Healthcare artificial intelligence sits at an inflection point. The pace of change and innovation for AI is moving at a rate we have never seen before, and as advanced decision-making becomes more rapidly accessible, healthcare organizations are mobilizing to adopt the necessary infrastructure and cutting-edge tools to move their workforce towards a highly efficient, AI-powered workplace.
For many in the market, it is unclear how the best companies are building winning healthcare AI products. Large models adapt by the week, and AI labs like OpenAI, Anthropic, and Google seem to be displacing entire startup categories every month. We find ourselves in an ambiguous moment for product building, as no doubt AI in 2026 and 2027 will look starkly different from the applications that we’re seeing today. This transformation is creating a new guard for how companies build and deploy products, as well as how they fundamentally operate, think, and act when solving their most pressing problems.
With so many companies (both public and private) moving to adopt AI and build their own AI-native products, we’ve gathered experts leading the charge in this regard to learn from them and understand how they think about building winning AI products in healthcare.
A Recap:
Last year, we saw healthcare technology funding rise above the $10 billion mark with an estimated 42% of that capital going directly to AI investments. With an even higher estimated percentage of funding going towards AI through H1 than last year, enterprises and investors continue to show a renewed focus on administrative workflows, a trend that we anticipate will continuing throughout 2025.
As we have seen across several Flare Capital portfolio companies (highlighted in the chart above), the core business principles for AI-based administrative solutions have not fundamentally changed, even as the underlying AI technology evolves. Winning teams still lead with a clear ROI narrative, target <12-month payback periods, build durable product moats, and integrate tightly (and seamlessly) into existing clinical and back-office workflows.
Organizations are pursuing a healthcare AI profit pool that McKinsey pegs at roughly $360 billion in annual value, as deep-learning breakthroughs and transformer-based LLMs give way to autonomous reasoning engines and agent-based workflows. Early-stage VC dollars and enterprise pilots are already shifting toward these next-gen capabilities (with more detail forthcoming in Flare Capital’s future AI reports).
Yet many health systems, payers, and life sciences companies remain stuck in a maze of sub-scale pilots launched primarily to maintain a competitive edge. Interviews with our executives across the sector have revealed that AI adoption may actually be advancing more slowly than anticipated, with less enterprise-wide impact than initially promised. Reasons for this include large data silos, integration headaches, and a persistent trust gap that continue to prevent the jump from proof-of-concept to production.
Adding to these challenges is an increasingly uncertain regulatory climate. Looming federal action (i.e., the anticipated “Big Beautiful Bill”) plus state-level regulatory rules and medical boards have prompted organizations to adopt more of a “wait-and-see” stance. Even still, capital keeps flowing: private-market investors are rewarding startups that can pair differentiated technology with seasoned/tech-forward teams. The result: category leaders are beginning to surface in key niches within the AI market.
Amid these changes, one question was fundamental to our discussion: how should companies design and scale world-class healthcare AI products to thrive in this environment?
Navigating the Dynamic AI Landscape
This year, we sat down with experts from across the healthcare AI industry as part of our Expert Roundtable series on Building Winning Healthcare AI Products. We heard directly from the experts who shared their thoughts and insights on what it takes to build winning AI products in the healthcare space, and where the industry might be headed.
The discussion saw an impressive panel of founders and operators debate live topics, such as technological moat, how to properly integrate into clinical workflows, and how best to navigate product builds for startups.
The first panel saw a group of three founding CEOs of early-stage healthcare startups. The second panel included two seasoned healthcare AI veterans, with direct oversight of AI implementation and evaluation within large organizations.
In our first panel, a few key paradigms emerged from the discussion:
When it comes to differentiation, our experts shared ways in which they see companies differentiating amidst the competitor-rich environment:
The Flare Capital team expanded on certain expert insights above*
As for how a product can differentiate, true defensibility in healthcare AI comes from mastering the “last-mile” problems that generic models leave unsolved. That means building pipelines that ingest messy clinical data (e.g., EHR, imaging, claims) and normalizing it to standards (i.e., FHIR). By layering rigorous validation into the process of product build, every prediction can then meet clinical safety and regulatory thresholds. Additionally, the latency, uptime, and privacy safeguards must be engineered into the tech stack such that care teams can rely on the system in “real” workflows. If an insight arrives too late, or can’t be audited by a healthcare professional, it’s useless.
At the same time, our panelists shared how platforms themselves must stay model-agnostic to some degree. Foundation models are advancing on a monthly and quarterly cadence, so the architecture should be able to support hot-swapping new checkpoints, A/B testing them against gold-standard datasets, and then rolling back if the metrics regress.
In practice, this calls for abstraction layers (e.g., model routers, feature stores, and evaluations) to decouple product logic from any single vendor. The winners here will be teams that treat data quality and clinical validation (along with continuous model governance) as core competencies rather than one-off tasks.
Generative AI has fundamentally shifted our day-to-day operations, pushing healthcare toward a new baseline in which clinicians and algorithms can collaborate/co-manage care. We already see ambient clinical-documentation copilots drafting visit notes in real-time, freeing physicians from keyboards; radiology suites where generative models can auto-compose preliminary imaging reads; and revenue-cycle models that can assemble prior-authorization packets and slash administrative lag.
