Why AI Fails in Surgical Care, and How to Get It Right

The following is a guest article by Dr. David Atashroo, Chief Medical Officer of Perioperative at Qventus

Automation and reliable improvement in surgical care operations are emerging patient safety obligations. The potential for AI to transform perioperative care and improve patient outcomes is no longer theoretical. But obligation needs a concrete strategy for execution. We as a field need to better understand why so many AI initiatives fail, and what it takes to succeed.

More than 80% of AI projects fail to reach production across industries, with similar findings in healthcare. For surgery specifically, the gap between promising algorithms and clinical impact remains large. Understanding why requires examination of not just the technology, but the organizational and operational contexts in which it must function.

Where AI Operationalization Fails

Many healthcare AI initiatives begin with impressive demonstrations of model performance. Leadership invests, IT deploys, and clinicians ultimately ignore the tool because it fails to fit existing workflows. Nothing changes, and the tool sits unused while the problems it was intended to address persist. This pattern repeats because organizations are often sold on what AI and machine learning can do rather than what clinical operations actually need.

An algorithm that predicts surgical complications provides no value if its outputs are not integrated into decisional workflows, or if clinicians lack trust in its recommendations. Though the technical achievements we see in model development are impressive and necessary, they are insufficient for operational impact. The difference between a promising prototype and clinical environment-ready software cannot be overstated.

Further, many AI solutions operate outside of primary clinical systems, creating additional work and untenable friction. When AI tools ask even more of healthcare workers already stretched thin, we cannot be surprised when adoption suffers.

Operational Design Principles

Algorithms Alone Do Not Drive Outcomes

Even the most advanced machine learning models are just a component of interventions driving real impact. Health systems must shift thinking from “What can this model do?” to “How exactly will this help our clinicians?” The technical sophistication of an algorithm matters far less than its meaningful integration into the reality of clinical practice.

Involve Clinicians Early in the Development and Implementation Process

Clinician involvement is essential for appropriate problem identification and assessing solution usability. Effective automation requires understanding not just of clinical context but the considerations and judgment that drives decision-making. We must think intentionally about which processes can be fully automated, when we need human oversight, and when we need escalation pathways. This simply cannot be determined without clinician input.

Confidence is Built with Guardrails

Effective guardrails must include mechanisms for ongoing monitoring and transparent communication of limitations. Models degrade over time as patient populations shift, clinical practices evolve, and patient population patterns drift. Continuous performance monitoring, with clear thresholds for intervention and retraining, is essential for building clinician trust.

Guardrails empower clinicians to understand and appropriately use AI tools rather than blindly trusting or dismissing them.

Make the Right Thing the Easy Thing to Do

AI interventions should reduce friction in clinical workflows and make the right decision the easy decision. Behavioral science has long proven that when the right action requires additional steps or cognitive effort, adoption will suffer regardless of potential benefits. Human-in-the-loop designs should present AI recommendations in formats that minimize effort to accept appropriate suggestions, while also allowing clinical judgment to override when needed. Regardless of format, the goal should always be the same: align the path of least resistance with optimal care.

Aim Before You Build

Picking the right use cases is paramount. Surgeon-executives and health systems constantly face limited bandwidth and constrained resources. A spray-and-pray approach will not deliver outcomes. Because of AI’s generative nature, organizations cannot abandon core principles of technology development – rigorous evaluation of data infrastructure, clinical champions, and governance capacity is essential. Prioritizing depth over breadth will generate clinical value and organizational learning to drive subsequent initiatives.

Seek Vendors Providing Deep Solutions

The right path forward requires clarity about what to expect from AI partners. Organizations should demand solutions that integrate deeply with existing workflows rather than asking clinicians to adapt to yet another standalone tool. Vendor solutions must demonstrate understanding of the specific clinical context in which they will operate, with clear plans for implementation support, ongoing monitoring, and drive for further improvement.

The Advantage of Integrated Delivery Systems

These principles reveal a structural insight: integrated delivery systems may be uniquely positioned to realize AI’s benefits in surgical care. Unlike many fragmented care environments, integrated systems can achieve data continuity across the surgical continuum – from preoperative evaluation, through the actual procedure, to postoperative follow-up and long-term outcomes – that enables comprehensive solutions that learn from complete patient journeys.

