By Dr. Scott Schell, Chief Medical Officer, Cognizant
LinkedIn: Scott Schell
LinkedIn: Cognizant
Healthcare organizations did not adopt AI for novelty. They adopted it because margins are tightening, labor is constrained, and the operational burden of modern healthcare is becoming increasingly difficult to sustain.
In several areas, AI has already delivered meaningful value. Ambient documentation platforms are reducing clerical effort during clinical encounters. Revenue cycle tools are preparing well-formed prior authorization requests, accelerating claims workflows and identifying denial risks earlier in the process. Inbox triage systems sort and prioritize large volumes of patient communication that otherwise accumulate faster than physicians can reasonably respond to them.
Those gains are material.
However, many organizations now face a second-order effect that received far less attention during AI’s early healthcare adoption phase: AI may accelerate tasks without necessarily simplifying the surrounding workflow.
Those are not the same thing.
In practice, many clinicians and operational teams are finding that work is not disappearing as much as changing form. The manual effort of producing content or processing transactions is increasingly being exchanged for supervisory work: reviewing, validating, correcting, escalating, and managing exceptions generated by AI systems operating at scale.
This hidden workload matters because healthcare is not a low-consequence environment. A hallucinated restaurant or clothing recommendation is forgettable. An incorrectly summarized clinical history, a missed escalation left buried inside inbox triage, or improperly routed prior authorization could devolve into something else entirely.
As a result, healthcare organizations are discovering that AI is often migrating operational burden rather than eliminating it.
Ambient documentation provides a useful example. The technology substantially reduces keyboard time during patient encounters. In exchange, many physicians now spend part of each visit reviewing AI-generated summaries for omissions, incorrect attribution, diagnosis errors, timeline compression, or subtle contextual errors that may not become obvious until later in the episode of care.
The work simply changed. It did not fully disappear.
A similar dynamic is emerging in revenue cycle operations. AI accelerates intake, coding support, denial prediction, and document handling. However, when systems generate outputs at high volume, organizations still need personnel and workflows capable of identifying edge cases, resolving ambiguity, and intervening when automation confidence degrades.
Healthcare workflows contain enormous variation. Exceptions are not rare events; in many specialties and operational domains, they are the workflow. This explains why some early AI narratives became overly simplistic: organizations initially confused task acceleration with workflow simplification. While related, they are not interchangeable.
A clinician may exchange faster documentation with greater cognitive responsibility for verification. A call center may process requests more rapidly while creating downstream rework for utilization management teams. AI-assisted inbox handlers may decrease time while creating concern about what the system may have missed.
These tensions do not represent failure of AI technology itself. Rather, they reflect the reality that healthcare delivery is fundamentally a coordination problem. AI inserted into fragmented workflows without operational redesign has the potential to relocate friction from one part of the system to another.
This “relocation” is increasingly important as the economics of healthcare AI continue to evolve. Early discussions often focused on physician burnout and “pajama time.” While those are significant issues, healthcare executives are now focusing on harder operational outcomes:
- Does the technology materially improve throughput?
- Does it reduce avoidable labor expense?
- Does it improve revenue capture or reduce leakage?
- Does it allow clinicians to spend more time on revenue-generating activities?
- And ultimately, does the operational benefit exceed the cost of acquiring, integrating, governing, monitoring, and supervising the AI itself?
Those are more mature questions. They are also the right ones.
Organizations seeing the strongest returns from AI are not treating it as isolated productivity tools. They are redesigning workflows around them, including escalation pathways, exception handling, interoperability, and clear accountability for how AI outputs move through clinical and operational processes.
As a result, successful AI deployment increasingly resembles operational engineering more than software installation.
That distinction becomes increasingly important as healthcare organizations move from contained pilots into enterprise-scale deployment. While small pilots can tolerate manual oversight and informal correction loops, enterprise systems cannot. Once embedded into scheduling, inbox management, revenue cycle operations, utilization review, or clinical documentation, even minor AI inefficiencies can propagate rapidly across organizations.
Healthcare is now moving beyond the question of whether AI can generate outputs. The more strategic question is whether organizations integrate those outputs into workflows in ways that genuinely reduce friction, maintain trust, and create measurable financial returns.
That work is harder than deploying a model, and it is likely where the next phase of healthcare transformation will occur.