AI is everywhere right now—from your inbox to your imaging studies.
But in healthcare, the real promise of artificial intelligence isn’t the ChatGPT-style diagnoses.
It’s smarter utilization.
AI has the potential to cut hundreds of billions in waste by automating the worst parts of our system: waitlists, scheduling, documentation, claims, billing, and more. It’s not a silver bullet—but it’s the first real tool we’ve had in a while that can target the root inefficiencies clogging up care delivery.
So how exactly is AI being used to improve utilization and quality in healthcare—and is it actually working?
In this article, I’ll walk through the biggest pain points AI is tackling today, what the research and ROI show so far, and what it’ll take for these tools to move from early adopters to the mainstream.
The Problem with Administrative Costs in U.S. Healthcare
National healthcare expenditures are approaching $5 trillion—up from $1 trillion when I was born and just $50 billion when my parents were born. For two decades, everyone has talked about "bending the cost curve," which has proven far easier said than done.
This $5 trillion spans across healthcare domains from payers to providers to pharmaceuticals. Yet one area that often escapes precise measurement is administrative costs.
Studies estimate that 15-30% of all healthcare dollars go toward administrative work. Put simply, when you pay $100 for medical services, up to $30 funds administrative tasks like revenue cycle management, prior authorization, and back office operations. More troubling still, researchers estimate half of these administrative costs are pure waste—that's 7.5-15% of national healthcare expenditures wasted!
On a per capita basis, the U.S. spends about $1,000 per person on healthcare administration. For comparison, Germany—the second-highest spender—allocates just $300 per capita for administration. The U.S. spends over three times more per person on paperwork than any other advanced nation!
The Role of AI in Streamlining Administrative Tasks
Artificial intelligence offers a promising solution to reduce administrative costs by streamlining existing systems or creating entirely new ones.
AI comes in two main forms that you're likely familiar with (or you can watch my class on generative AI here):
Machine Learning (ML): makes predictions from structured data.
Natural Language Processing (NLP): interprets unstructured language in notes and conversations.
When combined, these technologies create powerful tools that automate workflows and significantly reduce administrative burden.
Healthcare 101 Course: AI in Healthcare
Understanding AI's Role in Utilization: The "Self-Driving Car" Analogy
To understand how AI is transforming healthcare utilization, consider the levels of automation we see in cars.
Just like self-driving cars have different levels of autonomy, AI in healthcare operates on a spectrum (read more from Sahni et al here):
Level 1: Basic Assistance - Think clinical decision support tools that flag potential drug interactions or remind providers about overdue screenings. The human still does almost everything, but the AI provides helpful alerts.
Level 2: Partial Automation - Here, AI handles specific tasks while humans supervise. Radiology AI that pre-reads images and highlights suspicious areas is a great example. The radiologist still makes the final call, but the AI narrows their focus.
Level 3: Conditional Automation - Now we're getting to the exciting part. AI systems that can manage entire workflows like referral management—determining which specialty a patient needs based on their symptoms and history, then routing appropriately. Humans step in only for exceptions or complex cases.
Level 4: High Automation - This is where AI handles complete processes with minimal human intervention. Think claims automation that processes straightforward claims without human review, or AI scribes that document entire patient encounters, letting doctors focus entirely on the patient.
The exciting news? We're rapidly approaching levels 3-4 in many healthcare systems. Major health systems are already deploying AI that can autonomously manage bed assignments, predict sepsis hours before clinical signs appear, and automatically code complex medical records.
AI Use Cases Across the Healthcare Ecosystem
AI is transforming healthcare operations across the board. Let's look at how different sectors are leveraging these technologies to improve care and cut waste (read more from Sahni et al here):
Hospitals
Hospitals are using AI to predict bed capacity needs, optimize discharge timing, and identify patients at risk for readmission—helping to improve patient flow and resource allocation.
Clinical workflows are being streamlined through automated referral systems and enhanced analytics that surface actionable insights for care teams.
Perhaps most impressive are the sepsis detection systems. In randomized controlled trials, these tools have shown significant reductions in mortality rates and length of stay by flagging at-risk patients hours before clinical signs appear.
Physician Groups
Scheduling optimization is key for outpatient settings. Systems like Third Way Health use AI to analyze no-show patterns and dynamically adjust schedules, reducing wasted appointment slots. I’ve also covered this here.
Ambient documentation tools (like Nuance DAX and Suki) are transforming the patient encounter by automatically generating notes from doctor-patient conversations.
The impact here is twofold: reduced physician burnout from documentation burden and improved throughput that allows providers to see more patients without sacrificing quality time.
Payers
Prior authorization—the bane of both providers and patients—is being automated through AI systems that can predict which services will be approved.
Fraud, waste, and abuse detection has become more sophisticated with tools like the CMS WISER model that’s encouraging AI tools to be built to detect suspicious billing patterns more accurately than rule-based systems.
Risk stratification algorithms help payers and providers identify high-risk patients for targeted care management interventions, potentially preventing costly hospitalizations.
Who's Building the Tools?
