Calling Code: AI Healthcare Tools

Healthcare’s LLM moment is well on its way, with Generative AI well placed to reduce labour costs and increase efficiency in this famously labour-intensive industry. In the US, Gen AI could lead to 5 to 10% in savings, translating into $200 billion to $360 billion annually, by automating documentation, research and diagnostic assistance, creating bespoke reports, and facilitating round-the-clock AI healthcare support. In the long term, this impact could go beyond efficiency gains and lead to significant improvement in patient outcomes.
Introducing new tech in healthcare has unique challenges – accessing data at scale, integrating into workflows, navigating regulation, and building trust with clinicians. However, Gen AI has the potential to take on these challenges and disrupt the sector, evident in the investment Healthcare AI is attracting. Healthcare attracted the most amount of AI-related private investment across all sectors in 2022. Major tech companies such as Microsoft, Google, and Amazon Web Services are forging key partnerships with healthcare giants like Epic and Mayo Clinic to drive LLM adoption.
Use Cases: Improving clinician and admin efficiency is the primary focus area for LLMs currently
Healthcare tools address a broad range of use-cases, from pharmaceutical research to patient care and administrative tasks. Healthcare has long relied on Machine Learning to improve medical imaging, treatment prediction, and drug development. LLMs specifically (which are the focus of this article), are driving a new wave of innovation in the space, by optimizing effort-intensive processes that involve unstructured language-oriented tasks.
LLM adoption is the most prominent in provider tools (tools that support clinicians and administrators in delivering patient care)
- LLMs can boost productivity for healthcare providers by handling the bulk of language-related tasks: The healthcare industry is inundated with labour-intensive tasks across stages of clinical interaction, from scheduling to recording patient data to billing processes and follow-ups on treatment adherence. LLMs are particularly well-suited to streamline these processes due to their ability to process, interpret and act on large volumes of text. They automate tasks such as extracting key information from documents, categorizing and summarizing patient data, and even generating coherent narratives from unstructured clinical notes. Moreover, LLMs can streamline communication workflows, such as scheduling and patient engagement, by interpreting and responding to inquiries or instructions. This reduces the administrative burden on healthcare providers, enabling them to dedicate more time and attention to direct patient care.
- AI adoption/innovation in certain provider tools is relatively easier (compared to patient tools) due to lower regulatory constraints: Most provider tools using LLMs do not directly interact with patient treatment, rather provide insights, process support to clinicians, thereby requiring relatively less regulatory oversight or FDA approvals. This is a major factor in provider tools emerging as a early adopter of LLMs.
Early adoption of LLMs is being seen in tools for documentation, clinical decision-making, administrative workflows, and patient engagement
- Documentation (Medical Scribing, Medical Coding): Doctor-patient consultations create substantial manual work, especially in transcribing and coding these discussions into electronic health records (EHR). LLMs can transform physician notes and interactions into structured EHRs, and collate these notes, lab panels, and imaging to determine the right diagnosis codes. Tools like Deepscribe, Suki, Athelas, Abridge are already being used by providers across the industry, and claim to have reduced documentation time by 75-90%
- Clinical Decision-making (Care decision making, research & diagnostic support): Clinical decision-making requires healthcare providers to navigate through extensive medical literature and patient data to derive actionable medical insights for diagnoses and treatments. LLMs streamline this task by processing this information, generating key insights for diagnosis and treatment, and offering real-time updates and symptom-based diagnostic suggestions (of course, physician oversight on these suggestions is key). Notable LLM applications include Navina AI 's creation of comprehensive patient profiles from unstructured data,Glass Health 's AI-enhanced notebook for doctors, and Path AI's use of Generative AI in disease diagnosis via pathology slide analysis.
- Admin Workflows (revenue cycle ops, prior authorization): Administrative workflows in healthcare are notoriously paperwork-intensive, often mired in manual data entry, complex documentation, claims processes and reviews. LLMs are capable of automating form filling, streamline billing processes, and manage prior authorizations – reducing the manual effort required. Banjo Health's tool ‘Composer’ employs LLM-powered span recognition to identify clauses in clinical criteria documents and uses Generative AI to convert these into questions for clinical reviewers, accelerating the prior authorization process by up to 95%. Basys.ai is a recent entrant into the prior auth space and uses Gen AI to encode payer policies and guidelines, allowing automatic matching of requests with a policy requirements.
- Patient Engagement: Patient engagement involves effort-intensive tasks like data collection, patient triage, communication, treatment adherence and follow-ups, often leading to significant workloads for both patients and administrators. LLMs are capable of reducing this workload for patients and administrators. Notable Health has introduced an AI assistant for scheduling, bill pay and patient navigation. HealthNote has a chatbot which automates patient interactions pre and post appointments.
Market Map: Established as well as new AI-native startups innovating with LLMs
Emerging tools in the sector face high barriers to entry
The healthcare tools sector is an expansive one, and barriers to entry are high – its rigorous R&D demands and tight regulatory controls pose a tough environment for new entrants. Gaining traction among primary healthcare providers, such as doctors and hospitals, is particularly difficult because of the entrenched preferences for existing technologies and apprehensions regarding the risks and compliance issues associated with new technologies. This landscape poses a significant challenge for new entrants seeking market penetration. Similar to the legaltech sector, the high bar for accuracy in healthcare means mature startups are primarily driving LLM adoption instead of new entrants.
Established healthcare players are well positioned to adopt Gen AI into their existing tools
Recognizing the potential of LLMs, a number of industry incumbents such as Epic, Cedar, Paige AI and Augmedix are enhancing their software with GenAI/LLM capabilities. These established players, having already navigated the sector's regulatory complexities, are well-positioned for AI-driven enhancements. With their established customer networks, they also carry a data and distribution advantage.
