In the medical field, AI technology is gradually changing the way doctors work. OpenEvidence, as an AI diagnostic assistant tool designed specifically for doctors, has rapidly gained popularity among American doctors through precise clinical support and innovative business models.
In the previous research map, we pointed out that the medical field is likely to be the first area where Vertical Agent will land, and one of the most representative companies is OpenEvidence, an AI professional diagnostic Copilot designed for doctors. In the face of the explosive growth of medical knowledge and the severe overload of clinical information, OpenEvidence is committed to helping doctors improve diagnostic efficiency and decision-making quality with a product form similar to vertical Deep Research.
Founder Daniel Nadler, a serial entrepreneur in the field of AI, mentioned in an interview that OpenEvidence has rapidly gained popularity among doctors in the United States, with more than 100,000 doctors currently using their products every month. In February 2025, OpenEvidence completed a Series A funding round exclusively invested by Sequoia Capital, raising $75 million and raising a post-investment valuation of over $1 billion.
Unlike traditional medical software that relies on complex procurement processes for hospital systems, OpenEvidence adopts a direct-to-user growth strategy: providing services directly to individual doctors, eliminating lengthy approvals and centralized procurement. With functional design that precisely meets daily clinical needs and word-of-mouth spread among doctors, the product has achieved viral growth. Its business model embeds precise advertising by cooperating with pharmaceutical companies and medical device manufacturers, and smoothly inserts AI Agent into the advertising budget allocation of traditional medicine representatives and academic conferences, opening a new monetization path.
01. Background
In today’s medical field, doctors are facing unprecedented challenges. The explosion of medical knowledge has made clinical diagnosis and information processing extremely complex: medical knowledge is updated every five years, literature is growing at a rate of one article every two minutes, and PubMed has indexed 36 million abstracts, with 1 million new articles added each year. Google Scholar contains about 400 million articles, citations, and patents. In this context, doctors need to process a large amount of clinical information during the diagnosis and treatment process, but traditional search tools are difficult to quickly find specific information hidden in the literature, resulting in the problem of information overload.
After 10 years of interaction design, why did I transfer to product manager?
After the real job transfer, I found that many jobs were still beyond my imagination. The work of a product manager is indeed more complicated. Theoretically, the work of a product manager includes all aspects of the product, from market research, user research, data analysis…
View details >
At the same time, the problem is further exacerbated by the uneven allocation of medical resources. The World Health Organization (WHO) reports that doctors in low-income countries have only 1/9 of the frequency of access to cutting-edge medical evidence compared to high-income countries, forming a significant “cognitive scissors gap”. Even in the United States, the penetration rate of rural hospitals purchasing clinical decision-making systems (such as UpToDate) is less than 1/7 of that of teaching hospitals.
In addition, with the advent of an aging society, complex cases of multi-disease coexistence are becoming more frequent. Data shows that the average patient over the age of 65 takes more than 5 drugs and has more than 300 million combinations of drug interactions, compared to less than 7% of traditional guidelines coverage in such scenarios. In such a complex diagnosis and treatment environment, it is significantly more difficult for doctors to make decisions based on experience alone.
Can LLMs with world knowledge + long context improve this dilemma? Experiments show that there are still obvious limitations in the application of general AI large models in the medical field. Researchers at Cohen Children’s Medical Center in New York fed 100 pediatric case reports into ChatGPT and found that it had an error rate 83% higher than experienced doctors; In Medscape’s 150 case tests, ChatGPT’s area under the curve for diagnostic accuracy (AUC) was only 66%. These data show that AI’s diagnostic capabilities in dealing with complex diseases are still insufficient. This limitation directly impacts clinicians’ trust in general-purpose AI. When AI’s suggestions are inconsistent with doctors’ clinical reasoning, doctors often tend to trust their own judgment and ignore AI’s suggestions, resulting in AI’s potential in actual diagnosis and treatment not being fully exploited.
Faced with multiple challenges such as the explosion of medical knowledge, uneven resource allocation, and the limitations of general AI, the healthcare industry needs to find new solutions to improve the efficiency and accuracy of diagnosis and treatment, while bridging the gap in medical resource allocation. OpenEvidence is a representative company in this direction.
02. Products and technologies
OpenEvidence is a chatbot product focused on assisting in medical diagnosis, designed to provide efficient and precise clinical support to doctors and medical students. Its interactive interface design is unique,Each answer is marked with a cross-reference number and a reference list is attached at the end of the text to ensure the traceability and verification of the information.
