A recent conversation on the podcast The Business of a Clinic, hosted by Jared Aron, co-founder of Coherent Healthcare, ended up ranging across most of how we look at this market. For that reason, it felt like a good moment to set our digital health investment thesis out properly. That includes the categories we have deliberately walked away from and the ones we are holding off on for now.
Roughly 80% of the fund is committed to digital health, a concentration we chose about three years ago. It only makes sense if the thesis behind it keeps moving. Every six to twelve months we take the whole thing apart and ask which categories deserve to come in. Others come out, either because the opportunity has closed or because the timing is wrong for us. The second case is not a rejection, and we expect to look at those again a year later.
What We Stopped Looking At, and What That Taught Us
The AI interpretation layer on top of personal health data is the cleanest illustration of how a category closes. Two years ago, we met startups building exactly that, and almost every player has since shipped its own version. Several are very well-funded, while Oura and Whoop have built an answer that works well enough on its own. Whether we were late or the category simply stopped being defensible, the conclusion was the same. It is no longer part of the thesis.
Hormone tracking followed a comparable path over a shorter horizon. We saw a considerable amount of it this year and found the field splits in two. On one side, very well-funded players we had already been tracking for a couple of years. On the other, teams still working through technological development. That split suggests to us that the underlying technology is probably not yet there, and we believe this is yet to come from leading laboratories and institutions before it reaches mass market. We have paused on pure hormone tracking and expect to reconnect with the category next year, which is a timing call rather than a verdict on the space.
Demand was never the problem in either category. In the first case we saw no clear path for new companies to build something incumbents would struggle to copy. In the second case we believe further development is needed for the underlying technology.
Where Our Digital Health Investment Thesis Sits Today
Our most recent deep dive left us with three areas carrying the weight, all resting on AI and data.
The first is the safety layer around implementing AI in healthcare, which matters across both public and private settings. That said, our attention sits mainly on the private side. In practice, most of what we are underwriting here is compliance rather than security, and we are gradually widening the frame toward adjacent categories such as regulated software as they mature.
The second, and our core focus, is AI-driven operating efficiency inside private clinics. We treat it at length below, because the model a company chooses there determines almost everything about whether it lasts.
The third is preventive solutions built on top of the payer side, where we are particularly interested in companies working alongside reimbursement paths, since this is the mechanism through which a lot of otherwise stranded innovation eventually reaches patients. Similarly, hybrid models combining prevention with care delivery have held their place through every review we have run.
Alongside those, a handful of categories have earned their way into the thesis during the year rather than being planned. Lab automation on the pharma side was not something we initially read as an opportunity and now is. Health fintech has always interested us, and with margins compressing across the sector, the timing looks better than it did.
New regulated software and clinical pathways like mental health and sleep have moved up for the same underlying reason. Neither has produced a long-term winner. In mental health you can make the case that companies like Headspace have already established a category leader. Even so, there’s plenty of room for new companies going after new clinical pathways and deeper operational value. A clear example is the wave of companies using voice AI agents to detect whether a pathology is present. In sleep the same applies to sleep apnea, where something better should already exist and does not.
Horizontal, Vertical, or Service
Clinic efficiency is the area where we have spent the most time and, in our reading, the corner of healthcare most genuinely prepared to absorb new technology. Our view on how to build there has shifted with what we have seen.
The Limits of a Horizontal Product
The early cohort we met was verticalized, with products built for a single clinic type such as dental, whereas most of what reaches us now is horizontal, highly homogeneous, and sold across dental, MSK, and several other practice types at once. The sheer number of providers tells you the need is real, and it also tells you that defensibility is thin, which is a difficult combination to underwrite at pre-seed.
The structural problem with a purely horizontal product is that clinics only overlap so far. Perhaps 30-40% of what they do resembles what the clinic next door does, and patient intake sits inside that overlap. Move toward the clinical work, though, and one practice has very little in common with another. A horizontal tool therefore ticks a box during procurement and reveals its real costs later. Usually somebody internally acquires the job of making it work, which adds headcount. The efficiency the clinic thought it was buying quietly cancels out. When the ROI does not appear, the problem is rarely the technology on its own.
The Limits of Going Fully Vertical
Going fully vertical is a stronger position, though it carries a different ceiling. Expanding into new geographies or neighboring sub-types of clinic within the same vertical is genuinely hard. The product tends to accumulate exceptions until it resembles a Frankenstein. There are also hybrid modular approaches with several credible players working on them.
