The Quiet Revolution: Operational AI That No One Notices (But Everyone Feels)

A revenue cycle team discovers that denial rates for a specific payer group have decreased by 12 percent for the last quarter. Measures for coding accuracy have improved. Turnaround times for prior authorization have shortened. No team member in the department attributes this success to the AI tool implemented in the claims adjudication workflow six months prior. They think it is because the team has improved. The tool is now invisible. This is the definition of success.

The AI initiatives that create the most sustained impact in healthcare operations are the ones that are no longer recognized as AI. They become a part of the workflow. The line of distinction between a deployed tool and a tool that is integrated into the workflow is used to describe the difference between a pilot and a full-scale capability. Most health systems have much more of the former and much less of the latter.

AI in healthcare operations encompasses claims processing, scheduling optimization, supply chain management, bed capacity planning, referral routing, and workforce allocation. Unlike in traditional healthcare, these are not clinical applications. They do not impact diagnostic or therapeutic decisions directly. They design the framework within which those decisions are taken.

The Annual Survey from Deloitte on Health System Executives has identified operational efficiency as the top priority area for investment in AI yearly in comparison to patient engagement and clinical decision support. This makes sense. Operational AI focuses on areas in which the processes are already defined alongside measurable inputs and outputs, and known performance metrics. The business case is more straightforward, the risk is less, and the timeline to achieving a measurable return is shorter.

Research published in Health Affairs has analyzed the effect of operational AI on the financial performance of health systems and has found positive results for health systems with advanced operational AI in revenue cycle performance, labor costs, and capacity utilization. The improvements are not substantial in any given area, but are spread broadly which results in a compounding effect.

Clinical AI, due to the tangible implications, draws the most attention. Whether it is a diagnostic model that detects cancer, an anticipating algorithm that detects the risk of sepsis or an ambient tool for automatic documentation, these applications are all linked to the final health outcome of a patient which leads to stories that drive funding and public interest.

Operational AI lacks these stories. An algorithm that decreases operating room downtime by 15% remains unknown. A claims processing model that results in fewer denials fails to attract attention and a capacity planning tool that leads to faster bed turnover results goes unnoticed.

As the impact scope of operational AI continues to expand, the funding, oversight, and recognition of operational AI will continue to remain limited to the underdefined and underperforming aspects of AI. The lack of funding, oversight, and recognition creates a gap in review. The gap in systems that process large volumes of claims, distribute multi-million dollar labor costs, and schedule hundreds of clinicians, stands in contrast to the oversight systems in place for a diagnostic model that operates at a smaller volume.

Operational AI tools make decisions that affect patients indirectly. A clinician scheduling tool that underestimates future demand for a service line leads to access issues. A claims processing tool that denies certain procedure codes leads to loss of revenue and increases frustration for patients. A capacity planning tool that deprioritizes certain units creates inequitable distribution of resources.

The effects of operational AI systems that are implemented at an individual transaction level are often invisible. They are only visible at a higher aggregate level, and only if the organization is actively tracking the outcomes. The operational AI oversight structure needs to include the performance monitoring, bias review, and responsibility provisions that are used for clinical AI, but tailored to the operational AI context.

The AHA has operational AI oversight recommendations that encourage cross-functional review, and operational AI tools tend to impact many involved parties. From an oversight perspective, review limited to a single organizational unit, while the cross-functional scope of operational AI extends well past that unit, underestimates the systemic impact that can pose the greatest risks and opportunities.

Operational AI tools that generate sustainable value possess a unique design quality: they are embedded in the workflow, rather than layered on top of it. Any tool that is designed to function in tandem with the primary workflow system, where the tool recommends an action to the user, the user leaves the workflow system to access a different interface, and then returns to the workflow system to perform the action, introduces friction and causes a decline in adoption over time.

This standard of integration applies to the user interface, and also encompasses the data integration, which must be a two-way street. The tool must actively pull data from operational systems and push data back to the operational systems in a way that triggers an action without user intervention. For example, if a predictive staffing tool generates a report that a manager has to take to finalize the schedule, the manager has relieved their burden of analysis, but the operational burden remains. Consequently, the value of that tool is very limited.

If health systems are looking to implement operational AI tools, they need to evaluate the integration of the tool in the operational workflow that it is intended to improve, in addition to assessing the analytical capabilities of the tool. Without a doubt, the most accurate model that exists does not deliver value if the final step between recommendation and action requires human effort that the operational environment cannot consistently provide.

Deloitte has focused on health system executive priority surveys annually concerning AI investment. Health Affairs has focused on the effects of operational AI maturity in health systems. The American Hospital Association has offered oversight recommendations pertaining to operational AI. This edition connects to operational themes examined in Editions A, C, and J of this newsletter.

Christopher Hutchins
Founder & CEO, Hutchins Data Strategy Consultants