The following is a guest article by Sathiyan Kutty, Chief AI Officer at Emids
Healthcare has invested billions of dollars in artificial intelligence pilots over the past several years, yet only a small percentage ever make it into full production. The problem is not a lack of powerful algorithms. It is that healthcare is one of the most context-heavy industries in the economy, shaped by layered workflows, regulatory guardrails, reimbursement complexity, and high-stakes clinical decisions that most AI systems are not built to navigate. That struggle is not unique to healthcare — a recent MIT study found that nearly 95 percent of generative AI projects fail to achieve a meaningful return on investment globally — but the stakes in healthcare make it especially consequential.
A Stanford University study found that while automation is reshaping entry-level roles, demand for human oversight and domain expertise is rising. AI adoption, in other words, does not automatically translate into AI transformation. Without the right operational grounding, deployment stalls.
Why Traditional Delivery Models Fall Short
For decades, healthcare organizations have relied on traditional system integrators to execute large-scale technology initiatives. That model works well for infrastructure modernization or enterprise EHR deployments, where requirements are defined up front and customization is expected. But AI is fundamentally different. It is dynamic, continuously learning, and deeply intertwined with workflows that evolve daily.
The Healthcare AI Adoption Index published by Bessemer Venture Partners reports that only about 30 percent of AI pilots in healthcare successfully transition into production environments. This is not simply a technical failure. It reflects the difficulty of encoding healthcare’s operational nuance into scalable systems. Data remains fragmented across electronic health records, imaging systems, claims platforms, and patient engagement tools. Regulatory requirements shift at both the state and federal levels. AI models trained without deep awareness of these realities often perform well in controlled pilots but break down under real-world pressure.
Governance gaps further widen the divide. Reporting from HIMSS indicates that while approximately 88 percent of health systems have experimented with AI, roughly 80 percent lack mature governance frameworks to oversee its deployment. Without structured oversight, auditability, and accountability, AI initiatives remain isolated experiments rather than enterprise-grade capabilities. In a live clinical setting, even a technically sound model must be explainable, traceable, and compliant.
The Rise of Forward-Deployed Context Engineers
Closing this gap requires more than additional data scientists. It requires a different way of building and deploying AI.
Forward-deployed context engineering (FDCE) extends the Service as a Software concept into a delivery model. Instead of building AI systems in isolation and deploying them into complex environments, FDCE embeds domain experts directly within live workflows to continuously refine how systems interpret data, apply policy, and generate outputs. This approach collapses the gap between development and operations, enabling AI systems to evolve alongside the environments they operate in.
Where Context Actually Lives
These experts operate at the point where AI meets real-world execution. They sit at the intersection of clinical workflows, reimbursement logic, compliance policy, and technical implementation. Their role is not simply to improve model accuracy, but to ensure that outputs align with how care is delivered, documented, reimbursed, and audited in practice.
From Context to Execution
FDCE provides the mechanism to bring these layers together into the AI lifecycle. This ensures that decisions are not only accurate but aligned with how work actually gets done.
This distinction becomes clearer when applied to real workflows.
In practice, this becomes most visible in high-friction workflows such as prior authorization for payer organizations. An AI system designed to support prior authorization cannot rely on clinical guidelines alone. It must account for plan-specific policies, CMS mandates, documentation completeness requirements, provider submission patterns, and turnaround time SLAs.
With FDCE embedded into the development lifecycle, these variables are translated into the system itself. The result is not just automation of intake or triage, but a system that can dynamically prioritize cases, identify missing documentation based on policy logic, and surface recommendations aligned with both clinical intent and reimbursement rules. Without that grounding, outputs remain disconnected from real-world execution.
On the provider side, the same principles apply within clinical and revenue cycle workflows.
In clinical documentation, AI systems can analyze physician notes, identify gaps, and suggest improvements aligned with coding and billing requirements. When grounded in context, these systems reflect specialty-specific workflows, payer expectations, and audit standards. The result is improved documentation quality, reduced rework, and faster reimbursement cycles without increasing clinician burden.
