Bessemer Venture Partners released a new report that identifies the most valuable directions for healthcare AI startups, the best addressable markets for healthcare AI and the VC firm's own investing criteria.
There's been a surge in AI-enabled healthcare startups: In 2024, 38% of digital health companies that raised series A rounds were AI-enabled, according to Rock Health data, showing a clear top-down prioritization of AI in the healthcare sector.
Healthcare AI also is seeing tangible success with an exponential rise in Food and Drug Administration-approved AI/ML-enabled devices, 5x more in 2023 than in 2018, according to the Bessemer report.
A McKinsey survey found that 70% of payers and providers reported pursuing implementation of generative AI. Three-quarters of these same groups have increased IT investments over the last year and expect this trend to continue, a Bain & Company survey found.
One challenge is that healthcare markets are fragmented, with many sub-verticals each offering around $1 billion in total addressable market (TAM). This opens up opportunities for AI companies that can thrive within a highly specialized landscape, Morgan Cheatham, vice president at Bessemer Venture Partners, and colleagues wrote in the report.
AI has the potential to transform the healthcare industry when used innovatively, according to the venture capital firm. It pointed to existing AI companies Abridge, Subtle Medical, OpenEvidence and Atropos, which it says have begun to reconstruct healthcare categories with their solutions.
Bessemer identified six areas where it is interested in investing: interactive systems, multimodal technologies, simulation, evaluation infrastructure, sensors and specialty foundation models. This includes AI solutions with dynamic and real-time interaction; virtual environments that simulate healthcare for risk-free training; standardized frameworks to evaluate AI; advanced data capture; and the ability to augment or automate specialized tasks.
Combining multimodal data for novel data analysis is an area ripe for development. The biggest opportunities for healthcare AI companies are aligning their chosen modality and business model, leveraging multimodal AI and building vertical specific infrastructure, the report says.
Healthcare AI companies should assess how the modality of an offering, and the pricing approach, will impact the market, Bessemer partners wrote. For example, AI solutions already come in myriad forms, such as AI-enabled software, AI copilots, AI agents and AI-enabled services. The potential revenue generation for each modality depends on the business model.
The most common business models for AI solutions are usage-based pricing and performance-based pricing, the report found. For example, healthcare technology companies can either charge customers for the number of times they use the model or take a cut of the savings that are generated from deploying the technology.
AI companies should consider how their business model and AI model modality will impact their total addressable market and revenue generation.
“We predict that the most successful healthcare AI companies will focus on critical junctures where high volumes of valuable data are generated, upstream of essential workflows,” Cheatham and Bessemer Partner Steve Kraus wrote in the report.
Healthcare AI companies should also consider how to pair multimodal data for novel discoveries. Healthcare data are abundant and can be tapped through clinical records, medical imaging, audio, video, patient-reported outcomes, wearable device data, time-series information, exposure data and genetic sequencing results, the report says.
Advanced AI technologies that can draw insights from multimodal data will be some of the most valuable solutions, Bessemer says.
“When rich biomedical data is combined with population-level information, operational insights, and financial metrics, AI can not only diagnose diseases earlier and with greater precision, but it can also identify factors preventing patients from getting better as well as opportunities for enhancing the function and efficiency of the healthcare system” Cheatham and Kraus wrote.
Bessemer also says that there is opportunity in vertical specific infrastructure for certain issues in healthcare that lack solutions. For instance, healthcare companies continue to struggle with cybersecurity, and healthcare data are targeted by online criminals, but cybersecurity solutions often are not built for the healthcare market.
The popularization of AI also necessitates greater data generation and management capacity. Companies like Protege, Gradient and Omny Health are making headway, the report says; however, Bessemer predicts that the market will only continue to need more health data for training and testing models beyond what these companies currently provide.
One particular problem Bessemer cites is a need for a better way to de-identify and re-identify patient data at scale.
Benchmarking, model monitoring and governance are three areas that are also ripe for innovation. Bessemer sees a need for more solutions to measure model performance in real-world environments. It is also concerned with the impact model drift could have at scale.
AI assurance labs and implementation science centers are beginning to address the issue of AI oversight, the report noted.
The healthcare industry needs to stay aware of good and bad evolutions in AI technology. Healthcare AI can learn from the genomics industry, which has been designed to communicate when new variants of diseases are detected, the Bessemer report authors noted.