September 2023: How to build sustainably differentiated tech enabled healthcare services businesses
For digital health startups delivering care, clinical innovation is neither necessary nor sufficient
Healthcare and financial services are often compared because both are regulated industries that comprise a large part of GDP (~18% of US GDP for healthcare [1], and ~8% of US GDP for financial services [2] respectively). There are lots of parallels; around slowness to adopt new technology, being some of the last industries embracing the shift from on-premise to cloud, etc. One parallel is that both industries have meaningful service offerings that are super capital intensive. For financial services, that offering is typically lending (vs. payment acquiring or card issuing for example which are 2 orders of magnitude more capital-light). For healthcare, that offering is care delivery; the literal function of providers seeing patients and helping them with their care (vs. EHR software, scheduling software, etc).
Over the last 15 years, many venture backed startups have tried to disrupt lending with varying levels of success; usually this means figuring out a way to utilize software to create a sustainable lending advantage. The results have been mixed (and much less successful than other fintech products in terms of reach relative to incumbents. Chime and Cash App are top 10 consumer debit issuers nationally, but I don’t know of a single consumer lending startup that would be considered top 10 along any axis). Over the last decade, venture backed founders have started to go after care delivery in healthcare; this has meant figuring out a way to utilize software to create a sustainable advantage in care delivery. The most successful examples (at least in terms of scale) are probably Teladoc & Livongo, and maybe to a lesser extent One Medical/Oak Street Health/Signify (I say a lesser extent because these have all been acquired rather than running as standalone entities, but at least Oak Street Health and Signify were acquired for higher valuations than Teladoc’s current valuation as of Q3 2023).
At Carbon, I spend a lot of time in the weeds of this (trying to apply software to create a step function improvement in some aspect of the delivery of care). I’ve been at this for less than 4 years (and Carbon’s been around for 7 years) and my experience is deeply shaped by COVID. This essay is my attempt to sketch out a simplified model of the material factors at play, and outline what can work and what (I think) is repeatable. Treat these approaches as works in progress, and if I’m wrong about something or have the wrong frame, please call it out.
In 2023, traditional healthcare services companies are:
Scale players split between for profit and nonprofit ownership. For profit healthcare players are often owned and run as financial businesses (often by private equity firms), with margins driven by operational excellence, rather than developing advantage through technology
Massive market/reach - most of your care, and most people you know, are using a traditional provider (either as a PCP, or for ER visits, or specialists) whether for primary or specialty care
Margins lower than software, and growth generally slower than software businesses
Highly capital intensive, typically driven by labor and fix asset costs
Typically use off the shelf technology tools, and do not develop their own software
Before I dive in, a couple of caveats; first, these observations come from intimate knowledge of the Carbon Health business and our successes and failures over time, and secondhand knowledge of other businesses. Second, this line of thinking applies explicitly to companies responsible both for building the technology and delivering the care to patients (if your solution is pure software sold to providers, or pure services + mostly buying technology off the shelf, this line of thinking likely doesn’t apply) [3]. Third, for a lot of the examples I cite, the book is still being written, and given how difficult it is to “know” something (especially something you’re not a part of) there’s lots of either a) ways I could be wrong, b) important dimensions I could have missed or c) reasons a company could succeed or fail that are not related to the quality of the business or application of the lever. With all that being said, I’ve identified 4 levers, and as a startup building in this space, you need to have at least 2 and ideally 3 to scale and reach more patients faster than you would if you just grew out of free cash flow. Those levers are:
a differentiated way to get paid vs incumbents,
step function improvement in clinical outcomes (that your payor values),
a differentiated operational model (how you deliver the service for lower cost) and
a differentiated patient acquisition model (how you acquire patients for cheaper).
Ideally, your investment in technology moves these levers and helps you differentiate - sometimes that’s a novel technical approach and sometimes it’s just encoding a business insight that’s hard to operationalize without building specifically around it. If your technology isn’t driving differentiation, you should really ask yourself if you need to invest in it (because hiring and maintaining a technical organization is expensive). If the unlock isn’t technical, that doesn’t mean you’re in a bad business; it just means this model I’m trying to describe might not apply to you. I’m at the front end of identifying a potential fifth lever (exceptional operational telemetry) but I haven’t seen it work yet personally, so I’ll describe it with the caveat that I might tell you I’m wrong in a few months. If you’re building a tech enabled healthcare services company, investing in them, or considering joining one as an early employee, this is written for you, and if you have strong feedback (and particularly if you disagree, ping me).