It provides dual-mode answers of “nursing guidelines” and “clinical evidence”, focusing on practical suggestions and theoretical data support respectively. Each question answer is followed by a list of possible follow-up questions, making it easy for users to engage in multiple rounds of in-depth interaction.
In addition, OpenEvidence provides comprehensive clinical diagnosis and treatment support. The symptom analysis module quickly resolves fuzzy symptoms, provides potential etiology analysis, and recommends relevant examination paths. The treatment decision support function is based on the latest research, recommends treatment options and compares drug efficacy and resistance data, especially for rare diseases and complex crossover cases. The system also has a built-in real-time guideline access function to support quick retrieval of clinical guidelines and standards, such as CHA2DS2-VASc scores.
Administrative and workflow assistance functions are just as useful. The system can automatically generate medical documents such as advance authorization letters and patient discharge guidance, and automatically attach cited literature. Additionally, the integrated 50+ clinical calculator covers high-frequency scenarios such as disease scoring, drug dosage calculations, and more, simplifying complex calculation processes.
Medical Knowledge Tracking Learning Function (TL; DR) selects newly published papers every day, generates visual charts and specialty classification summaries, helping users quickly grasp the cutting-edge trends of the discipline. This structured knowledge update mechanism provides a convenient way for doctors and medical students to continue learning.
OpenEvidence dialog interface
OpenEvidence products are primarily intended for the following user groups:
- Doctors (core users): including general practitioners, specialists, and community doctors with insufficient medical resources in remote areas, used to inquire about rare cases, optimize diagnosis and treatment plans, and improve efficiency workflows.
- Nurses, physician assistants, pharmacists, etc.: Used to query nursing guidelines, medication information, etc.
- Medical students: Learn the latest clinical guidelines, prepare for medical exams (such as the USMLE), and get structured knowledge support.
- Medical researchers: Use to track discipline trends, follow the latest paper abstracts and visualize data, and shorten literature review time.
For a physician, a typical use case for OpenEvidence products is as follows:
Based on its product design ideas, OpenEvidence’s core value proposition is to help physicians quickly access the most up-to-date and relevant medical evidence and provide straightforward clinical diagnostic recommendations. The system supports complex clinical decision-making by integrating interdisciplinary medical information, especially in rare and marginal case processing. More importantly, OpenEvidence aims to democratize medical information, giving all doctors, not just practitioners in large healthcare institutions, equal access to high-quality medical resources.
In terms of technical performance, OpenEvidence has shown significant reliability. As the first AI system to score over 90% in the United States Medical Licensing Examination (USMLE), it outperformed ChatGPT in all three USMLE exams, with an overall error rate of 77% lower than ChatGPT. This low error rate significantly reduces the hallucination problem of general-purpose AI large models, significantly improving doctors’ trust in AI assistants. With these features, OpenEvidence is redefining the standards of utility and trustworthiness for medical AI.
The USMLE is the only admission test for medical students in the United States to qualify for clinical practice, and the exam content is divided into three steps, focusing on basic medicine, clinical medicine, and skill application.
So how does OpenEvidence technically deliver on its value proposition? Since its inception in 2022, OpenEvidence has chosen a different technology path. At the time, the industry was generally focused on developing general-purpose LLMs with larger parameter scales, while the OpenEvidence team focused on developing small, specialized models. Although this choice sacrifices the generalization ability of the model to a certain extent, it significantly improves its accuracy and reliability in the field of professional medicine.
In 2023, the OpenEvidence team’s research paper, “Do we still need clinical language models,” further validated this decision. The paper pointed out that in the medical field, small specialized models outperformed large general-purpose models, and won the Best Paper Award at the Machine Learning and Healthcare Conference that year.
Founder Daniel Nadler once quoted American science fiction writer Ted Chiang as saying, “A large language model is a JPEG compression of the world,” implying that it sacrifices detail accuracy in the pursuit of broad applicability. OpenEvidence, on the other hand, focuses on high-quality compression in the medical field, ensuring information accuracy and reliability in this critical domain. This focus allows OpenEvidence to provide more accurate and credible information support in the field of medical AI.
03. Commercialization and competition
The first advertising-led business model in the Chatbot space
OpenEvidence disrupts the sales model of traditional medical software, employing an innovative GTM (Go-To-Market) strategy similar to the “consumer internet”, throughProvide free products directly to doctors and medical researchers, and use the word-of-mouth communication of these professional consumers to quickly open the market.
