The Impact of Artificial Intelligence on the Health Economy, Workforce Productivity, and Administrative Efficiency: A Systematic Review
1Institute of Health Policy and Management, College of Public Health, National Taiwan University
2Schulich School of Medicine and Dentistry Department of Epidemiology and Biostatistics, Western University
3Institute of Environmental and Occupational Health Sciences, College of Public Health, National Taiwan University
4Department of Environmental and Occupational Medicine, National Taiwan University Hospital
5National Institute of Environmental Health Sciences, National Health Research Institutes
6Department of Emergency Medicine, National Taiwan University Hospital
7NTU Research Centre for AI Technology and Healthcare
8Nossal Institute for Global Health, The University of Melbourne
9Global Health and Population, Harvard T.H. Chan School of Public Health
*Correspondent to: johntayulee@ntu.edu.twABSTRACT
Background
Healthcare systems globally are under increasing financial and operational strain due to aging populations, rising expenditures, and workforce shortages. Amid these challenges, artificial intelligence (AI) has emerged as a promising tool to enhance system-level performance, particularly in cost reduction, productivity gains, and administrative efficiency.
Objective
This review aims to synthesize existing evidence on the macro and system-level impact of AI implementation across three key domains: the health economy, workforce productivity, and administrative efficiency.
Methods
A systematic review methodology was employed, allowing for the integration of diverse data sources and study types. Literature was systematically identified through PubMed and Google, covering publications from 2020 to July 2025. Studies were included if they evaluated AI’s impact on national or regional health expenditure, labor restructuring, or process efficiency. Thematic synthesis was guided by a conceptual framework modeling AI as a system-wide catalyst.
Results
Twenty-four studies were included. AI implementation demonstrated potential cost savings of 5–10% in national health expenditures, driven by automation in hospital operations and administrative processes. AI-supported interventions reduced diagnostic time by up to 90% and treatment costs by over 30% in specific applications, such as cancer diagnosis and radiotherapy. Administrative tools, including AI-assisted documentation and claims processing, achieved efficiency gains of up to 40%. However, reliance on simulated models, short-term studies, and single-center data limits generalizability.
Conclusions
AI presents significant potential to enhance health system efficiency and reduce costs. Real-world implementation studies, standardized outcome metrics, and robust governance frameworks are essential to validate these gains and ensure equitable, sustainable adoption.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Funding Statement
This study is funded by Yushan Fellow Program from the Ministry of Education (MOE), Taiwan (MOE-112-YSFMN-0003-002-P1)
INTRODUCTION
Health systems around the world are facing growing pressure. Spending on healthcare is rising faster than many economies can keep up with due to rapid aging population1. The percentage of GDP spent on health has shown a significant increase over the past decades. According to projections, healthcare spending in the United States is expected to grow faster that the GDP, reaching 19.7% of GDP by 2032, up from 17.3% in 20222. This trend reflects a broader global pattern where health spending as a share of GDP has been increasing, driven by factors such as economic development, demographic changes, and policy shifts. For instance, global health spending was 8.6% of the global economy in 2016 and is projected to reach 9.4% by 2050 3.
The escalating cost occurs while many health systems struggling with insufficient healthcare funding from both the public and private sector. For example, the UK’s National Health Service (NHS) faces a projects £ 4.8 bill shortfall in its 2024/2025 revenue budget, raising concerns about potential service cuts without additional funding 4. In the United States, proposed federal cuts of $ 880 to Medicaid over the next decade threaten to reduce access to essential services cater for vulnerable populations 5.
Additionally, many health system has shortages in the healthcare workforce, making it challenging to meet the growing demand6. This issue is compounded by the maldistribution of healthcare professional across geographical area, leading to inequalities in service provision in underserved and remote areas. Adding to this burden is the increasing amount of administrative work that clinicians and managers face growing administrative burden, diverting valuable time from direct patient care. According to a time and motion study, physicians in ambulatory practice spend 49.2% of their time on electronic health record (EHR) and desk work and 27.0% on direct clinical face time with patients during office hours7. Another study in the Journal of General Internal Medicine found that physicians spend 20.7% of their time on EHR input alone and 7.7% on administrative activities, with 66.5% of their time on direct patient care (including multitasking with EHR) 8.