Clinical engagement and buy-in are also often more achievable within integrated systems. Surgeons, anesthesiologists, nurses, and care coordinators who work within a unified system rather than independent groups can collaborate, provide ongoing feedback, and participate in continuous improvement. This alignment enables the iterative co-design processes that are critical for successful AI deployment.

The value of this approach extends beyond implementation and into care delivery itself. Consider obesity care. Digital weight management platforms, compounding pharmacies, and direct-to-consumer prescribing services have expanded access and helped to engage patients. While these innovations fill important gaps, obesity is rarely a single-modality problem. Some patients benefit from medication management, while others require behavioral health support, nutritional counseling, or bariatric surgery. When care is delivered through isolated point solutions, recommendations are constrained by the tools available within that platform rather than the full spectrum of treatment options.

Integrated delivery systems offer a more holistic view. They can combine AI-enabled insights with a comprehensive set of clinical interventions and evaluate patients longitudinally across the continuum of care. In this model, AI helps guide patients toward the most appropriate intervention rather than simply optimizing a single service line.

The shared governance structures of integrated systems also facilitate the accountability frameworks necessary for responsible AI deployment.

Conclusion

For surgeon-executives and digital health leaders, this framework offers both caution and opportunity. Without a deep understanding of clinical workflows, robust guardrails, and rigorous use case selection, even promising pilots will remain just that.

The opportunity is equally clear: organizations that master these principles can transform surgical care operations in ways that support clinicians in their efforts to improve patient outcomes. As AI capabilities continue to advance, the gap between organizations that can effectively deploy these tools and those that cannot will only widen. The time to build the foundation for AI deployment in surgical care is now. 

About David Atashroo

David Atashroo is the Chief Medical Officer at Qventus. In this role, he leads the design and direction of the Qventus Perioperative Solution, which uses AI and automation to optimize OR utilization and drive strategic surgical growth. Qventus has generated significant results by maximizing efficiencies for notable health systems, such as eliminating over 36,000 excess patient stays, generating an annualized contribution margin exceeding $110M, and enabling 14,000 new surgeries to be scheduled. He holds a doctorate in medicine from the University of Missouri-Columbia and trained in plastic surgery at the University of Kentucky before completing his postdoctoral fellowship at Stanford University School of Medicine. In addition to his role at Qventus, Dr. Atashroo continues his clinical practice at the University of California, San Francisco.

The following is a guest article by Dr. David Atashroo, Chief Medical Officer of Perioperative at Qventus

Automation and reliable improvement in surgical care operations are emerging patient safety obligations. The potential for AI to transform perioperative care and improve patient outcomes is no longer theoretical. But obligation needs a concrete strategy for execution. We as a field need to better understand why so many AI initiatives fail, and what it takes to succeed.

More than 80% of AI projects fail to reach production across industries, with similar findings in healthcare. For surgery specifically, the gap between promising algorithms and clinical impact remains large. Understanding why requires examination of not just the technology, but the organizational and operational contexts in which it must function.

Where AI Operationalization Fails

Many healthcare AI initiatives begin with impressive demonstrations of model performance. Leadership invests, IT deploys, and clinicians ultimately ignore the tool because it fails to fit existing workflows. Nothing changes, and the tool sits unused while the problems it was intended to address persist. This pattern repeats because organizations are often sold on what AI and machine learning can do rather than what clinical operations actually need.

An algorithm that predicts surgical complications provides no value if its outputs are not integrated into decisional workflows, or if clinicians lack trust in its recommendations. Though the technical achievements we see in model development are impressive and necessary, they are insufficient for operational impact. The difference between a promising prototype and clinical environment-ready software cannot be overstated.

Further, many AI solutions operate outside of primary clinical systems, creating additional work and untenable friction. When AI tools ask even more of healthcare workers already stretched thin, we cannot be surprised when adoption suffers.

Operational Design Principles

Algorithms Alone Do Not Drive Outcomes

Even the most advanced machine learning models are just a component of interventions driving real impact. Health systems must shift thinking from “What can this model do?” to “How exactly will this help our clinicians?” The technical sophistication of an algorithm matters far less than its meaningful integration into the reality of clinical practice.

Involve Clinicians Early in the Development and Implementation Process

Clinician involvement is essential for appropriate problem identification and assessing solution usability. Effective automation requires understanding not just of clinical context but the considerations and judgment that drives decision-making. We must think intentionally about which processes can be fully automated, when we need human oversight, and when we need escalation pathways. This simply cannot be determined without clinician input.