Why this matters
Research suggests it takes hospitals and clinics 17 years for research evidence to become widespread [4]. In practice the 17 years stat masks extremely wide variability; things with obvious order of magnitude higher performance get adopted faster (GLP-1s would be an example). Incremental improvements that don’t cut above the narrative clutter take even longer. The promise of being able to grow fast, is that by reducing this lag, you’re able to reach more patients with clinical innovation much more quickly than otherwise.
Differentiated reimbursement model vs. incumbents
This lever means you get reimbursed for delivering care in a way that diverges from the model utilized by the dominant providers of that type of care/specialty in your space. This is the difference between treating diabetes as an endocrinologist, and getting reimbursed in a traditional fee for service (FFS) model, vs treating diabetes utilizing continuous glucose monitors, and being reimbursed via a remote patient monitoring (RPM) model (with the RPM example as the "differentiated payor or reimbursement model).
Tactically speaking, your goal with utilizing this lever, is to create a care delivery model that is primarily reimbursed differently from the dominant models for the specific specialty, patient population, or even region that you cater to, utilizing a payment model or buyer that is fast growing and eventually can become the dominant payment model in your space. By picking this correctly, you ride the growth of that reimbursement model to become a dominant player in your space. For example this can mean going into a space that is primarily paid as fee-for-service, but your reimbursement model is value based, or some other PMPM model. In all cases, the way you know a differentiated reimbursement model is working, is it should result in higher contribution margins (typically through higher revenue per unit for whatever atomic unit of care you organize around), and should cause heartburn for incumbents as their operations and costs don’t initially support it.
How you get paid also affects both patient conversion and your go-to-market (GTM) approach. For patient conversion, any model that subsidizes the out of pocket cost to a patient (either subsidized by their insurer, employer or the government) will result in orders of magnitude more patient volume than an unsubsidized model. I think this is true unless your offering is a step function improvement over the the standard of care in terms of patient appeal. For example the actual latent demand for weight loss/weight control products is so widespread, that we’re seeing tons of startups chasing the Ozempic opportunity as direct-to-consumer offerings[3] because there’s tons of anecdotal and clinical evidence supporting the idea that it enables you to lose a material amount of weight. For GTM, patients often won’t try new things (even with promising clinical results) unless they are subsidized/covered by the patient’s insurance or employer. This is because historically, utilizing healthcare providers that are unsubsidized or out of network often results in larger or surprise out of pocket expenses.
Of all the levers, differentiated payor models are most alien to technology founders digging into healthcare. They typically require a fairly nuanced knowledge of the specific specialty area that you only really get from working in it, or being a chronic patient in that specialty for a long period of time. Innovating here also typically requires a strong network with the various payors and relationships with decision makers, because as a startup, that network gets you a leg up by enabling you to a) know the right decision makers, b) understand their incentives and the decision process of their organizations, and c) get them to take a risk on you early (otherwise you spend the capital you raise negotiating an agreement, and by the time it’s signed, you need to raise again). I think to do this right, you need to have payor contracts basically negotiated and ready to go prior to even starting the company or raising any outside capital. This is the reason why a strong network with payors is important; startups that do this typically form out of the founders realizing there’s strong payor demand for a particular reimbursement model or mode of care in a particular specialty and they form a company around it, with payor support from day one.
Examples of the differentiated reimbursement model in action
At Carbon, we’ve explored the differentiated payor model a handful of times, most recently in our diabetes offering which is funded via Remote Patient Monitoring, and in our partnership with BCBSMA which is a value based care arrangement.