Against this backdrop, Artificial Intelligence (AI) and Machine Learning (ML) have captured attention as tools that could potentially reshape how health systems function. AI is not new—it began with early attempts to simulate human thinking using symbolic systems and rule-based expert programs in the 1980s. These tools could be useful in narrow areas but were limited by their inability to handle complexity and real-world uncertainty. What has changed in recent years is the rise of new forms of AI, particularly deep learning and large-scale foundation models. These systems learn from enormous amounts of data and can now perform tasks that once required human intelligence—such as reading radiology scans, understanding natural language, writing computer code, or even engaging in meaningful conversation. Tools like GPT and Gemini represent a leap forward not only in accuracy but also in general intelligence. In healthcare, they are already being tested for clinical decision support, predictive risk modelling, and automating administrative tasks. At present, AI tools have been increasingly used and accepted in clinical practice, particularly in imaging and diagnosis. LLMs have also been used for documentation, administrative, or coding work in clinical settings 9,10.
But alongside this rapid progress, concerns remain. These systems rely heavily on the data used to train them—data that may be incomplete, biased, or not representative. AI decisions can also be difficult to interpret or explain, which raises issues of trust, accountability, and ethics 11. While much attention has focused on AI’s technical capabilities, its broader impact on health systems as a whole requires deeper investigation. From a policy prospective, it is essential to examine the macro-level impact of AI on health economy. This study reviews the current evidence on how AI is influencing three key areas: (1) overall health spending and cost control, (2) administrative efficiency, and (3) workforce productivity.
RESULTS
Summary of Included Studies
This review includes 24 studies examining the macro-level impact of artificial intelligence (AI) across three key domains: health economy, workforce productivity, and administrative efficiency. Of these, 9 studies were conducted in the United States, and the remainder spanned global or multi-country settings, including Taiwan, Japan, Bengal articles12-19.
Medical imaging emerged as the most common application area, with nine studies focusing on AI’s role in interpreting radiologic, endoscopic, or pathological images 15,16,20,21. These studies investigated tasks such as detection15,16,22, diagnosis16,22-24, reduction of interpretation time 15,16, and workflow optimization 15,16,23.
Nine studies12,14,17-20,22,24,25explicitly addressed the health economic impact of AI, including cost savings12-14,19,20,24, budget impact19,20,24, and reduction in treatment costs14. Administrative efficiency was the focus of three studies 13,26, each addressing the potential of AI to automate and streamline non-clinical tasks such as clinical documentation 13,18, and claims processing 13,19.
Impact of Health Economy or Systems-Level Spending
The study, conducted by a team of researchers from Harvard, offers one of the most comprehensive assessments of the potential financial impact of AI in healthcare sector19. It estimates that over the next five years, AI could generate annual net savings of $200 billion to $360 billion in the U.S. healthcare sector without compromising quality or access, based on 2019 dollars. This represents a 5% to 10% reduction in total U.S. healthcare spending. The estimate is based on data from the National Health Expenditure, specifically total hospital revenues and total hospital costs19.
Complementing these quantitative projections, the EIT Health & McKinsey report highlights AI’s systemic role in sustaining healthcare delivery under financial and workforce pressures, particularly in Europe, by enhancing productivity and enabling more proactive models of care18. At a broader economic frontier, McKinsey’s assessment of generative AI estimates an annual global value creation of USD 2.6–4.4 trillion, with USD 60–110 billion accruing specifically to the pharmaceutical and medical-product industries, pointing to further downstream effects on healthcare affordability and innovation17.
Costs Reduction or Saving for Treatment
Few studies identifying cost savings associated with the use of AI. s. One example is provided by Mori et al., who analyzed AI-aided colonoscopy for polyp diagnosis. They compared traditional methods with an AI-aided approach for detecting diminutive colorectal polyps. Their findings indicated reductions in average colonoscopy costs of approximately 18.9% in Japan, 6.9% in England, 7.6% in Norway, and 10.9% in the United States. Thus, the use of AI results in substantial cost savings for colonoscopy20. A study on metastatic colorectal cancer demonstrates that deploying high-sensitivity AI for MMR/MSI status determination, followed by confirmatory testing for negative cases, resulted in projected savings of $400 million— approximately 12.9% of total diagnostic and treatment costs—compared to next-generation sequencing alone 14. Another cost analysis focusing on machine learning-guided radiotherapy interventions showed a statistically significant reduction in acute care visits and a halving of treatment costs per course ($1,494 vs. $3,110), suggesting strong cost-efficiency when targeting high-risk cases using AI triage 12,14.
