has also been getting worse rather than quietly disappearing. In the same AMA survey, 74% of physicians said prior-authorization denials had increased over the previous five years, and 95% said the process delayed necessary care.
AI spending is moving quickly into the category. Menlo estimated that prior-authorization AI had already passed $100 million in annual spending after growing roughly tenfold year over year.
Government infrastructure should make some parts easier. CMS is forcing more standardized electronic prior-authorization processes, including faster response deadlines and FHIR-based APIs for requirements, submissions and responses. As that plumbing improves, simply checking whether authorization is required will become a weaker business.
The hard part remains wide open. Someone still has to identify the right policy, find the relevant evidence in the chart, recognize what is missing, prepare the submission, watch for a response, answer requests for more information and build an appeal when the request is denied.
The opportunity looks especially good in oncology, rheumatology, gastroenterology, infusion medicine and other specialties where expensive treatments meet complicated payer rules. A company that reliably gets treatment approved is solving a much bigger problem than one that only tells the staff which form to complete.
Is healthcare patient-access AI still open, or is voice already crowded?
Healthcare patient-access AI is still open, although simply answering the phone with a voice agent is getting crowded fast.
Patient access is enormous. Menlo estimates that healthcare spends more than $100 billion a year on patient access and engagement while software captures only around 5% of that spending. AI products in the category were already growing roughly twentyfold year over year in its survey.
Assort Health shows how quickly the first layer has matured. The company recently raised $120 million at a $1.2 billion valuation and says revenue grew twentyfold over 15 months. Its platform has handled more than 190 million patient interactions across 62,000 care protocols and now stretches across scheduling, intake, medication requests, payments, outreach and referrals.
Assort's latest referral product is a useful example of where the category is moving. The company says its agent now processes inbound referrals, checks eligibility, identifies missing information, finds the right provider, contacts the patient and books the appointment inside the EHR. In its reported deployments, 79% of referrals converted into scheduled appointments, and the system found a better-fit provider in about half of cases.
Other companies are heading in the same direction. Tennr began with referral documents and has pushed into benefits checks, authorization work, missing-document follow-up and automated phone calls. Commure recently launched its own end-to-end referral and intake product.
Basic voice technology will keep getting cheaper. A phone bot that can understand speech, answer common questions and book a simple appointment is unlikely to remain special for long.
The bigger prize is owning the whole path from "I need care" to "the right care actually happened." Scheduling, eligibility, intake, referrals, authorization, outreach and rescheduling all sit around the same patient journey. We think patient access stays very open for companies that can close those loops rather than just handle conversations.
If you want more recent data on this point, please see our latest healthcare AI market report.
This chart, featured in our healthcare AI market deck, looks at Tempus AI’s strategy in healthcare AI
Is payer AI still early enough for a new startup?
Payer AI is still unusually open today, although selling to health insurers is slow enough to kill a startup that underestimates the market.
Menlo's survey found domain-specific AI adoption at only 14% among payers, compared with 27% at health systems. Payers represented roughly $50 million of healthcare-specific AI spending in the survey, a tiny amount beside the size of U.S. insurance administration.
New companies are clearly finding room. Anterior raised another $40 million earlier this year, bringing its total funding to $64 million, to expand AI across health-plan clinical and administrative workflows. Cohere Health has moved well beyond prior authorization into appeals, payment accuracy, claims operations, quality and care management.
Provider automation is starting to put pressure on the other side. Providers can use AI to code faster, submit more complete authorizations, identify underpayments and appeal denials at much higher volume. Health plans eventually need their own automation to process those transactions.
The catch is procurement. Menlo found an average AI buying cycle of about 11.3 months among payers, compared with 6.6 months at health systems and 4.7 months among outpatient providers. A payer sale therefore took more than twice as long as an outpatient sale.
For a well-capitalized company willing to spend years inside complicated workflows, payer AI still has a lot of empty space. We especially like medical-necessity review, appeals, payment integrity, policy digitization and care-management operations. For a small team that needs fast sales and instant product feedback, we would start elsewhere.
Is nursing AI still open?
Nursing AI is one of the most interesting healthcare AI markets right now because nurses still spend huge amounts of time documenting care, while the product category is only beginning to take shape.
Nursing work is much more continuous than a physician visit. Nurses enter observations, medications, assessments, interventions, care-plan changes and handoff information throughout a shift. Research on EHR use has found hundreds of flowsheet entries during a single twelve-hour shift, which gives AI many more potential touchpoints than simply drafting one note.