New entrants are primarily innovating with Foundational Models and clinician tools
Most clinician tools using LLMs do not directly interact with patients or take any decisions, thereby requiring relatively less regulatory oversight or FDA approvals. Emerging startups are thus innovating with clinician tools which provide insights and support to providers (such as Ambience for documentation, Atropos Health for clinical research, Decoda Health for patient interaction), and healthcare-specific foundational models that power these tools (Hippocratic AI , OpenEvidence).
Funding Landscape: Robust funding activity in 2022-23, with 30+ deals across stages
The healthcare tools sector typically sees larger funding rounds than other enterprise solutions, driven by its advanced technology needs and intense R&D. The sector's complex use cases, strict regulations, and tough market entry conditions lead to investors placing focused bets on high-conviction companies.
The sector is seeing significant investment in established tools, specifically focused on incorporating generative AI
- Abridge, a startup converting patient-clinician dialogues into EMR-integrated notes, rasied $30M in its Series B funding, with Mayo Clinic as a strategic investor. Abridge’s AI automates over 91% of clinician notes across various specialties.
- Athelas hit unicorn status with its $132M Series B funding. Athelas uses generative AI in Athelas Scribe to document medical conversations and significantly cut the cost of clinical intake. Following its recent merger with Commure, Athelas is poised to further explore LLMs in healthcare .
~70% companies in the sector have raised funds in 2022-23
The funding landscape has been robust in the recent years. Among emerging tools, those spearheading the development of Foundational LLMs for the healthcare sector, like Hippocratic AI and OpenEvidence, have secured some of the largest funding rounds.
Active, yet concentrated funding landscape
Investors are making few high-conviction bets rather than many speculative ones. The sector shows a limited presence of unfunded startups – a deviation from typical trends observed in other AI tools sectors, indicating the formidable entry barriers that define this space.
Hippocratic AI, incubated at General Catalyst, is building a patient-facing healthcare LLM designed for non-diagnostic tasks such as dietary guidance, pre-op conversations, communication of negative test results and billing explanations. Hippocratic AI has developed a framework to incorporate professional certification, reinforcement learning from human feedback (RLHF) via healthcare experts, and bedside manner training into their conversational LLM.
OpenEvidence is an LLM that uses AI to aggregate, synthesize, and visualize clinical evidence for improved decision-making in healthcare. OpenEvidence, in partnership with global healthcare leader Elsevier Health, powers ClinicalKeyAI – an AI tool that supports clinical decision making at the point-of-care by providing quick access to the latest evidence-based medical knowledge through conversational search.
Ambience Healthcare, backed by the Open AI Startup Fund, has raised $36M funding, with a recent Series A round. Ambience's AutoScribe, an AI medical scribe, captures real-time clinician-patient conversations; supporting multiple languages, accents, and speakers. The tool allows clinicians to customize the charting style for better relevance to their function, and provides longitudinal context by integrating historical data from the EMR to capture all relevant info for the visit.
Employee Growth: Positive but slowing growth; Hippocratic AI & OpenEvidence expanding teams rapidly
Employee growth serves as a key barometer of organizational health and confidence in future growth, underlining the sector's momentum in innovation and market outreach.
The sector's employee growth has been consistently positive, despite some fluctuations. After a peak 9% increase in Q2’22, employee growth has stabilized, commensurate with the broader AI market trend where the initial AI hype is giving way to a focus on value. However, overall steady growth trend remains a positive indicator for the sector's health - with the last two quarters seeing nearly 70% of companies in the AI healthcare sector expanding their teams. The growth is relatively evenly distributed across departments like engineering, research, and sales & marketing.
Emerging AI-native startups building healthcare-specific Foundational Models, e.g. Hippocratic AI and OpenEvidence, have seen extraordinary employee growth, with >50% QoQ growth in Q3’23. Employee expansion at these startups has been concentrated mainly in healthcare services, engineering, and business development teams.
Looking Forward
- LLMs are rapidly being adopted in provider tools to boost healthcare productivity
Large Language Models (LLMs) are creating significant impact in provider tools, effectively handling complex, unstructured language tasks like documentation (Deepscribe), research (OpenEvidence) and patient interaction (Healthnote).
- Established startups are well-positioned to drive AI innovation in the healthcare sector
Established startups – having successfully overcome initial barriers to entry, navigated regulatory processes, and built loyal customer networks - are in the driving seat for LLM adoption. A substantial portion of investment is flowing towards these established players, particularly those enhancing their offerings with Gen AI capabilities, such as Abridge and Athelas.
- Emerging AI-native startups are primarily innovating on foundational models and clinician tools
Hippocratic AI and OpenEvidence are well-capitalized and seeing strong traction building foundational models for healthcare. Among clinician tools, Ambience Healthcare, Glass Health, and Atropos Health are tackling interesting use cases like medical documentation and clinical decision support.
- Big tech offerings in Healthcare are on the horizon, can potentially reshape market dynamics
Big tech has joined the bandwagon of Gen AI in healthcare – Microsoft's 2022 acquisition of Nuance Communications and 2023 AI collaboration with Epic, Google’s trial of MedPaLM-2 chatbot with Mayo Clinic are recent examples. How Big Tech approaches the market and its impact on emerging tools will be critical to follow.
- Concerns around unethical practices, cybersecurity, and bias will be important to address
Gen AI in healthcare has raised concerns around unethical data practices, privacy and cybersecurity risks, and potential biases against underrepresented communities. Addressing these will be crucial to drive adoption as regulatory frameworks around Gen AI take shape.