Additionally, a retrospective study showed substantial cost reductions through improved medication adherence and patient outcomes using AI. Findings indicated that after the introduction of a pharmacist-led, AI-supported program, medication adherence improved across all disease measures. Importantly, adherent patients demonstrated notable reductions in overall healthcare expenditures compared to nonadherent patients, with cost savings of 31% for hypertension, 25% for hyperlipidemia, and 32% for diabetes. However, external validity may be limited by claim data constraints, potentially impacting the reflection of actual clinical outcomes due to missing or inaccurate data22.
Efficiency Gains in Medical Imaging
In radiology, the implementation of an AI tool for pneumothorax detection was shown to significantly improve report turnaround time, reflecting measurable gains in workflow efficiency16. Similarly, a prospective randomized study found that integrating an automated AI platform into chest CT interpretation reduced reading time by 22.1%, demonstrating AI’s potential to enhance radiologist productivity15. Similarly, in breast cancer screening, an international evaluation published in Nature demonstrated that an AI system outperformed radiologists in both the US and UK, reducing false positives by up to 5.7% and false negatives by 9.4%, while cutting second-reader workload by 88%37.
These findings align with Transforming Healthcare with AI, which identifies diagnostic speed and accuracy as key areas where AI is already delivering impact. The report emphasizes that imaging-heavy specialties such as radiology, pathology, and ophthalmology are being actively reshaped by AI, describing imaging applications as “low-hanging fruit” due to their readiness to provide immediate, tangible benefits in clinical settings18.
Efficiency Gains in Medical Documentation
One study identified clinical documentation as a significant component of administrative costs and discussed how tools such as NLP can extract relevant data from clinical notes. This can help ease documentation demands from payers and accelerate claims approvals, emphasizing the link between documentation processes and administrative efficiency13.
Similarly, Analysis by McKinsey group notes that AI can take on administrative tasks, including the use of NLP to process clinical notes—highlighting its role in automating clinical documentation and improving workflow efficiency in this area18.
Recent empirical work further illustrates these benefits in practice. Rozario and colleagues demonstrated that applying a machine-learning optimization model to operating room scheduling significantly reduced overtime by 21%, translating to an estimated cost saving of USD 469,000 over three years26. In allied health private practice, Evans et al. found that using an AI scribe decreased time spent on clinical notes and letters, reduced after-hours documentation, and improved both productivity and therapeutic alliance between clinicians and patients33. Together, these findings highlight AI’s dual potential: streamlining administrative processes while simultaneously enhancing the efficiency and quality of clinical workflows.
Efficiency Gains in Claim Management
In terms of administrative efficiency, AI tools have shown some of the clearest and most immediate benefits13,18,19. Studies suggest that automating claims processing, prior authorization, quality assurance, and credentialing could eliminate as much as $168 billion in annual U.S. administrative spending13. AI applications in this area include centralized digital claims clearinghouses, NLP-based documentation extraction, and fully electronic prior authorization systems13. For instance, centralizing claims processing was estimated to save $300 million annually, while electronic authorization pipelines could yield an additional $417 million in savings. The McKinsey and NBER analyses both highlight administrative automation as a primary driver of system-level savings, estimating that 40–50% of AI-derived cost reductions for hospitals and physician groups stem from back-office and billing functions.
Potential in Claims Management and Member Services: AI could lead to significant savings by addressing repetitive administrative processes such as processing prior authorisations or adjudicating claims. AI-enabled identification models can streamline operations and inform efforts to reduce fraud, waste, and abuse19.
