Interoperability ≠ AI-Ready

Healthcare’s AI revolution is moving at full speed.

Hospitals, insurers, and life sciences companies are all racing to integrate AI into workflows that improve patient care, optimize operations, and accelerate drug discovery. The market opportunity is massive—AI in healthcare is projected to reach $490 billion by 2032, driven by rising demand for predictive analytics and automation.

But despite the enthusiasm, one major challenge remains: the data isn’t AI-ready.

The Messy Reality of Healthcare Data

Right now, 80%+ of hospital data is unstructured. Clinical notes, imaging reports, patient histories—they exist in a mess of PDFs, free text, and fragmented systems. AI can’t learn from chaos.

To illustrate the point, here’s an extremely simplified example of unstructured data:

Relatively easy for a human to understand, but extremely difficult for a computer to make this information useful.

Interoperability: The Healthcare Data Problem Everyone Acknowledges

Naturally, the industry has focused on interoperability—standardizing data formats so that systems (healthcare providers, payers, AI systems, etc) can talk to each other. Interoperability is the essential first step because without it, data remains siloed and inaccessible. By creating a common language for healthcare data, interoperability lays the foundation for AI adoption—ensuring that information can flow freely between systems before it can be structured, labeled, and made AI-ready.

FHIR, HL7, data fabrics—this is where most health IT investment has gone. The idea is that if we just make data more accessible, everything else will follow.

Here's how the same unstructured data would look if it were structured in a traditional table and as FHIR-compatible data, using the example information from before:

Much easier for software to understand! But there’s a problem:

AI Needs More than just Structure

FHIR and HL7 help move data, not prepare it for AI.

Structured data isn’t the same as labeled, contextualized, AI-ready data. While structured data follows standardized formats like FHIR and HL7, it primarily ensures that information is organized and shareable between systems.

However, AI models require more than just structured data—they need clear connections between different data points, meaningful categorization, and additional processing to make sense of complex clinical information. Without this extra layer of refinement, AI struggles to draw useful insights, detect trends, or make reliable predictions from healthcare data.

Take autonomous vehicles. In the early days, companies had millions of hours of raw camera and LiDAR footage. But that footage wasn’t useful for AI until it was labeled, structured, and annotated—which is why companies like Scale AI emerged to fill the gap.

Healthcare is in the same position. Hospitals have tons of data, but it’s not ready for AI models to learn from.

The Missing Layer Between Interoperability and AI

Making the leap from interoperability to AI-ready data requires more than a format change—it requires an entirely different infrastructure approach that includes context, AI integration, and compliance.

Even if existing interoperability players pivot, they face major hurdles:

• Ensuring AI-ready data includes context – Quantitative data alone is not enough; AI-ready data must also incorporate qualitative insights, such as physician notes, patient-reported symptoms, and real-world evidence. Without this, AI models lack the nuance needed for accurate predictions and insights.

• Developing AI models alongside AI-ready data – AI-ready data is only truly valuable if the company has the capability to build and refine AI models that can leverage this structured information. Companies that only focus on interoperability may struggle to transition to AI-first solutions without deep expertise in model training and validation.

• Navigating regulatory and compliance complexities – Unlike interoperability, which primarily focuses on ensuring data flows between systems, AI-ready data requires stricter compliance considerations. This includes privacy-preserving techniques, bias mitigation, and alignment with evolving AI-specific healthcare regulations, which interoperability-focused players may not be equipped to handle.

Who Wins? The Builders of AI-Ready Healthcare Infrastructure

The winners in this space aren’t just focused on moving data. They’re solving the fundamental problem that keeps AI from working in healthcare.

Health systems need an AI infrastructure layer that takes them the last mile—turning messy data into structured, AI-ready intelligence. This means:

Handling Multimodal with a structured approach

Healthcare data comes in many forms and require different handling methods. Clinical notes contain essential context but are unstructured, imaging data requires complex annotation, and real-world patient-reported data often lacks standardization. Without a structured approach, these fragmented datasets remain disconnected, limiting AI's ability to derive meaningful insights.

The key to solving this is a multi-step pipeline that:

• Structures and labels raw clinical notes using NLP and AI-driven annotation.

• Enriches imaging and sensor data by linking it with clinical histories.

• Creates a unified, AI-ready dataset that ensures models can extract patterns across multiple specialties.

This approach makes AI models more effective by providing them with a comprehensive, context-rich dataset rather than isolated data points. The result is improved predictive accuracy, more reliable decision-making, and greater adoption of AI-driven solutions in healthcare.

Enhancing AI Models with Reinforcement Learning Human Feedback (RLHF)

AI models must continuously evolve to keep pace with real-world clinical complexities. Reinforcement Learning Human Feedback (RLHF) enables AI to improve over time by incorporating direct input from clinicians who validate and refine its outputs.

With RLHF, AI models are not just trained once; they undergo an iterative learning process where:

• Clinicians review AI-generated insights and provide corrective feedback.

• The model adjusts based on real-world validation, improving accuracy and reducing errors.

• New data refines performance, ensuring AI evolves alongside medical advancements.

This approach creates AI systems that are more precise, context-aware, and adaptable across multiple specialties. Unlike static AI models, RLHF ensures continuous learning, making AI-driven clinical insights increasingly reliable and valuable to healthcare providers.

Engaging both technical and clinical stakeholders with specialty-tuned AI models

CIOs need structured data that supports AI applications across the organization, but the frontline clinicians are the ones driving specific use cases. Companies that can not only structured data but also develop AI models that are fine-tuned for specialty applications help streamline adoption by ensuring that AI delivers immediate value to the clinicians using it.

To bridge this gap, specialty-tuned AI:

• Aligns with existing clinical workflows – AI models trained on specific specialties integrate seamlessly into day-to-day decision-making, minimizing the learning curve for clinicians.

• Reduces the burden on health IT teams – Instead of requiring custom-built AI for each department, hospitals can deploy pre-trained specialty models that require minimal adjustments.

• Ensures faster adoption – Clinicians see immediate value from AI models that are already optimized for their specialty, making them more likely to engage and advocate for broader adoption.

A scalable AI strategy isn’t about replacing clinical expertise—it’s about enhancing it with AI models that work the way clinicians do. Organizations that take this approach will see AI adoption accelerate across departments without friction or resistance.

What Does This Mean for Investors?

For investors looking to back the next wave of AI-driven healthcare innovation, the companies that will stand out are those that solve the last-mile problem of AI adoption.

Companies that only clean data, or only build AI models will not be enough. The winners will build the foundational AI data layer for healthcare providers by:

• Cleaning data and building AI models side by side.

• Creating a Continuous Data Pipeline by structuring historical data and actively enriching and curating real-world clinical data, ensuring renewable datasets that improve the AI models over time.

• Create the "Operating System" for AI in healthcare, allowing hospitals to integrate AI solutions at scale rather than relying on fragmented, specialty-specific tools.

Please subscribe to our newsletter if you haven’t, and share our newsletter with a friend. Stay tuned to our newsletter for more insights into healthcare innovation!

Join us at The Healthcare Syndicate as we back the most ambitious founders 10Xing the standard of healthcare!