The essence of this GTM strategy is a model that relies on product quality and user experience-driven growth (PLG), avoiding the complex process of traditional medical SaaS relying on hospital system procurement (including multiple meetings, AI committee reviews, budget approvals, etc., usually a complete set of approval decisions has a cycle of 2-5 years, and the approval results are uncertain), thus greatly reducing the threshold for doctors to use and promoting the rapid dissemination of products among the doctor community.
Founder Daniel Nadler also noted that the free strategy allows OpenEvidence to bypass the lengthy enterprise procurement process. For example, the U.S. Department of Veterans Affairs (VA), as one of the largest healthcare systems in the country, typically takes three years to procure new technology. OpenEvidence has received a lot of recognition and praise from VA physicians by making it free to VA physicians: many say OpenEvidence’s products have greatly improved the quality of care for veterans by helping them make treatment decisions at critical moments.
The GTM strategy chosen by OpenEvidence has proven to be fruitful: OpenEvidence Achieved zero users to cover 10%-25% of practicing physicians in the United States in just one yearThe explosion of the product, with about 100,000 doctors using the product per month and MAUs reaching 30-400,000, validates founder Nadler’s philosophy that “make a good enough product, and it will spread itself.” At the same time, OpenEvidence has also been recognized by the New England Journal of Medicine (NEJM), further enhancing the level of trust placed in OpenEvidence by professional users such as doctors.
It’s worth mentioning that OpenEvidence has been around since 2023Established an exclusive strategic partnership with the New England Journal of Medicine (NEJM).。 NEJM editorial executives took the initiative to contact OpenEvidence to seek cooperation, hoping that its common tools would include NEJM content, and at the same time rejected data cooperation invitations from other AI large model companies. The collaborative framework values the purity and academic value of the NEJM brand. This is due to the fact that upstream and downstream reciprocal chains can be formed around OpenEvidence: content providers such as NEJM gain exposure, OpenEvidence platforms gain traffic through more powerful model services, and physician users obtain high-quality information. Specifically, OpenEvidence brings tens of millions of visits to NEJM, and NEJM provides OpenEvidence with exclusive permission to train with NEJM’s full text, updating the knowledge base in almost real-time and continuously improving the knowledge level of AI large models.
In addition, OpenEvidence became a member of the Mayo Clinic Platform Accelerator in March 2023, entering a 20-week incubation program to accelerate innovation with Mayo’s clinical resources and technical support. Mayo Clinic opens clinical guidelines and partially de-identified clinical datasets to OpenEvidence for validation and optimization of AI models, as well as clinical expert guidance to ensure recommendations align with evidence-based medicine practices.
OpenEvidence also takes 2B to expand user channels. In August 2023, OpenEvidence reached an agreement with media conglomerate Ziff Davis to integrate its technology into Ziff Davis’ health-related websites, including MedPage Today for health professionals and Everyday Health for general readers, further expanding user reach.
In terms of monetization methods, OpenEvidence is based onPrecision advertisingas the core. Through advertising for pharmaceutical companies, medical device manufacturers and other medical-related industries, the closed loop of commercialization is realized. Due to the high-value attributes of the doctor group, their daily diagnosis and treatment decisions are highly related to drugs and medical devices, attracting advertising investment from relevant manufacturers.
The advertising content is closely integrated with clinical decision-making scenarios, such as automatically recommending PD-1 inhibitor advertisements from relevant pharmaceutical companies when doctors view immunotherapy papers, and marking the number of citations. or after the doctor enters a case of diabetes with kidney disease, recommends SGLT2 inhibitors with real-world study data. We believe that this precision advertising model based on professional user pools may be faster than the advertising commercialization of general-purpose large models.
Competitor: Traditional Medicine Database
In terms of the competitive landscape, OpenEvidence’s real competitors are UpToDate Clinical Medicine Database。 UpToDate is the world’s largest clinical community, now covering more than 10,000 clinical topics across 25 specialties, all written in the form of textbooks/teaching guides by renowned clinicians from around the world following the principles of evidence-based medicine, and rigorously peer-reviewed with up-to-date references to ensure content quality. As of now, UpToDate is used by 2 million healthcare workers, with the enterprise version costing $50-100/user/month, totaling ~$57.6 million in annual revenue.
OpenEvidence has the following advantages over UpToDate: AI-powered interaction capabilitiesIt’s not a static text page, which means users can get answers after asking targeted questions, eliminating the tedious query process.
