Sri Madabushi shares key takeaways from his discussion with experts Ajai Sehgal and Dr. Grace Terrell on aligning data and AI strategy from the technological and clinical perspective.
There is widespread enthusiasm within healthcare for AI’s potential to transform operations, reduce costs, and revolutionize patient care. Yet a stark reality remains: a large majority of AI initiatives struggle to move past the initial trial phase. Although opinions vary on the exact scale of the problem—with some leaders pointing to successful rollouts—there is a broader trend. Organizations often get trapped in experimental testing, draining budgets and executive support without ever delivering real value.
Balancing data architecture with clinical realities offers a clear lesson: healthcare AI falls short when leaders attempt to build advanced models on top of fragmented, legacy data systems.
I sat down with two of IKS Health’s experts at our Vitalize Summit, an invitation-only gathering of senior healthcare executives. Ajai Sehgal, Group Chief AI Officer, provided the technical data framework perspective, while Grace Terrell, MD, MMM, Chief Medical Officer, grounded the conversation in the realities of a frontline clinical practice.
Why AI pilots fail: Lack of a data strategy
When asked why promising AI applications routinely stall before enterprise deployment, both Sehgal and Terrell cited a disconnect between vendor promises and real-world infrastructure.
- Sehgal: Vendors often build point solutions using perfectly curated data. However, when a health system plugs that tool into its own architecture, the algorithm encounters a fragmented, ungoverned data ecosystem. AI doesn’t stop working when it’s given bad data; instead, it can confidently deliver a completely wrong answer that can be nearly impossible to detect.
- Terrell: The failure is one of historic design, not clinical vision. Healthcare IT originally grew out of legacy architectures built strictly for billing and scheduling. Electronic Health Records (EHRs) were data layers added on top of these financial engines, creating systems that were not designed for longitudinal patient care.
Ultimately, an AI strategy without a solid data strategy is bound to fail. Deploying AI point solutions without data governance and data availability leaves technology roadmaps without a foundation.
The contradiction: Too much data, not enough insight
I asked them to explain the healthcare paradox: why, despite the immense volume of data housed inside EHRs, organizations still struggle to extract meaningful insights.
- Sehgal: EHRs were designed as billing frameworks; they compress rich clinical narratives into flat diagnostic codes. To optimize operations using AI, health systems must extract data that extends far beyond the EHR—including supply chain metrics, vendor cost structures, and patient segments. To solve this, leading health systems are pulling their operational data out of disconnected vendor systems and bringing it together in the cloud.
- Terrell: Critical data remains locked away from the point of care. Information is rarely synthesized to offer proactive support. Instead, the legacy system retroactively penalizes a clinician if an individual patient is missing a specific prescription. Personalized healthcare will remain an impossibility with outdated software that enforces backward-looking compliance. Today, success requires freeing this trapped, isolated data and evolving past old computing models.
By merging disparate data streams into a unified cloud environment, organizations bring their data directly into the scalable infrastructure needed to drive generative AI.
The human impact of fragmented operations and data silos
When prompting Terrell and Sehgal to address the human impact of fragmented operations and data silos, I asked them to highlight the effects on clinician workload and attention, and why addressing these issues is a strategic responsibility for the CIO.
- Terrell: Physicians are drowning in transactional documentation. Specialists are increasingly forced to act as general internists because primary care physicians are completely bogged down by administrative chores. The true opportunity with AI is to re-engineer the clinical environment using human-centered design. Technology should operate seamlessly, where a user inputs a simple intent and an automated backend coordinates the execution. AI should automatically absorb more of the clerical tasks, returning doctors to the actual work of healing.
- Sehgal: Having a data-first strategy does not mean turning clinicians into technologists or data engineers. Instead, healthcare CIOs must move beyond mere accountability and actively champion the strategy behind backend data management.
By repositioning data governance as a value-creating engine rather than a compliance hurdle, CIOs can ensure AI operates seamlessly in the background, lifting the administrative burden so clinicians can focus purely on medicine.
Board alignment: Reframe AI around data and mission
I asked how CEOs should manage a board demanding immediate AI payoffs, while trying to justify investments in long-term data infrastructure first.
- Sehgal: Board members attend conferences, hear the hype around generative AI, and return believing they are falling behind. Executives need to reframe this narrative: a data strategy is the AI strategy. Because successful AI outcomes are impossible without accessible data, leaders must position data infrastructure investments as the only way to scale AI beyond a mere proof of concept.
- Terrell: Leaders need to speak directly to the board’s core mission. Board members in community and rural hospitals are thinking about economic survival, local jobs, and keeping their doors open. CEOs must explain that a robust data infrastructure is a strategic differentiator that directly protects the hospital’s long-term mission.
Healthcare IT has long focused on measuring the process of care rather than the actual results of that care. Aligning data strategies with precise clinical endpoints gives boards a clear, mission-driven reason to invest.
Actionable data-first strategies for healthcare leaders
As we concluded our discussion, I asked Sehgal and Terrell for actionable strategies to help leaders immediately sharpen their data-first focus. They agreed on the following:
- Execute a data completeness audit: Look past the AI demo and map out the required data signals. If that data isn’t already cleanly accessible within current systems, stop the pilot before investing capital in a guaranteed failure.
- Audit internal use and workarounds: Before deploying AI, audit the human design layer to understand how employees actually interact with current software, handle workflow variances, and use manual workarounds to bypass friction.
- Commit to a parallel, use-case approach: Instead of trying to clean an entire data warehouse at once, isolate a single, high-value use case and prepare only the specific data assets needed for that individual workflow.
Ultimately, healthcare AI falls short when it lacks a strong foundational data strategy. By designing a data infrastructure that respects the realities of frontline clinical practice and focuses on immediate, targeted wins, organizations can build a world-class system while delivering measurable financial and clinical outcomes.
Watch the panel discussion featuring these technology and clinical leaders as they share how to put these principles into action.