An easy way to visualize this [5]:
Differentiated Clinical Model: step function (10% or more[6]) improvement in outcomes along some axis
A differentiated clinical model helps drive consistently better clinical outcomes vs the dominant players in that specialty <> patient space. This usually means you either
Have a way to extend clinician supply for the same cost, so that your cost of delivering care is sustainably lower than the incumbents, for the same level of quality, or
Centralize a care journey which has complex coordination across multiple steps, providers and (sometimes) payors, or
Introducing a novel diagnostic or therapeutic approach that improves care incrementally, but is not yet widely adopted as the standard of care [7]
The first approach solves for provider scarcity which is an industry wide problem (there are pretty much not enough doctors for anything in any specialty[8]). The second approach unifies the care journey by giving the providers and patients visibility and control/influence over all aspects of it and reducing fidelity that’s lost as data jumps from app to app and provider to provider. The third is an arbitrage on adoption (the average clinical innovation takes 17 years to become widespread). Some examples of #1 & #3 include Steady Health, which utilized CGMs as a novel diagnostic, and used that to extend supply of endocrinologists (because the same endocrinologist could care for many more patients). An example of #2 is Partum Health (where I’m invested) which centralizes all aspects of the perinatal journey beyond the OB, including Doula services, lactation services, pelvic care, and more.
Novel diagnostics & therapeutics
A more exciting, but rare way to do this is to introduce an absolutely novel diagnostic or therapeutic approach that improves on existing paradigms by a step function, by either a) dramatically reducing time to diagnose, b) dramatically reducing cost to diagnose, c) dramatically improving diagnostic precision or d) increasing access to diagnosis.
The best recent example of a novel diagnostic approach I’ve seen is Network Eye Care; they provide diagnostic and therapeutic services focused on diabetic retinopathy (including macular edema); this can be done in clinic (hence expanding access) without specialists (hence reducing costs), quickly (hence increasing access to diagnosis and improving speed) and runs on hardware that’s already widely adopted, and has already been FDA approved. While it will take years for this to be widely adopted, the improvement over the standard of care is incredible.
The best recent example of a novel therapeutic approach I’ve seen recently is Virta Health; they’re built around a specific therapeutic model for reversing diabetes, and are steadfast in their commitment to it. Other recent examples would be companies like Heading Health utilizing psychedelics applied in mental health, and startups like Sunrise deploying Ozempic and Wegovy for weight loss.
In the short term, these approaches are functionally an arbitrage on billing codes - the billing codes and associated contracted rates give you an economic ceiling to work with, and by deploying these approaches you’re either a) trying to expand the unit margins (for example in cases where you reduce the cost) or b) trying to shift the rendering and billing provider from those services (for example, moving a test from a sendout lab to point of care means that the provider administering the test gets paid for it, rather than the lab getting paid for it).
In the long term, these novel approaches expand the TAM - by decoupling the diagnosis or therapy from provider supply, you dramatically expand who it can reach. In the long term, expanding supply should probably bring the overall cost of the service down, but I’m too new to healthcare to have observed this actually occurring yet.
More generally, differentiating on clinical outcomes could impact your business profile in two ways. First, it could help reduce your patient acquisition costs; either because you’re D2C and patients want it so much that they’ll do anything to get it (in the case of GLP-1s), or because payors want it so much that they’re willing to steer patients to you to get it. Second, it could result in higher revenue per unit (without you needing to change your reimbursement model) because payors will be willing to pay higher contract rates to incentivize you to take their patients.
Differentiated Operational Model: step function (10% or more) improvement in unit margins
A differentiated operational model involves applying some insight that allows you to deliver care at a similar quality as incumbents, but at meaningfully reduced cost that is driven by removing administrative overhead & cogs (this reduced cost can manifest as a reduced price, or more typically as the same price but with higher margins). A lot of the RPM models combine differentiated care models and differentiated operational models; the care models tend to be sensor or measurement driven rather than episodic, thus enabling the patient to be home (or at least not required to visit a provider for intervention). Simultaneously, the reduced need for in person encounters drives down the need for space and by extension the cost of real estate.
If your model has an operational lever, you should be looking for ways to remove a dimension of fixed cost by changing how care is delivered, or remove a meaningful portion of variable cost by automating away an administrative part of the care journey. Executing well against a differentiated operational model should result in higher margins for the same billable service (or higher margin per patient if you’re in a value based care arrangement), or the ability for providers to take on incrementally more visits per unit of time, or a larger patient panel.
