Bhupesh Nadkarni
The Biggest Risk as AI Enters Care Delivery
AI in healthcare has moved beyond pilot projects; it's now being implemented at scale. AI is now integrated into clinical workflows, operational decision-making, and patient access points, but with significant consequences. Commonly cited risks include bias, privacy concerns, and hallucinations.
While these are significant, a more fundamental risk exists: deploying AI into production environments without governance robust enough to withstand clinical scrutiny, regulatory review, and public accountability.
My Perspective
I am Bhupesh Nadkarni. I have spent 25 years architecting solutions for healthcare technology.
Currently, I collaborate with healthcare leaders facing pressure to modernize, especially with AI. They require speed, precision, and systems that perform reliably in real-world settings. I also lead a team of 800 AI-enabled engineers who design and implement these systems at scale.
In my experience, the milestone in healthcare is not merely a model generating an answer. It is the organization's ability to defend that decision and make it auditable.
Trust Is the Production Requirement
Healthcare is sustained by trust rather than novelty.
Clinicians place trust in systems they can validate, patients in those they can understand, and regulators in those that are traceable, auditable, and accountable. If AI-integrated systems do not meet these standards, they should not be deployed in production, regardless of their performance in demonstrations.
To be credible, AI must be built on three principles:
Transparency, Explainability, and Responsibility
These are not aspirational values. They are essential conditions for safe deployment and sustaining the scaling needs of the industry.
Pilots Are Controlled. Production Is Accountable.
Pilot environments allow for greater flexibility, whereas production environments do not.
During pilot phases, oversight is intensive, volumes are limited, edge cases are managed manually, and risks are contained.
In production, AI becomes part of how care is delivered and how decisions are made. It can influence documentation quality, coding integrity, utilization decisions, and patient access. It also creates records that can be reviewed months later under audit pressure.
Consequently, the central question shifts. The question is no longer whether the model can generate an answer. Instead, it is whether the outcome can be justified to clinicians, patients, and auditors.
If this question cannot be answered with clarity, the system is not ready for production deployment.
Governance Is the Gap That Creates Real Risk
When AI capabilities advance more rapidly than governance structures, predictable failure patterns emerge.
Accountability Becomes Unclear
When an AI recommendation influences a decision, ownership must be explicit. In healthcare, blurred accountability quickly becomes unacceptable.
Exceptions Become Commonplace
Production workloads regularly reveal edge cases. Without clear escalation procedures and defined decision rights, teams may either over-rely on the tool or abandon it entirely. Both scenarios compromise safety and hinder adoption.
Auditability Is Compromised
If an AI output cannot be linked to its inputs, underlying guideline logic, and supporting evidence, it becomes indefensible. At this stage, the system ceases to provide decision support and instead introduces operational risk.
Governance is essential because it safeguards care delivery while enabling innovation to scale.
Being Transparent, Explainable, and Responsible Is How AI Earns Clinical Confidence
Transparency
Transparency requires that a system's behavior is visible, including the Patient Health Information (PHI) utilized, inaccessible data, underlying assumptions, and existing uncertainties.
Explainability
Explainability ensures that the rationale is immediately clear to clinicians. The system should reference the relevant clinical context, applicable guidelines, and factors influencing its recommendation, and indicate what could alter the output.
Responsibility
Responsibility requires the health system to operate within established boundaries, respect defined roles and approvals, support overrides and escalation, and be monitored in production based on real-world outcomes rather than solely on model metrics.
If these conditions are unmet, confidence will erode in clinical, operational, and audit contexts.
The Non-Negotiable Questions Production AI Must Answer
Before AI can be considered production-ready, governance must address several unavoidable questions:
- Does this recommendation tie back to the care guidelines clinicians follow?
- Can the rationale be explained to a clinician in plain language?
- Can it be explained to a patient without causing confusion?
- Can it be defended in an audit with evidence and traceability?
- Who owns the decision when AI is involved?
- What happens when the system is wrong?
- What happens when inputs are incomplete, outdated, or conflicting?
If these answers lack clarity, scaling AI will increase risk rather than mitigate it.
What Strong Governance Looks Like in Practice
Strong governance lives in real workflows. Policies only matter when they are enforced.
- Clear ownership for each AI-enabled workflow across clinical, operational, and technical responsibilities.
- Defined decision boundaries so it is clear where AI can recommend, where it can act, and where it requires approval.
- Case-level traceability, including sources, guideline references, and context used.
- Readable audit logs that compliance teams can review without translating engineering artifacts.
- Continuous monitoring tied to operational and clinical outcomes, with thresholds that trigger review and correction.
This approach enables healthcare organizations to scale AI while maintaining control.
The Future Belongs to Systems That Can Be Defended
The future of AI in healthcare will be defined by systems that consistently deliver safer care and more efficient operations under real-world conditions, not by impressive pilots.
When governance is robust and guiding principles are explicit, AI transitions from being perceived as a risk to becoming a dependable asset.
That is how we move from hype to impact.
That is also how we earn lasting confidence from the people delivering care every day.