Yet adoption still collides with the challenge of siloed data, regulations, legacy EHR transitions, and a lack of trust around “black box” outputs from different AI companies. Most decisively, the incentives in healthcare remain fractured between IT budgets, clinical needs, and payer contracts each pulling in different directions, which can cause pilots to stall when the value doesn’t accrue evenly. Breaking through this requires governance models that share upside across departments and technical architectures that make high-quality data, compliance, and explainability the “default” rather than the exception.
Translating Data into Insights, and Insights into Action
Leading startups can turn their relationships with stakeholders in the organizations from use case or SOW-driven into longer-term, actionable insights, which ultimately extend well beyond their initial scope of work. Startups must begin developing a trust-based relationships before parlaying the opportunity to solve the organization’s next, ideally larger, hair-on-fire problem.
When asked about tracking and measuring success, our experts shared some great insights below:
As we look outside of healthcare, product leaders across OpenAI and Anthropic for example are echoing similar guidance for startups building in this dynamic environment. Taking tangible steps — such as preparing teams to write increasingly more effective AI evaluation tests – will become the de-facto skills for product builders as teams build, improve, and trust AI systems not to make mistakes, especially with autonomous workflows.
Our roundtable experts noted a similar shift and described ways in which they’re developing smaller clusters of specialized models to work either in tandem or finetuning model to keep compute cost low without sacrificing quality output. An example of this in action includes Microsoft’s healthcare agent orchestrator, piloting with Stanford Health Care to enable data scientists and developers building AI agents for oncology tumor boards. Despite this vision, examples of this work are still early.
As we pan out, the fact is, today’s best AI model is the worst we’ll be using going forward.
Product teams must continue to build products with this in mind. As we heard from one of our panelists, Suchi Saria, who draws on experience across Bayesian Health, Coalition for Healthcare AI (“CHAI”), and universities Johns Hopkins and Stanford; Suchi highlighted how solutions should emphasize product design such that the infrastructure is ahead of the model that exists today, without over-promising features to the buyers, and expect that the pace of model innovation will continue to climb.
Core Capabilities for Safe, Scalable AI Products
In our second panel, we were joined by two senior executives from Flare Capital’s strategic Limited Partners, Lana Bender, VP AI and Behavioral Science at GuideWell and Dr. Barry Stein, VP and Chief Medical Informatics and Clinical Innovation Officer at Hartford HealthCare. Both drew on their experience deploying AI solutions across payer and provider organizations for 20+ years.
Barry and Lana shared how their organizations are thinking about building solutions, and the often-under looked aspects that startups miss (or must get right) when integrating across large enterprises.
Insights included:
Above all, perhaps the “ecosystem partnerships” piece is the more ambiguous task to achieve for early-stage startups, especially since these startups may be spinning out of great organizations or already have tech talent, making it seem like an unnecessary endeavor. Meanwhile, we’ve seen large organizations developing their own direct partnerships with larger enterprises (i.e., NVIDIA, OpenAI, Anthropic) and expanding on existing relationships (i.e., Microsoft, Salesforce). When startups fail to recognize the importance of scale through channel partnerships, it can cause that organization to get left behind.
To capitalize on these capabilities, startups must design engagement models that align incentives across payers, providers, and technology vendors from day one. That means being able to structure contracts that share upside on clearly defined clinical and financial outcomes – and embedding co-development milestones to give the enterprise partners real “skin in the game.” Furthermore, it’s important that startups select channel partners whose data assets and distribution help shorten time-to-impact. In practice, what this looks like could be coupling a rapid-cycle pilot (e.g., a 60-day deployment) with a pre-negotiated expansion path, contingent on hitting adoption and outcome thresholds.
By pairing an evidence-based commercial model with the pillars outlined above, startups can shift AI projects from isolated proofs-of-concept to more enterprise-wide programs that deliver compounding value.
Observations
If healthcare is going to keep pace with the broader AI ecosystem, the next eighteen months should be treated as an execution window for startups and organizations alike. We expect AI budgets will continue to rise, but capital alone will not be enough to bridge the gap between breakthrough model performance and organizational impact. Success is going to hinge on an organization’s ability to pair the larger spend commitments with disciplined governance, data interoperability, and incentive-aligned business models that translate AI-efficiency into measured outcomes – both clinically and financially.
Startups and organizations that can institutionalize these frameworks (rapid-cycle product evaluation, partnerships to extend reach, and hard-wired trust and transparency into workflows), as shared by our experts, are the ones that will successfully convert pilot bets into enterprise-wide value. Those that delay will face a widening performance divide as model capabilities and company expectations continue to accelerate.
Summary
What did we miss? We’d love to hear from you, the teams and AI product builders, working on these solutions. Reach out to Jon George or any member of the Flare Capital team to continue the discussion.
Be sure to join us for our Flare Capital Expert Roundtable Series each quarter as we discuss the important topics impacting our healthcare system.
Stay in touch with Flare Capital:
· Subscribe here for more insights
· Follow us on LinkedIn, YouTube, Spotify, Instagram, X (Twitter), and our website.