The AI healthcare ecosystem is rapidly expanding with solutions coming from three main sectors:
Incumbents
Epic and Oracle are integrating AI directly into their EHR platforms, providing clinical decision support tools that work within existing physician workflows.
Microsoft/Nuance's DAX (Dragon Ambient eXperience) has emerged as a leading ambient documentation solution, automatically transcribing and structuring clinical conversations to reduce documentation burden.
Big Tech
Google Health has developed various healthcare AI tools, including models for medical imaging analysis and predictive analytics.
Amazon AWS provides the cloud infrastructure powering many healthcare AI applications, while offering their own healthcare-specific services.
IBM Watson, despite early challenges, established foundational approaches for clinical AI that influenced the current generation of solutions.
Specialized Startups
Qventus provides machine learning solutions for real-time hospital operations, including predicting likely discharges and coordinating tasks to streamline the process. Hospitals implementing these tools report significant reductions in length of stay and improved bed availability.
LeanTaaS uses AI algorithms to optimize scheduling for operating rooms and infusion clinics, minimizing idle time and increasing patient throughput. Many academic medical centers utilize these tools to run more cases with the same resources.
Bayesian Health, founded by Johns Hopkins faculty, has partnered with leading institutions like Duke Health and Cleveland Clinic to deploy AI systems that detect patient deterioration and sepsis early, contributing to improved survival rates.
Other notable players include Viz.ai, Aidoc, PathAI, Cleerly, and Third Way Health, each specializing in different aspects of healthcare optimization.
The most successful implementations share a common characteristic: they integrate seamlessly into existing workflows rather than requiring clinicians to learn entirely new systems or platforms.
ROI and Cost Impact
AI-driven healthcare solutions promise significant financial returns across the healthcare ecosystem. According to research from Harvard University and McKinsey & Company, wider AI adoption could save the U.S. healthcare system 5% to 10%—translating to $200 billion to $360 billion annually.
These savings materialize differently across key stakeholders:
Hospitals: $60B–$120B potential annual savings (4–10% of total hospital costs)
Primarily from clinical operations improvements like OR optimization
Quality and safety enhancements including early deterioration detection
Physician Groups: $20B–$60B annual savings (3–8% of total costs)
Schedule optimization and reducing no-shows
Enhanced efficiency in clinical documentation and workflows
Payers: $80B–$110B annual savings
Streamlined claims management and prior authorization
More effective care management and reduced readmissions
Overall AI net savings opportunity, Sahni et al (2023)
While implementing AI solutions requires upfront investment, the key insight is clear: AI isn't cheap to implement—but the waste it replaces is far more expensive.
Barriers to Adoption: The TFD Framework
Despite the compelling ROI and growing evidence supporting AI in healthcare, adoption remains inconsistent across the industry. To understand why, we can examine barriers through my TFD Framework:
Trust
Healthcare organizations are currently navigating what Geoffrey Moore describes as "crossing the chasm" - the challenging gap between early adopters and the early majority.
Source: Lars de Bruin, Business to You
This critical transition requires:
Proof points from respected peer institutions that demonstrate not just technical success but meaningful clinical and financial outcomes
Evidence that meets both technical and clinical standards — providers need to trust both the ROI calculations and the clinical soundness of AI solutions
Transparency in how algorithms make decisions, especially for clinical applications where patient safety is at stake
Without trust, healthcare organizations remain stuck in pilot purgatory, unable to scale promising technologies across their systems.
Frictionless Experience
The most successful AI implementations share a crucial characteristic: they integrate seamlessly into existing workflows rather than requiring clinicians to learn entirely new systems. This means:
Solutions must embed directly into the tools clinicians already use daily
AI cannot add cognitive load to already overburdened clinicians
Implementation must account for existing operational processes and adapt to them, not vice versa
Even the most sophisticated AI will fail if it creates additional work or disrupts clinical flow. The most successful solutions become nearly invisible, operating in the background while enhancing clinical decision-making.
Distribution Moat
Access to providers represents a significant barrier for many AI solutions. This gives a substantial advantage to:
Solutions that embed directly into EHRs and other platforms that providers already use daily
Platforms with existing clinical workflows that can be enhanced rather than replaced
Beyond these three core barriers, healthcare organizations also face additional challenges, including legacy technology constraints, siloed data, nascent operating models, misaligned incentives, industry fragmentation, and talent acquisition difficulties. These systemic issues can further complicate adoption even when TFD barriers are addressed.
The digital divide between well-resourced systems and safety-net providers presents a particular concern, as AI adoption risks exacerbating existing healthcare inequities if only wealthy systems can afford implementation.
The Big Picture
AI is transforming healthcare utilization management from basic assistance to autonomous workflows. Using the automotive automation framework, we're rapidly progressing toward levels 3-4 across hospitals, physician groups, and payers. The economic case is clear: potential industry-wide savings of $200-360B annually. However, successful implementation requires addressing the TFD framework—building trust through clinical evidence, creating frictionless experiences within existing workflows, and solving distribution challenges. For healthcare leaders, AI represents a strategic capability that will differentiate high-performing organizations in the coming years.