Differentiated operational models are what I’m most familiar with because Carbon deploys this throughout our system; we have an integrated technology stack (including our own EHR) which gives control and visibility into care delivery and enables optimization on cost, reduces provider turnover, and centralizes and reduces IT spend. For Carbon this is an example of being 2x better at 10 things than 10x better at one thing; I can’t point to one feature that is our operational lever. Instead, we have unreal execution speed relative to incumbents, so we’re consistently first to market. For instance, our hands-free charting is the single largest deployment of generative AI reaching patients today (this won’t always be true, but it speaks to the power of our model). I believe over the next several years, generative AI will dramatically simplify many parts of the administrative stack across billing, medical coding, scheduling, post visit clinical workflow, call centers, fax & secure messaging, prior-auth, and more. This is one of our core bets, and our operational lever is that we can harness new technology faster than anyone else.
Durably lower patient acquisition costs
There are a few ways to acquire patients that materially reduces your cost of patient acquisition. First is having a product, service offering, or channel approach that cuts through the clutter so well, that your conversion to a finished visit is higher than average, and thus your net cost of patient acquisition is lower. An example of this would be like selling Viagra or Ozempic as a direct-to-consumer (DTC) offering. Consumers clearly want it, and incumbents struggle to execute DTC offerings as they often require nuance that doesn’t translate to their core business.
Next is trying to acquire patients in a novel channel, with a solution that’s at parity with the incumbents. If incumbents or competitors are not in a particular channel, and that channel has net lower costs than others but converts at the same rate, that translates into lower overall CaC. For instance this might mean acquiring patients via ZocDoc or Solv Health for a high LTV procedure or service that’s still mispriced. Of all the approaches I think this is the most ephemeral - there’s just too much incentive for competitors to drive spend in your channel or for the channel to raise prices, that as a result it’s hard to believe channel arb is long run sustainable.
A third mechanism for durably reducing patient acquisition costs is solving a problem that a particular insurance, health system, or payor group has with a large population. By developing a care model in partnership with that entity, they might be willing to assign or steer patients to you. As a result you spend less to acquire those patients and you’re able to deploy that excess capital elsewhere.
A fourth mechanism is systemically improving retention and reducing churn. One Carbon advantage is we’ve been able to have much higher retention than the average urgent care, driven by strong patient experience/NPS; this means you’re less likely to have to re-acquire patients at the same frequency of competitors.
This takes real skill because healthcare is mostly not impulse purchases, so you often have to be both differentiated AND catch patients when there choosing a provider. Durably lower patient acquisition costs allows you to pay back your fixed costs (of rent, technology, or clinical staff) more quickly, getting units/service lines to profitability more quickly, and allows you to reinvest those profits to grow faster than others.
This is Carbon’s other lever; we’ve continuously innovated on patient acquisition and retention via investing in long tail SEO, having an active and aggressive approach to SEM, optimizing customer support, retention, scheduling, price transparency, and more. We’re not perfect at these and we have some obvious areas to improve on, but our patient acquisition costs are step function better than competitors, and have been that way for years.
Wildcard: Exceptional Operational Measurement & Telemetry
I have this as a wildcard because its becoming increasingly obvious to us that this creates leverage, but its a work-in-progress. In addition, I haven’t seen any other startups describe this as part of their advantage. You’ll notice a throughline for all of these levers is measurement; there’s some baseline to how good the standard is, and you’re trying to be better.
In healthcare, the atomic unit of value you’re providing is a patient and provider collaborating on the patient’s care. There’s a lot of other things that happen in care, but basically everything is in service of patient care; you’re either providing it, paying for it, documenting it, or some combination. This atomic unit is completely people oriented - two or more humans in the offline world have to spend time together for this.
It is insanely hard to optimize people unless you’re good at measuring what they are doing, and a LOT of what they do in healthcare is essentially offline or otherwise quite difficult to capture in a software system, so click or view based metrics won’t do. As a result, a lot of lessons that software people learn about how to measure, end up performing poorly when applied to an offline environment).
Anything you want to measure, you have to design to be measured, otherwise it will be gamed, and even well designed measurement systems are highly susceptible to gaming. Since most things happen offline and are subsequently documented, you have to engineer your system to measure what’s happening there, because otherwise super important things will happen that you won’t capture, either on the care side (thus affecting quality of care) or the administrative side (thus affecting the efficiency of care delivery). You’ll drown if you’re not intentional about what you’re measuring, or you’ll optimize the wrong things.