Why We Back AI-Enabled Services
The model we find most convincing is AI-enabled services. The company combines AI with a fairly similar underlying architecture, and points it at the return the clinic actually wants. That means going deep into the operational side of a practice, much as we mapped out in our post on where the money is flowing in Digital Health.
That satisfies both halves of the buyer’s question, since the clinic gets the technology it needs. It also gets a measurable outcome out of the other end. The standard objection is that services scale less well, and our answer is that tooling has improved sharply. Buyers are more comfortable than a year ago, and adoption is accelerating accordingly. Depth is what makes this play defensible, because a company that gets far enough inside learns the business properly. Competitors cannot replicate that from outside, and the ROI compounds the deeper it goes until the relationship looks less like a supplier and more like a partner whose work shows up in the clinic’s margin. We see the same dynamic outside healthcare, where service-wrapped cold calling agents convert more reliably than the pure agent equivalents.
What Has to Be True on the Buyer’s Side
A thesis about clinic software is also a thesis about clinic readiness, and the two move at different speeds. A large share of buyers want to integrate AI without having noticed that their own data is untidy. Whether they clean it internally or bring in a third party, the work has to happen first. Larger groups with established procedures tend to arrive further along, while smaller clinics that moved recently usually do not. The practical consequence is that some technologies cannot sell until their market has done this preparation.
This is one of the reasons we expect certain sectors to unlock for investment over the next twelve months rather than today. Data readiness is available to anyone who wants it now, the early adopters already have it, and any forward-looking organization that has not started will have started within the year, at which point the companies selling into them become investable on a different set of assumptions.
Operating Efficiency Ahead of Clinical
The same logic explains why we weight operating efficiency above clinical efficiency for now. Patient intake, triage, and the wider patient-facing operation are less regulated and easier to adopt, which is why that part of the market is moving quickly and why the window to enter early is closing, since in another twelve months the leading companies there will be raising Series A and above. On the clinical side the answer depends entirely on the pathology, with cardiac attracting significant investment while sitting behind regulatory clearance that takes considerably longer than a year, whereas sleep and mental health remain open precisely because nobody has closed them.
There is one more effect worth naming, which is that clinics burned by a horizontal deployment tend to emerge from the experience understanding that a service or vertical model suits them better. Those scars are expensive for the clinic and useful for the market, because they shorten the distribution cycle for the companies building the right way, and that is a large part of why we think the private side of this market works well over the next few years.
The Bio and Digital Pharma Side
Digital health is where most of the fund sits, but the same review exercise that shapes our digital health investment thesis runs across bio and digital pharma, where the categories have proved more stable than on the clinic side. Three of them have held their place through successive reviews. The first is AI tooling applied to scientific research, where the value shows up in how quickly a team gets from hypothesis to usable result. The second is tech bio in the broad sense, meaning new molecular technologies that change what can be attempted at the bench rather than making existing work faster. The third is data connectivity and liquidity solutions, which matter here for much the same reason data readiness matters inside a clinic, since very little else works while the underlying data stays locked in systems that do not speak to each other.
Biomanufacturing is the most recent addition, and it came into the thesis the way most categories do, which is by appearing repeatedly in the pipeline before we had a formal view on it. The stability of the bio side is worth reading alongside the changes in digital health, because the two move on different clocks. Categories built on software adoption open and close within a couple of review cycles, whereas the ones built on scientific capability tend to stay open for longer and reward patience instead of speed.
Where This Leaves Us
Concentrating most of a fund in one sector only works if you are willing to keep editing what that sector means, which is why we would rather describe our digital health investment thesis as a set of live judgments than as a fixed list of categories. The filters that do most of the work are defensibility and timing, and applying them honestly means accepting that some genuinely large markets are closed to us because somebody got there first, while others are open only once the technology or the buyer catches up with the idea.
Inside clinic efficiency the same discipline points toward depth over breadth, toward companies that get far enough inside a practice to understand it and to prove the return in numbers the clinic recognizes. Everything else follows from how quickly buyers get their own data in order, which is the variable that decides when whole categories become investable. Over the next twelve months we expect that variable to move, and we would rather be early to the sectors it unlocks than tidy about the ones we have already crossed off. On the bio side the variable is a different one, closer to what the underlying science makes possible, which is why we hold those positions for longer before revisiting them.