For example, when a doctor asks the question, “How can poor patients replace rifaximin for small intestinal bacterial overgrowth,” OpenEvidence’s deep vertical model skips million-word topic summaries and pulls subphase clinical data for metformin combination therapy directly from the metadata to output solutions accurate to the third line of paragraph four. Doctors using UpToDate need to view a summary of the topic and click on the link to read it to find the specific answer they are seeking in the article.
04. Team and financing
team
Daniel Nadler, the founder of OpenEvidence, holds a PhD in economics from Harvard University, and his academic background covers the intersection of economics and artificial intelligence, which allows him to deeply understand the complex technical problems of artificial intelligence and apply it to practical business scenarios.
He is a successful serial entrepreneur across disciplines, and his entrepreneurial experience is equally remarkable. In 2013, he co-founded Kensho Technologies, an AI quantitative trading tool focused on serving Wall Street, with Peter Kruskall. Kensho proposed the “Warren algorithm” by analyzing millions of market data points to find correlations and arbitrage opportunities. In 2018, Kensho was acquired by Standard & Poor’s for $550 million, making it the most expensive deal in AI at the time, marking Nadler’s first major success in AI.
In 2021, Nadler founded Xyla, which focuses on developing high-accuracy LLMs. The Xyla team includes more than a dozen PhD and PhD students, and is equipped with a supercomputer in the Nevada desert. At the same time, the pandemic has made Nadler aware of the challenges doctors face in accessing the latest medical knowledge. Therefore, he incubated OpenEvidence in Xyla to focus on improving clinical decision-making accuracy through LLMs. To solve the problem of high computational costs, Nadler and his team used “Retrieval Augmented Generation” (RAG) technology, which combines user data with pre-trained large models to provide more targeted and reliable outputs while avoiding the “hallucination” problem in AI answers.
In addition to his entrepreneurial status, Nadler has also demonstrated a multifaceted talent. In 2016, his debut poetry collection, Lacunae: 100 Imaginary Ancient Love Poems, was published and named Book of the Year by National Public Radio. He has since ventured into the film industry, serving as a producer on The Brooklyn Secrets and Palmer. These experiences demonstrate his diverse achievements in the fields of technology and art.
The team led by Nadler adheres to the academic elite style, with most of the members coming from top laboratories at Harvard and MIT, and prefers academic research to be implemented. He believes that “top talent only wants to work with top talent”, and said in an interview: “My experience at Kensho made me realize that if you let someone with a high IQ and extremely fast learning speed solve a problem, they will progress far more than an average team that is 100 times larger. ”
CTO Zachary Ziegler is a PhD candidate in computer science at Harvard University under Alexander Rush, a top scholar in the field of NLP, with a solid academic background and specialized knowledge of machine learning. He previously served as the head of artificial intelligence at IMAX AI, founded Xyla with Nadler in 2021, and led the development of OpenEvidence as CTO, focusing on solving AI hallucinations in the medical field.
financing
In February 2025, OpenEvidence completed Sequoia Capital’s exclusive $75 million Series A financing, with a post-investment valuation of more than $1 billion. Sequoia partner Pat Grady, who led the round, believes that OpenEvidence adoption is similar to how consumer Internet products are spread: “Not many medical tools spread like consumer apps, but this is one of them.”
05. Conclusion
OpenEvidence’s core value proposition is essentially the use of AI to solve the contradiction between “information explosion” and “indexing efficiency” in the medical field. What is even more instructive is its business model: it directly “sticks” doctors like a consumer Internet platform – relying on ultra-high-accuracy, ultra-low-hallucination clinical AI tools to solve rigid needs, relying on word-of-mouth communication to achieve fission, and then transforming the professional user pool into a highly scenario-based and accurate advertising field favored by pharmaceutical companies and device manufacturers.
As a result, OpenEvidence not only taps on the urgent need for clinical AI tools among doctors, but also cuts into the advertising budget previously allocated by traditional healthcare companies to traditional medicine representatives and academic conferences – think of new drug data popping up when diabetic doctors check medication for complications, and oncologists can see the latest device research when formulating plans…… There may not be many advertising scenarios with a higher ROI than these “products that accurately push treatment plans when doctors make decisions”.
This approach of “precise monetization in vertical fields” provides a new idea for startups focusing on AI applications to think about how to break through the encirclement in involution. With the implementation of Vertical Agents in more fields such as healthcare, law, and finance, we may see a more efficient business model than general-purpose AI Chatbot inline advertising taking shape – and OpenEvidence has clearly grabbed the first move in the professional medical track.