Confidence is Built with Guardrails

Effective guardrails must include mechanisms for ongoing monitoring and transparent communication of limitations. Models degrade over time as patient populations shift, clinical practices evolve, and patient population patterns drift. Continuous performance monitoring, with clear thresholds for intervention and retraining, is essential for building clinician trust.

Guardrails empower clinicians to understand and appropriately use AI tools rather than blindly trusting or dismissing them.

Make the Right Thing the Easy Thing to Do

AI interventions should reduce friction in clinical workflows and make the right decision the easy decision. Behavioral science has long proven that when the right action requires additional steps or cognitive effort, adoption will suffer regardless of potential benefits. Human-in-the-loop designs should present AI recommendations in formats that minimize effort to accept appropriate suggestions, while also allowing clinical judgment to override when needed. Regardless of format, the goal should always be the same: align the path of least resistance with optimal care.

Aim Before You Build

Picking the right use cases is paramount. Surgeon-executives and health systems constantly face limited bandwidth and constrained resources. A spray-and-pray approach will not deliver outcomes. Because of AI’s generative nature, organizations cannot abandon core principles of technology development – rigorous evaluation of data infrastructure, clinical champions, and governance capacity is essential. Prioritizing depth over breadth will generate clinical value and organizational learning to drive subsequent initiatives.

Seek Vendors Providing Deep Solutions

The right path forward requires clarity about what to expect from AI partners. Organizations should demand solutions that integrate deeply with existing workflows rather than asking clinicians to adapt to yet another standalone tool. Vendor solutions must demonstrate understanding of the specific clinical context in which they will operate, with clear plans for implementation support, ongoing monitoring, and drive for further improvement.

The Advantage of Integrated Delivery Systems

These principles reveal a structural insight: integrated delivery systems may be uniquely positioned to realize AI’s benefits in surgical care. Unlike many fragmented care environments, integrated systems can achieve data continuity across the surgical continuum – from preoperative evaluation, through the actual procedure, to postoperative follow-up and long-term outcomes – that enables comprehensive solutions that learn from complete patient journeys.

Clinical engagement and buy-in are also often more achievable within integrated systems. Surgeons, anesthesiologists, nurses, and care coordinators who work within a unified system rather than independent groups can collaborate, provide ongoing feedback, and participate in continuous improvement. This alignment enables the iterative co-design processes that are critical for successful AI deployment.

The value of this approach extends beyond implementation and into care delivery itself. Consider obesity care. Digital weight management platforms, compounding pharmacies, and direct-to-consumer prescribing services have expanded access and helped to engage patients. While these innovations fill important gaps, obesity is rarely a single-modality problem. Some patients benefit from medication management, while others require behavioral health support, nutritional counseling, or bariatric surgery. When care is delivered through isolated point solutions, recommendations are constrained by the tools available within that platform rather than the full spectrum of treatment options.

Integrated delivery systems offer a more holistic view. They can combine AI-enabled insights with a comprehensive set of clinical interventions and evaluate patients longitudinally across the continuum of care. In this model, AI helps guide patients toward the most appropriate intervention rather than simply optimizing a single service line.

The shared governance structures of integrated systems also facilitate the accountability frameworks necessary for responsible AI deployment.

Conclusion

For surgeon-executives and digital health leaders, this framework offers both caution and opportunity. Without a deep understanding of clinical workflows, robust guardrails, and rigorous use case selection, even promising pilots will remain just that.

The opportunity is equally clear: organizations that master these principles can transform surgical care operations in ways that support clinicians in their efforts to improve patient outcomes. As AI capabilities continue to advance, the gap between organizations that can effectively deploy these tools and those that cannot will only widen. The time to build the foundation for AI deployment in surgical care is now. 

About David Atashroo

David Atashroo is the Chief Medical Officer at Qventus. In this role, he leads the design and direction of the Qventus Perioperative Solution, which uses AI and automation to optimize OR utilization and drive strategic surgical growth. Qventus has generated significant results by maximizing efficiencies for notable health systems, such as eliminating over 36,000 excess patient stays, generating an annualized contribution margin exceeding $110M, and enabling 14,000 new surgeries to be scheduled. He holds a doctorate in medicine from the University of Missouri-Columbia and trained in plastic surgery at the University of Kentucky before completing his postdoctoral fellowship at Stanford University School of Medicine. In addition to his role at Qventus, Dr. Atashroo continues his clinical practice at the University of California, San Francisco.