The Five-Year Desert to Product-Market Fit and a $5.3B Valuation | Shiv Rao, Founder of Abridge
- Aisha Washington
- 1 hour ago
- 7 min read
Abridge is now valued at $5.3 billion and widely regarded as a leader in vertical AI. Its path there, however, included roughly five years without the kind of product-market fit that makes a startup’s future feel secure. In this conversation, founder Shiv Rao explains what sustained the company through that uncertain period and how its strategy evolved as healthcare and artificial intelligence changed around it.
The central lesson is not simply to persist. Rao distinguishes between remaining committed to a durable insight and becoming attached to one product, customer segment, or business model. Abridge repeatedly changed its approach while preserving its conviction that conversations between clinicians and patients contain an unusually valuable signal—one capable of improving documentation, workflows, and eventually the broader economics of care.
Holding the Thesis While Changing the Company
Abridge began in 2018 with a clear premise: healthcare is fundamentally organized around conversations. The exchange between a patient and a clinician is where symptoms, decisions, concerns, and plans become part of a medical story. If that exchange can be captured accurately, securely, and usefully, it can support an entirely new layer of healthcare infrastructure.
Rao says that premise remained non-negotiable even when the surrounding strategy did not. The team was willing to change products, routes to market, and revenue models. It was not willing to discard the belief that clinical conversations could become the foundation for better healthcare workflows.
That distinction matters because perseverance alone can preserve a bad idea. Rao’s version of resilience requires founders to ask which belief is fundamental and which decisions are merely current implementations. A founder may be deeply committed to a problem while still being prepared to close an unsuccessful product and begin again.
Abridge also benefited from technological and market shifts that were difficult to time precisely. Transformer-based systems, large language models, pandemic-era disruption, and severe clinician burnout gradually made its original vision more feasible and urgent. Being early gave the company time to learn, but it also required surviving until the market caught up.
From a Consumer App to Enterprise Healthcare
Abridge initially developed a consumer product that allowed patients to record and retain information from medical visits. In retrospect, Rao views parts of that period critically. During the pandemic, he believes the company might have gained more by concentrating on research, model development, and viable business architectures instead of pushing a direct-to-consumer application.
The consumer route also introduced a difficult transition. Healthcare institutions must trust vendors with sensitive information, and a company designed around individual users does not automatically inherit enterprise credibility. Abridge ultimately shifted toward serving clinicians and health systems, while reconsidering a business model that could have created uncomfortable incentives around sensitive data.
Yet the early patient focus was not wasted. It reinforced the importance of giving people agency over their healthcare stories. Rao says Abridge is now returning to patient-facing opportunities from a stronger position, using AI to help both sides of the clinical encounter rather than treating clinicians and patients as separate markets.
This history illustrates the company’s broader operating principle: a strategic retreat does not have to invalidate the original mission. Sometimes a startup must leave one route, build the necessary capabilities elsewhere, and return later with a more credible product.
Healthcare Is a Collection of Markets
One of Abridge’s important go-to-market lessons was that “healthcare” is too broad to function as a useful customer category. An independent physician, an academic medical center, and a large integrated delivery network operate under different constraints and purchasing processes.
For a company with ambitions to reshape clinical work, reaching large care-delivery organizations became particularly important. These systems concentrate substantial numbers of clinicians, workflows, and data relationships. Selling into them is slower and more demanding than remaining down-market, but it also provides the scale needed to create broad impact.
Abridge entered through a task that is nearly universal: clinical documentation. Doctors spend significant time producing notes, entering orders, selecting diagnoses, and completing billing-related work. Automating part of that burden offers an immediate benefit without requiring an institution to redesign its entire operating model.
Rao emphasizes that documentation is not an isolated administrative artifact. In the US healthcare system, documented care drives reimbursement. A clinical note can therefore connect the physician’s workflow to the priorities of a chief medical information officer, CIO, and CFO. By designing the documentation architecture deliberately, Abridge could move from relieving clinician burden toward supporting enterprise value.
An Intelligence Layer, Not a Replacement for Epic
Rao does not frame Abridge as a direct substitute for electronic medical record platforms such as Epic. Instead, he describes the company as an intelligence layer built above existing systems of record.
The clinical conversation provides the initial wedge. From that source, AI can help create notes, prepare charts, surface relevant context, support orders, and connect documentation to billing. This lets Abridge extend its role without demanding that health systems abandon their core infrastructure.
That positioning also creates a form of counterpositioning. An entrant can build around a workflow or business model that an incumbent may struggle to reproduce without disrupting its existing products. Rao advises founders to understand competitors well enough to identify such constraints, while keeping their daily attention on their own customers and product.
Abridge’s growing category identity suggests that this focus has resonated. Rao notes that some customers have even begun using the company’s name as a verb—a strong sign that a product has become embedded in everyday work.
