Reviewing our investment in DeepScribe: what did we get right, and what did we get wrong?
Digital Health: Reviewing our investment in DeepScribe: what did we get right, and what did we get wrong?
It’s always fun to revisit assumptions that we made at the start of an investment, and track them to see how they ended up. We invested in DeepScribe in September, 2019, and wrote an essay about why we made the investment (largely pasted below, but updated with a nicer graphic from Rock Health :)
Yesterday, DeepScribe announced their $30M Series A, so it seemed an appropriate time to look back at our original thesis. Since we invested in DeepScribe, they have built a best in class team of 40 employees and have grown to become one of the largest clinical documentation providers in America with just under 600 users on the platform.
There were several areas that turned out to be correct:
- DeepScribe had put into practice a novel statistical technique that theoretically would provide accurate transcriptions and and summarizations with a fraction of the data needed by other techniques; this has turned out to be accurate– their technology has worked brilliantly
- Doctors are willing to outsource the note-taking and summarization to an AI-enhanced system to reduce their paperwork; this has turned out to be accurate
In addition, there were several areas that turned out to be pleasant surprises. One of our concerns when we made the investment is that the founding team is very young– we knew that this business would be operationally heavy, and we were not sure if the team was well-suited to operational excellence. This has turned out to be a very positive surprise: DeepScribe is now operating at scale with physicians and quality control staff, and has built a highly functional customer acquisition machine. All of those aspects of their business are operations-heavy, and all are notably successful.
What did we get wrong? At the moment, it’s too soon to tell. Our belief is that AI-powered scribes will be ubiquitous, and will carry the load of physicians’ paperwork. There is still a massive gap between that assumption and reality. The pandemic has accelerated telehealth adoption, and in some ways that makes the adoption of digital transcription/ summarization services simpler. However, behavioral changes are more difficult than technology breakthroughs, so we are still in wait-to-see mode
Overall, it is fair to say that we are thrilled with the team’s progress on the technology and the execution sides. In addition, they are solving a huge societal problem: physicians really suffer under the weight of paperwork, draining the joy from the practice of medicine. DeepScribe can be an important part of the solution.
Below is our blog post from 2019 on why we made the investment
Foothill Ventures tends to be thesis driven in our approach to investment. That is, we have several theses about the way we think the world will evolve, and keep our eyes open for companies that fit these theses. We are occasionally opportunistic, but even in those instances, we use the opportunism as a chance to revisit our theses (ie., if Opportunity X is so appealing, is there a more generic insight that could lead to a rich vein of venture opportunity? :)
We’ve been interested in digital health for a while, but had not yet formed a coherent thesis. All of the elements are there: it is the largest part of the American economy ($3.8T — 18% of the economy); there are obvious sources of both inefficiency (eg., overwhelming amounts of documentation/ paperwork) and bottlenecks (eg., limited physical space and limited physician resources to see patients). In fact, ⅓ of physician time is spent on paperwork. Satisfaction is low for all parties: patients and physicians both feel that the system isn’t working.
At the same time, there are forces that are creating opportunity. The ACA is part of a more general shift towards outcome-based reimbursements, and a greater attention towards measurement. New reimbursement codes are being generated by CMS that make a host of digital services attractive for service providers, patients, and physicians.
Having said all of the above, there are so many potential opportunities, we had not yet settled on any single one as a thesis. Furthermore, for some of the more obvious areas (eg., remote patient monitoring services), we haven’t yet come to a conclusion about what elements will separate a winning team from a losing team in this coming goldrush. Clearly, a lot of VCs are pouring money into this general area: Rock Health tracked a greater than 2x rise in investments, to $29.1BB, in 2021 vs. $14.9 in 2020.
This pre-amble is intended to give context to the fact that we have been in a hybrid “opportunistic-thesis” mode. We know that there is an area that we want to explore, but we don’t quite know what we’re looking for yet.
We first saw DeepScribe at the Alchemist Accelerator demo day. It is extremely rare for us to invest in companies that we view through accelerators (the dynamics of the large accelerator cattle call is not very conducive to how we make investments), but we were struck by the thoughtfulness and articulateness (is that a word?) of DeepScribe’s founders, and wanted to learn more.
First, a short description of what they do: DeepScribe is an AI-enabled “scribe service”. Medical scribes are people with some medical training who assist physicians with patient interactions, by summarizing conversations, extracting medically meaningful data (ie., diagnoses, prescriptions, etc). Scribes help solve a major pain point that physicians have — they are buried in paperwork, which keeps them from helping patients and dramatically lowers their job satisfaction. The increased patient volume that the scribe enables should justify the increased cost. Scribes, though, have their own limitations: they are expensive and tend to have high turn-over (scribes are often medical students or nurses).
There are multiple companies and approaches in this domain. There are companies that are providing remote scribes or scribes on demand. This at least allows for some uniformity and scalability of operations. It also may allow for the use of offshore medical personnel (eg., in the Philippines or India) to handle the scribe duties at a lower price. However, these services still suffer from a fundamental efficiency problem (scribe : doctor ratio remains 1:1), and still marginalize the service to higher revenue generating physicians.
There are other companies (like DeepScribe) that are taking a technical approach to the problem. In general, they have focused on medical transcription (a part of the scribe problem) or being a generalized “alexa for physicians” (a superset of the scribe problem).
We fell in love with DeepScribe because (a) they took a highly pragmatic, human-in-the-loop approach to solving an important problem — how to provide scribe services efficiently while maintaining extremely high quality; and (b) they have taken a differentiated and novel technical approach to the problem.
On the pragmatic side: the DeepScribe team has been extremely clever in making decisions on what integrations need to happen (ie., with EHR systems), where humans still need to help polish the machine-generated output, and how to work seamlessly within a physician’s workflow to be effective. On the differentiated side, they have applied cutting edge statistical techniques from Professor Bin Yu’s lab at Berkeley that requires a fraction of the data to produce world-beating document summarization results (Drs. Murdoch, Feng, and Yu published on this subject here).
What does this mean? It means that DeepScribe is able to extract the relevant parts of physician-patient conversations accurately after a much smaller sample set of conversations. While the AI gets up to speed, they will still rely on quality control personnel to review and polish the output. Over time, though, the ratio of QA to physician should continue to drop. Eventually, they can offer a very low cost, full-automated solution AND a somewhat higher cost “concierge” solution that includes human assistance. Once they have mastered scribe services, other forms of medical transcription and documentation assistance become accessible.
Every doctor we talked to views this type of service as a godsend. Every current user (doctors) of DeepScribe is telling every other doctor that they know about the service. This sort of word of mouth, genuine love is extraordinarily rare. It allows the doctors to see more patients, make more money, and do more of the stuff they like (helping people) and less of the stuff they don’t (taking and summarizing notes)
So… to go back to the initial proposition: why did we invest?
It’s pretty simple:
- This (healthcare) is the largest part of our economy.
- It is riddled with inefficiencies and areas of dissatisfaction
- A large chunk of inefficiency and dissatisfaction has to do with the documentation load imposed on doctors
- Various AI approaches are well-suited to this task; DeepScribe has quickly built a product and service that handles this task brilliantly
On a selfish note, we like getting our feet wet in this exciting area. So much so that we made our 2nd digital health bet (with Moving Analytics) shortly after our investment with DeepScribe. These will certainly not be our last.
If you want to work on cutting edge AI to make doctors’ and patients’ lives better, check out DeepScribe’s job openings here