There is already evidence that automation can save meaningful time. Epic says nurses at Mercy using its AI to draft end-of-shift care-plan notes cut average documentation time from 3.5 minutes to about 32 seconds, an 85% reduction. The number of notes completed fully and on time rose 225%.
The category is moving quickly. Mount Sinai Medical Center recently started using Epic's ambient nursing documentation, which turns bedside conversations into draft charting. Microsoft Dragon Copilot is also being built around nursing flowsheets and clinical information retrieval. Separate hospital research is now testing ambient nursing AI directly inside Epic's Rover mobile workflow.
Those incumbent moves make another generic nursing scribe less compelling. The larger opening is the shift itself: flowsheet documentation, handoffs, discharge preparation, care-plan updates, missing tasks and the constant search for relevant information across the chart.
Nurses coordinate a large share of what actually happens to a patient inside a hospital. Software that understands that workflow and removes repetitive EHR work could become much bigger than a transcription product.
This chart, featured in our healthcare AI market deck, shows annual funding in healthcare AI startups
Are specialty healthcare AI operating systems still open?
Specialty healthcare AI operating systems may be the best place left to build because difficult medical workflows still vary enormously between oncology, fertility, rheumatology, cardiology and other specialties.
The attraction is simple. General models keep getting better, which makes generic healthcare features easier to copy. Specialty workflows remain much harder because they combine clinical knowledge with local operating rules, payer requirements, scheduling constraints, medications, documents and repeated edge cases.
Oncology already shows the pattern. Companies such as RISA Labs are automating patient-access and authorization work around cancer treatment. Triomics has built AI workflows around oncology care and research and has worked with organizations such as Memorial Sloan Kettering, MD Anderson, Yale and Mount Sinai.
Oncology is particularly difficult because a single patient's journey can involve pathology, genomic biomarkers, changing treatment guidelines, prior authorization, specialty pharmacy, infusion scheduling, clinical trials and repeated imaging. A product that understands how those pieces interact learns much more than an assistant that simply summarizes the chart.
The same idea works elsewhere. Fertility clinics have protocols, cycles, medications and patient communication that look very different from cardiology. Rheumatology has biologics and complicated authorization. Gastroenterology has procedure preparation, pathology, chronic disease and infusion workflows. Transplant medicine has an entirely different chain of eligibility, testing, matching and follow-up.
This is also where we think healthcare AI can still build real defenses. Every completed workflow produces specialty-specific data about edge cases, payer behavior, patient behavior and operating rules. Integrations become deeper. Outcomes can be measured within that specialty. Customers become much more reluctant to replace a system that understands how their practice actually works.
A new company does not need to become the AI layer for all of healthcare. Owning one expensive, complicated specialty end to end can be enough to build a very large business.
If you want more recent data on this point, please see our latest healthcare AI market report.
Is medical-imaging AI still open?
Medical-imaging AI is still open around deployment and workflow, but another standalone radiology-detection algorithm is a much weaker startup idea these days.
Radiology has dominated regulated medical AI for years. By the time the FDA's list of authorized AI-enabled medical devices passed well beyond 1,000 products, radiology represented the large majority of them. Hospitals no longer face a shortage of algorithms that claim to detect abnormalities.
The harder question is what happens after a model exists. Health systems have to integrate it into PACS and the EHR, decide which patients should trigger it, monitor performance, manage several vendors, prove that it improves care and make sure the result reaches the right clinician at the right time.
Real-world evidence has often lagged regulatory clearance. Reviews of authorized imaging AI products have found that prospective clinical testing remains much less common than the number of approved products might suggest.
That helps explain why companies such as Aidoc have expanded toward broader clinical AI platforms rather than remaining collections of isolated algorithms. Other vendors are building orchestration layers that let hospitals deploy, compare and manage multiple imaging models.
We still see room in imaging infrastructure, model monitoring, multimodal workflows and underdeveloped specialties. There is also room when the AI can connect an imaging finding directly to triage, follow-up or treatment.
The crowded part is the isolated prediction. The workflow around that prediction is much less finished.
This chart, featured in our healthcare AI market deck, compares the main business model options for ambient AI companies
Is healthcare AI governance becoming a real startup market?
Healthcare AI governance is becoming a real market now because hospitals are starting to manage many AI systems at once rather than a handful of pilots.
The problem changes once a health system deploys dozens of models and agents. Someone needs to know which models are running, what patient information they can access, how they were evaluated, whether performance has drifted, who approved each use case and what happened when a clinician overrode the output.