Across these workflows, early implementations are beginning to show measurable impact. Organizations are reporting reductions in manual intervention, improvements in turnaround times, and greater consistency in audit outcomes. While results vary by workflow and implementation maturity, the pattern is clear. Systems that embed context into decision-making deliver more reliable operational outcomes than those that do not.
From Projects to Platforms
Another shift is underway in how AI is delivered and commercialized in healthcare. Organizations are moving away from one-off project development toward software-led platforms infused with domain intelligence. Rather than building bespoke tools that require extensive customization, vendors are packaging reusable capabilities with embedded compliance guardrails and workflow integrations.
These developments reflect a broader evolution in how AI operates within enterprise environments.
Agentic AI represents a shift from passive intelligence to active orchestration. Unlike traditional automation or AI copilots that assist with individual tasks, agentic systems can execute multi-step workflows, adapt to changing inputs, and coordinate actions across systems within defined guardrails. In healthcare, this means moving from isolated recommendations to systems that can triage, route, validate, and escalate decisions while maintaining human oversight and regulatory compliance.
In practice, these systems operate as coordinated layers, where context informs decision engines, decisions drive workflow orchestration, and every action remains traceable through audit and feedback loops.
This reframes AI not as a project with a start and end date, but as an ongoing capability that learns from usage patterns and adapts alongside policy and workflow changes. It also reshapes commercial incentives. Contracts increasingly tie value to measurable outcomes such as fewer claim denials, faster chart completion, and reduced administrative burden rather than hours billed.
Why This Moment Matters
Healthcare does not lack experimentation. It lacks scaled execution. Each stalled pilot represents not just sunk cost, but growing scepticism among clinicians and executives who have seen promising demonstrations fail to translate into durable results. In a system where administrative tasks already consume a substantial portion of clinicians’ workdays, contributing to burnout and workforce shortages. AI deployed without context risks becoming another layer of complexity rather than a meaningful reduction of it.
What distinguishes organizations that move from pilots to production is not technological novelty. It is their ability to integrate operational context into deployment, governance, and accountability structures from the outset. Systems built with these realities in mind anticipate workflow constraints rather than discovering them late. Compliance is embedded rather than retrofitted. Learning occurs continuously within live environments rather than in isolation.
The organizations that succeed will not be those that deploy the most AI, but those that design systems where AI can operate safely within the realities of healthcare. The future will not be defined by model sophistication alone, but by whether those models can act, adapt, and be trusted within the workflows that define care.
About Sathiyan (Seth) Kutty
Sathiyan (Seth) Kutty is the Chief AI Officer at Emids, where he leads AI-driven innovation across healthcare payer, life sciences, and health tech markets. With over two decades of experience spanning analytics, AI, and technology-led growth, Seth has built a reputation as a sharp and pragmatic leader in the field.
His career spans some of the most recognised names in global technology and business, including Kaiser Permanente, Tesla, VMware, Intuit, and IBM Consulting, where he worked closely with C-suite leaders to translate advanced analytics into real business outcomes: revenue growth, operational efficiency, and market expansion.
Beyond his corporate career, Seth is a repeat entrepreneur. He founded and scaled a data and AI services company that went on to achieve a profitable exit, bringing a founder’s mindset to every leadership role he takes on.
Seth holds a Bachelor of Science in Electrical Engineering and Computer Science, and a Master of Science in Industrial and Operations Engineering with a specialisation in Operations Research from the University of Michigan. This blend of hands-on experience and academic grounding shapes his approach to building scalable, outcome-oriented AI platforms that deliver lasting value.
About Emids
Founded in 1999 and headquartered in Nashville, Emids is a leading global provider of digital transformation solutions across the healthcare and life sciences ecosystem. We deliver AI-led engineering, data, and platform services powered by healthcare-trained ontologies, context-aware intelligence, and our uniquely embedded Forward-Deployed Context Engineers (FDCEs). Our Service-as-Software approach helps payers, providers, biopharma, medtech, and health tech companies modernize operations, activate high-impact AI use cases, and deliver outcomes with speed, precision, and trust. #health
About John Roach