Two Tech Outsiders Are Disrupting Healthcare with AI By Simply Asking ‘Why?

For years, documentation has been one of the most frustrating aspects of healthcare. Physicians, already burdened with long working hours and administrative responsibilities, found themselves spending an increasing amount of time writing notes instead of engaging with patients. This growing issue became a personal struggle for Akilesh Bapu and Matthew Ko, who co-founded DeepScribe in 2017 to alleviate the documentation burden using artificial intelligence.

Bapu’s inspiration came from his father, an oncologist who often struggled to balance his professional and personal life due to overwhelming documentation demands. Ko had firsthand experience with the issue when he served as a care coordinator for his mother during her battle with breast cancer. He saw how excessive note-taking affected not only doctors but also patients, who often felt ignored as their physicians focused more on writing than on meaningful interaction.

Despite the widespread adoption of digital tools, the documentation process remained inefficient. Physicians were still spending nearly half their day writing notes, even with existing speech-to-text solutions that merely transcribed conversations without summarizing them. Bapu and Ko saw an opportunity to leverage advancements in artificial intelligence and natural language processing to create a solution that could truly understand and summarize patient-physician conversations without requiring additional manual input.

How DeepScribe Works

In 2019, DeepScribe introduced its ambient voice AI technology, which records, summarizes, and integrates patient conversations into a physician’s Electronic Health Record (EHR) system. Unlike basic transcription tools, DeepScribe’s AI extracts only medically relevant information, filtering out small talk and irrelevant details. The system also adapts to physicians’ unique conversation styles and writing preferences over time.

The process is simple: once a physician activates the DeepScribe application, it records the patient exam, processes the conversation, and generates a clinical note that is then uploaded directly into the appropriate EHR fields. The goal was to allow doctors to focus on their patients rather than on documentation.

Despite its ambitions, DeepScribe’s AI was not flawless. Current and former workers pointed out recurring issues, including errors in medical terminology, incorrect spellings of medications, and instances where the AI added medications a patient was not taking. These errors reflected a broader challenge in AI-driven documentation: the tendency of algorithms to “hallucinate” data.

To mitigate these errors, DeepScribe relied on a team of 200 human contractors to review AI-generated records, correct mistakes, and ensure quality. These workers also used Google searches to verify billing codes, further underscoring the need for human oversight. According to the company, DeepScribe’s AI constructed approximately 80% of each record, while the remaining 20% was refined by human scribes. However, the extent of human involvement raised questions about the true capabilities of AI in medical documentation. While DeepScribe marketed itself as an AI-powered solution, much of its accuracy still depended on human intervention.

Ethical Concerns Over Transparency

A major concern surrounding DeepScribe’s operations was the handling of sensitive patient information. While doctors informed patients that their conversations were being recorded, some employees reported that physicians assured patients that only AI analyzed the recordings without mentioning the human review process. This lack of transparency raised ethical concerns, with experts arguing that patients should be explicitly informed when third-party workers listen to their conversations.

DeepScribe claimed that its quality-assurance team was disclosed in sales presentations and on its website, though much of its marketing emphasized the AI aspect of the product. Whether physicians and patients fully understood the extent of human involvement remained unclear.

DeepScribe’s Expansion into Oncology with Flatiron Health

Despite these challenges, DeepScribe continued to expand its reach. The company recently announced a partnership with Flatiron Health, a health-tech firm specializing in oncology care. Through this collaboration, DeepScribe’s AI integrated with Flatiron’s OncoEMR platform, a cloud-based electronic medical records system used by over 4,200 providers across 800 community-based cancer care locations in the U.S.

By tailoring its AI to the complex workflow of oncology, DeepScribe aimed to reduce the time oncologists spent on documentation, allowing them to focus more on patient interactions. The AI captured key contextual details, interval history, and treatment plans while automatically organizing clinical notes with relevant ICD-10 diagnosis codes. According to DeepScribe, oncologists responded positively to the technology, with 91% calling it easy to use and adoption rates surpassing 80% within large oncology groups.

Leadership Changes and Future Direction

As DeepScribe continued its expansion, the company appointed Dr. Dean Dalili as its Chief Medical Officer. Dr. Dalili, an experienced hospitalist and healthcare executive, previously served as President of Market Operations at DispatchHealth and held leadership roles at Envision Healthcare and SCP Health. His background in managing physician networks across 800 hospitals and shaping healthcare policies positioned him as a key figure in refining DeepScribe’s clinical strategy and driving broader adoption.

DeepScribe reached a significant milestone when clinicians across the U.S. used its documentation solution to capture 1 million patient visits. This achievement made DeepScribe the most widely adopted ambient AI scribe solution in healthcare, marking a breakthrough in automating medical documentation.

“One of the things my co-founder and I love to say is that we’re not doctors, we’re not natively from medicine,” said Bapu. “That helps us bring in an outsider tech perspective and rethink these processes. Just asking ‘why’ helped DeepScribe be successful early on when a lot of folks native to healthcare took things for granted.”

Introducing Real-Time AI Documentation and DeepScribe Assist

DeepScribe introduced “Real-Time,” a groundbreaking addition to its platform that allowed clinicians to edit and clarify notes directly in their EHR during the patient encounter. This functionality, powered by DeepScribe’s proprietary clinical large language model, HealAI, represented a major advancement in AI-driven medical documentation. Unlike traditional AI documentation solutions that generated notes only after a visit, Real-Time enabled clinicians to receive immediate feedback, improving accuracy and efficiency.

DeepScribe also launched DeepScribe Assist, an ambient intelligence solution designed to enhance documentation and coding compliance. By leveraging structured data from patient conversations, DeepScribe Assist provided real-time feedback and ensured that critical diagnoses were accurately documented. A key feature of DeepScribe Assist was its ability to surface relevant Hierarchical Condition Categories (HCCs) during patient visits, aiding in risk adjustment and value-based care.

To streamline adoption, DeepScribe announced its availability on AWS Marketplace. This move allowed healthcare organizations to deploy DeepScribe’s AI documentation platform through a private offer, consolidating cloud spending and improving integration with AWS-based systems.

“Our inclusion in the AWS Marketplace is made possible by our work with AWS and the enterprise readiness of our AI clinical documentation solution,” said Ko. “The increased accessibility will enable more organizations to address clinician burnout by automating documentation.”

The Future of AI-Powered Healthcare

DeepScribe’s advancements, from reaching 1 million patient visits to launching Real-Time and DeepScribe Assist demonstrated its commitment to transforming medical documentation. By reducing administrative burdens, improving data accuracy, and enhancing compliance, DeepScribe was not just streamlining workflows but shaping the future of AI-driven healthcare.

Recently, in Stripe’s annual letter, the company stated that new industry-specific AI tools are helping players “properly realize the economic impact of LLMs and that the contextual, data, and workflow integration will prove enduringly valuable,” and DeepScribe is an example of this in healthcare.