Investors are already treating that layer seriously. Qualified Health raised a $125 million Series B earlier this year and works with health systems including Mercy, Emory Healthcare, University of Rochester Medicine, Jefferson Health and the University of Texas System. Its platform covers evaluation, deployment, governance, access control and post-deployment monitoring.
The size of that financing does not prove the category will stay independent. Epic, Oracle, Microsoft, cloud providers and security vendors can all bundle pieces of AI governance into products customers already buy.
A generic inventory dashboard therefore looks weak to us. The stronger opportunity is a control layer that sits in the actual path of AI deployment: test a model on local data, decide what the model may access, monitor its behavior, compare models, keep an audit trail and connect model performance to clinical or operational outcomes.
Hospitals are moving from "Can we use AI here?" toward "How do we control all the AI we are already using?" That second question can support a serious infrastructure market.
Can a new consumer health AI company still break through?
A new consumer health AI company can still break through, but generic medical question-and-answer products are being swallowed by platforms with vastly larger distribution.
Consumer behavior has moved very quickly. Rock Health's latest consumer research found that roughly one in three people now use AI chatbots for health questions, about twice the rate seen a year earlier.
That sounds great for startups until we look at where those questions are going. ChatGPT already reaches hundreds of millions of people and has built a dedicated health experience. Anthropic is pushing Claude further into healthcare as well. Consumers do not need to download a new app to ask a model why their knee hurts or what a blood-test result might mean.
The startup opportunity starts when the answer needs to become an action. Insurance navigation, appointment booking, medical-record collection, specialty care, longitudinal monitoring, diagnostics, prescriptions and clinician access all require more than a good model response.
Distribution can also come from owning the service behind the AI. A fertility company, chronic-care provider, diagnostic platform or specialty clinic can use AI as the interface while keeping a provider network, clinical protocol, data asset or care pathway underneath.
We would avoid "AI doctor in your pocket" if the actual product is only chat. Consumer healthcare still has plenty of room when the company controls what happens after the conversation.
This chart, featured in our healthcare AI market deck, illustrates how revenue is distributed across customer segments in the healthcare AI market
Can an autonomous AI doctor become a real company now?
An autonomous AI doctor could eventually become enormous, but right now it is one of the hardest ways to build a healthcare AI startup.
The technical direction is becoming more believable. Frontier models can already collect histories, reason through differential diagnoses, interpret medical literature, summarize longitudinal records and work with images and lab data. Companies such as Lotus Health are already testing AI-led primary-care models that combine automated conversations with diagnosis, prescriptions and referrals under clinical oversight.
Commercially, however, autonomous medicine stacks nearly every difficult healthcare problem on top of the same product. Diagnosis can create direct patient harm. Prescribing requires clinical responsibility. Medical practice is regulated. Reimbursement still revolves around existing providers and organizations. Patients also need to trust the system enough to act on its advice.
The current behavior of the biggest AI companies is informative. Their health products are still framed around assistance, information and clinician support rather than unrestricted autonomous diagnosis and treatment.
That boundary will probably move. We just do not know how quickly regulation, liability, reimbursement and consumer trust will move with the models.
For founders who want a high-probability business now, administrative and clinical workflow automation looks much better. For someone deliberately taking a ten-year bet on what healthcare delivery itself could become, autonomous care remains one of the biggest open frontiers.
If you want more recent data on this point, please see our latest healthcare AI market report.
So what’s actually still open in healthcare AI?
Healthcare AI is still very open, but the best opportunities now sit in difficult workflows rather than obvious AI features.
We would put specialty healthcare operating systems near the top. They combine large budgets with workflows that general-purpose software struggles to understand, and they give a startup a realistic way to accumulate proprietary operating data.
Prior authorization is also unusually attractive. The workload remains huge, the ROI is visible and digitization should make automation easier without removing the complicated clinical work that creates room for specialized companies.
Nursing is earlier. The latest deployments show that hospitals can save real documentation time, while a large part of the broader nursing workflow remains untouched. Patient access is further along, but the underlying labor pool is so large that there is still room for companies that own scheduling, referrals, eligibility and outreach together.
Payer AI has enormous potential and relatively low current penetration, although the sales cycle makes it a harder company to build. Healthcare AI governance also looks increasingly real as hospitals move from isolated pilots to fleets of models and agents.
We are much less excited by standalone ambient scribing, generic clinical search, isolated imaging algorithms and generic consumer health chat. Those markets can still produce big winners, but a new entrant is now fighting established healthcare AI leaders, EHR vendors or frontier-model companies from day one.