Choosing Models Around the User Experience
Abridge does not treat model ownership as an ideological question. Rao says the company uses frontier models, open-source systems, and internally developed models according to what produces the best outcome for clinicians.
At the time discussed, approximately 40% of model output came from in-house systems, though Rao cautions that the share changes as models are distilled, fine-tuned, and replaced. Internal models are especially attractive for bounded tasks where speed, control, and lower cost matter. More difficult problems may benefit from riding the frontier as general-purpose systems improve.
Latency is a decisive consideration. In a high-pressure clinical environment, even small delays can make a tool feel intrusive. Rao compares the desired experience to good air conditioning: essential, reliable, and largely unnoticed. The technology should disappear into the workflow so the clinician can concentrate on the patient.
This product standard shapes model selection:
Quality must be sufficient for consequential clinical workflows.
Responses must arrive fast enough to preserve the natural rhythm of care.
Cost matters, particularly as usage expands.
The architecture must adapt as frontier and open-source capabilities change.
Scaling also produces a learning advantage. More deployments expose the system to different specialties, languages, care settings, and clinician edits. Those interactions can help Abridge refine products and models for a much wider range of users.
Proprietary Advantage Comes From Workflows and Trust
Foundation-model companies create powerful tailwinds for vertical AI, and Rao prefers to collaborate with that progress rather than compete against it directly. Abridge’s defensibility, in his account, comes from regulated-industry expertise, proprietary workflows, accumulated feedback, and the ability to handle healthcare data correctly.
Enterprise healthcare data is rarely clean or uniform. Behavioral-health information cannot always be treated like internal-medicine data, and integrating either into a compliant workflow requires extensive infrastructure. A useful vertical AI system must do more than call a capable model; it must organize context, respect institutional rules, and deliver the result at the exact point where work happens.
Trust is equally important. Rao says Abridge does not build its business by selling healthcare data. When a proposed feature would use partner data in a new way, the company’s philosophy is to earn permission from health systems before proceeding. That discipline may slow some opportunities, but it protects the relationships on which long-term adoption depends.
Taste as a Human and Organizational Advantage
Rao identifies taste as an increasingly important differentiator in AI product development. For him, taste is the ability to recognize meaningful patterns, combine ideas in distinctive ways, and make choices that express a coherent human point of view.
Abridge encodes this idea in a cultural principle: people develop good taste by experiencing good work. Researchers need exposure to important papers; product teams need familiarity with excellent interface patterns; leaders need to observe strong organizations and decisions.
Exposure alone is insufficient, however. Taste requires selection. Teams should remain curious without indiscriminately chasing every new model, design trend, or management idea. The advantage comes from identifying what deserves attention and translating it into choices that feel consistent with the company’s mission.
Rao connects this to the belief that exceptional companies create culture rather than merely respond to it. To do that, they must remain close to the edge where technology, behavior, and expectations are changing.
Leadership Under Constant Pressure
As Abridge grew to around 450 employees, Rao’s job changed repeatedly. He describes founder leadership as a succession of “tours of duty”: identify the company’s most important challenge, determine where the founder can contribute uniquely, and commit to that role until the situation changes.
This is different from micromanagement. As the interval between decisions and actions becomes shorter, companies need executives capable of exercising judgment without waiting for the founder to settle every detail. Rao considers hiring such leaders one of the hardest and most consequential parts of scaling.
He also argues that there is no meaningful peacetime for a technology CEO in a rapidly moving market. Pressure can preserve urgency, while the absence of it may invite complacency. Abridge reinforces responsiveness through cultural practices such as its “Titanic rule,” which expects employees to acknowledge messages within three hours.
The company currently asks employees to spend three days each week in the office, although Rao presents the policy as something still being calibrated. He values whiteboards, spontaneous interaction, and physical proximity, while recognizing the burden on employees with difficult commutes. His broader point is that culture cannot be copied mechanically; it must fit the company, its executives, and its stage.
Building Toward More Human Healthcare
Rao’s long-term ambition is larger than automating medical notes. He wants AI to help create deflationary healthcare economics while giving clinicians better context, cues, and decision support.
Some repetitive, high-frequency, lower-risk tasks—such as routine medication refills—may be suitable for extensive automation. As cases become more complex, however, human judgment becomes more important, not less. Rao’s vision is therefore not centered on replacing clinicians. It is about removing administrative work and supplying relevant context so healthcare professionals can devote more attention to difficult care.
That framing is especially important as demand rises faster than the healthcare workforce can comfortably absorb. Abridge’s five-year “desert” demonstrates how long a consequential market can take to open. Its subsequent growth shows what can happen when a persistent thesis finally meets capable technology, urgent customer demand, and a product designed for the realities of the workflow.