From the outside looking in, my instinct is that one company that’s quite good at this is Teladoc, and mostly it’s because all their patient interactions are mediated by technology by default (primarily over video visits). Theoretically this allows them to easily measure and optimize provider <> patient interactions, in a way that’s quite difficult to do offline.
An uncomfortable hypothesis
The seed crystal for this essay was an observation I saw in our own business and a struggle [9] I started to see when I’d talk to early stage founders building tech enabled healthcare services startups. My observation: clinical innovation alone is not sufficient to drive rapid adoption of new evidence based care practices. In fact, you’re much more likely to succeed and scale a healthcare delivery business (regardless of whether it’s tech enabled) if you are differentiating on reimbursement, patient acquisition, or operational models than if you’re differentiating on the clinical model alone.
I think the main reason for this is decision making cycles for healthcare payers are super long, and that’s not even including the time it takes for you to get to the right decision makers and understand their decision process. As a result the time it takes to get payors (either government, insurance payors or employers) to adopt novel care models is functionally two fundraises away. I’ve now seen so many examples of this - in the early days our diabetes program clearly had super high quality clinical results, but struggled because the fundamental innovation was clinical, not payor based, and getting payor attention to engage with it was tough. In contrast, from the outside looking in I think companies like Lyra Health and Teladoc had a marginal clinical innovation (telehealth), but their operational models were novel (Lyra Health helped access new pools of high quality supply that did not take insurance) and at least in Lyra’s case, their reimbursement model was also novel - they connected those pools of supply directly to employers, bypassing insurance carriers entirely. Both are massive businesses today.
My hypothesis is that innovations in reimbursement, operations, or patient acquisition correlate more highly with success, than clinical innovations do. This is kind of a depressing thought. It means that, even if you discover an approach with better clinical outcomes for patients (and that is all you have), unless you also have figured out the economics in advance, you’ll likely be out of business before payors adopt it (because there’s no actual urgency for them to adopt it). If you’re starting a technology enabled healthcare business around a novel clinical insight, to improve your odds of success, you should invest just as much in figuring out which economic lever(s) you’ll utilize, to make sure you reach as many patients as possible, with unit economics that work, as quickly as possible.
Thanks to Meghan Doyle, Nita Sommers, Kalie Dove-Macguire, Bryan Roberts, Maurice Chiang, Claire Vo, and Bob Kocher for helping refine this while in draft form.
[3] The main reason this line of thinking doesn’t apply if you’re not building the technology is that, with few exceptions, if you’re primarily using technology off the shelf that’s available to everyone, it’s very likely the technology provider is competing with you to extract that surplus economic value.
[4] https://hbr.org/2019/08/4-ways-to-make-evidence-based-practice-the-norm-in-health-care
[5] In practice, a lot of the “Patient Pay” category are really employer paid services in disguise. These companies have a consumer business, but primarily try to scale by selling to employers.
[6] I think 10% is the floor here. I don’t have a good way to think about how large the improvement has to be. But in almost all cases that improvement should translate into money you can reinvest into the business, which will support your ability to grow and scale faster and reach more patients as a result.
[7] A model that is more patient centric/friendly/well liked than traditional models is notably missing from this list; the main reason is that I think lots of founders start here, and it ends up just not being the thing that matters eventually.
[9] You’ll notice pretty much nothing about Medicare, Medicaid or Medicare Advantage in this essay. I have an understanding of those models and how they work, but I do not know what moves the needle for them well enough to integrate them into my thinking just yet.
Great read here, thanks Kunle.
What are the points of failure do you think can happen specifically when you mentioned about partnering with payors for large populations as one option to reduce patient acquisition costs? Also what sort of revenue models are likely here? PMPM, revenue share, VBC/capitation?
Being nit-picky here, but would tweak this: "you need to have payor contracts basically negotiated and ready to go prior to even starting the company." Realistically, payor contracts take a very long time to negotiate even when the parties are already working together, and they will not waste their legal team's time negotiating with a non-entity. Founders need to understand the difference between agreeing in principle with payor business leaders vs. getting the contract done with legal. Legal can sometimes take many many more months than expected. I don't think you need to have it negotiated... you just need to understand HOW to negotiate the contract and already have an internal executive champion within the payor who will advocate for you.