Revenue cycle sits somewhere in the middle. The overall market is excellent, yet horizontal competition is already intense. We would enter through a narrow problem such as specialty coding, revenue integrity or denial prevention rather than trying to build another broad RCM platform immediately.
The pattern across the best opportunities is pretty consistent. Healthcare still pays a huge number of people to move information from one system to another, understand what that information means, decide what has to happen next and chase the process until it gets finished. AI is finally capable of taking over much more of that work.
If we were building in healthcare AI today, we would start with one expensive workflow inside one specialty, automate the full job rather than one screen, and make the outcome measurable. Once the software owns enough of the workflow, better models become an advantage instead of a threat.
That is where healthcare AI still has the most room.
| Healthcare AI opportunity | How open is it now? | Why we like or dislike it |
|---|---|---|
| Specialty healthcare AI operating systems | Very open | Deep workflows, strong expansion paths, real data advantage |
| Prior authorization | Very open | Huge manual burden and measurable ROI |
| Nursing workflow automation | Very open | Early category with enormous workflow volume |
| End-to-end patient access | Very open | Massive labor budget still barely converted into software |
| Payer operations | Open | Low AI penetration and very large administrative budgets |
| AI governance and control infrastructure | Open | Model proliferation is creating a new software layer |
| Specialty revenue-cycle automation | Open | Strong economics if the wedge is narrow and difficult |
| Imaging workflow and orchestration | Selectively open | Algorithms are crowded, deployment remains messy |
| Autonomous AI care | Wide open but early | Enormous upside with major regulatory and trust risk |
| Standalone ambient scribe | Mostly claimed | Strong incumbents and growing EHR bundling |
| Generic clinical AI search | Mostly claimed | OpenEvidence already has extraordinary physician adoption |
| Generic consumer health chatbot | Mostly claimed | Frontier-model companies own the distribution advantage |
| Another isolated imaging algorithm | Crowded | Too many point solutions and weak workflow ownership |
This chart, featured in our healthcare AI market deck, shows how symptom checker app technology has evolved over time
OUR METHODOLOGY
This analysis asks where a new healthcare AI company still has meaningful room to build today. We break the market into distinct opportunity areas and judge each one using recent, observable evidence rather than treating healthcare AI as a single category.
Depending on the market, we look at current AI adoption and spending, remaining manual workload, customer behavior, incumbent positioning, product expansion, funding concentration, regulatory change, deployment evidence and how much of the underlying workflow existing products already control.
We prioritize recent evidence because the competitive landscape is moving quickly. A market that looked early two years ago can now have established leaders, embedded EHR distribution or platform competition. At the same time, mature parts of healthcare can still contain large amounts of operational work that software has barely touched.
No single metric determines whether a category is open. Funding can show conviction without showing whitespace; rapid adoption can validate demand while making entry harder; and strong model performance does not mean regulation, procurement or workflow integration will move at the same speed. The labels from “very open” to “crowded” are therefore editorial syntheses of several signals, not outputs from a fixed scoring formula.
We give particular weight to workflow ownership. A product becomes more interesting when it can take responsibility for an expensive job from start to finish, learn from specialty-specific edge cases, integrate deeply with the systems around it and tie its value to a measurable operational or financial outcome.
We favored recent first-hand disclosures, regulatory sources, healthcare organizations and established industry research over secondary commentary whenever the underlying evidence was available directly.
Key sources used for the analysis include Menlo Ventures on healthcare AI adoption, spending, administrative expenditure and buyer behavior, Rock Health on Q1 2026 digital-health funding and funding concentration, Rock Health on consumer use of AI for health questions, the American Medical Association on prior-authorization workload and denials, and CMS on electronic prior-authorization requirements and FHIR-based infrastructure.
We also used Oracle on the expansion of Clinical AI Agent, Epic on nursing documentation results at Mercy, Microsoft on Dragon Copilot for nursing workflows, Commure on referral, intake and revenue-cycle expansion, Assort Health on patient-access scale and funding, and Anterior on payer-AI expansion.
For regulated medical AI and platform pressure, we used the FDA’s list of AI-enabled medical devices, Qualified Health on healthcare AI governance infrastructure, OpenAI on Health in ChatGPT, Anthropic on its healthcare and life-sciences products, and OpenEvidence on reaching one million clinical consultations in a day.
In our healthcare AI market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- Who are the top investors in healthcare AI?