Both McKinsey’s Reports (2020, 2023) highlight AI’s transformative potential but differ in scope: the former focuses on AI’s role in improving healthcare delivery and reducing system waste, while the latter explores cross-industry automation of knowledge work. Both reports also address labor implications—The Economic Potential of Generative AI (2020) points to potential displacement of high-skill tasks, whereas Transforming Healthcare with AI emphasizes relieving administrative burdens in healthcare and the importance of digital upskilling. While Generative AI focuses on theoretical value creation, Transforming Healthcare with AI ties economic gains to specific healthcare efficiencies and underscores the need for governance, funding, and workforce readiness to realize AI’s full potential18,38.
In summary, Figure 2 illustrates a mechanism-based framework that was consistently described across several studies included in this review. Beginning with inputs in the form of AI technologies—including applications in natural language processing, image analysis, and machine learning—the framework highlights the key mechanisms through which these technologies enact change. These include process automation, decision support, and resource optimization. As these mechanisms are embedded into healthcare operations, they contribute to measurable improvements in productivity, efficiency, and service quality. These system-level outcomes, in turn, drive economic impact through cost savings, improved health outcomes, and better use of resources. This pathway emphasizes that the economic benefits of AI are not simply a function of the technology itself, but rather emerge from its integration into and influence on health system processes.
Analytical Approaches and Statistical Methods
The studies included in this review employed a diverse range of analytical approaches to estimate the impact of AI on healthcare costs, productivity, and efficiency. At the broader economic level, macroeconomic modeling and bottom-up use-case analysis were used to forecast potential system-level savings. For instance, Sahni et al.19 estimate that wider adoption of AI could lead to savings of 5% to 10% in U.S. healthcare spending—approximately $200 to $360 billion annually in 2019 dollars. These estimates are based on specific AI-enabled use cases that utilize current technologies and are attainable within the next five years without compromising quality or access. Their approach involved mapping specific healthcare tasks (e.g., administrative work, diagnostics, operations support) to AI capabilities and estimating time and cost reductions from AI adoption across sectors like payer services and hospital operations. This analysis offers a macroeconomic view of how AI could reshape healthcare spending patterns if deployed at scale.
In contrast, scenario-based cost modeling has also been applied to specific clinical workflows. Kacew et al.14 evaluated eight diagnostic strategies for determining mismatch repair/microsatellite instability (MMR/MSI) status in metastatic colorectal cancer. Their analysis found that a strategy combining high-sensitivity AI with confirmatory high-specificity PCR or IHC testing resulted in the greatest cost savings—approximately $400 million (12.9%)—compared to next-generation sequencing alone. Additionally, a high-specificity AI-only approach provided the shortest time to treatment initiation (<1 day) and maintained a 97% rate of patients receiving guideline-supported therapy.
Experimental and quasi-experimental designs were also employed to directly quantify changes in healthcare productivity resulting from AI use. A notable example is the cluster-randomized trial by Abramoff et al.21, which examined the use of an autonomous AI system for diabetic eye exams in primary care. The study used queueing theory to model clinic workflows and measured productivity by the number of completed patient encounters per hour. Clinics using the AI system achieved a 40% increase in throughput compared to controls, highlighting the operational gains achievable through autonomous decision-making tools.
DISCUSSION
This review synthesizes current evidence on the macro-level effects of artificial intelligence (AI) across three critical domains of health system performance: the health economy, workforce productivity, and administrative efficiency. Across these domains, AI—particularly in the form of machine learning algorithms, generative models, and natural language processing systems— demonstrates substantial promise in generating cost savings, enhancing operational throughput, and improving workflow efficiency.
In the economic domain, analyses from sources such as McKinsey and the National Bureau of Economic Research estimate that scaled AI implementation could reduce national health expenditures by 5 to 10 percent. These projected savings are primarily attributed to automation of clinical and administrative processes and the deployment of predictive analytics. However, these forecasts are based on optimistic assumptions, and their applicability across healthcare systems with varying regulatory and infrastructural contexts remains uncertain.
AI has also shown promise in mitigating workforce inefficiencies. Applications such as ambient AI and large language models are increasingly used to automate clinical documentation and accelerate diagnostic workflows. Systematic reviews have reported significant reductions in diagnostic time—occasionally exceeding 90 percent. However, such gains are offset by ongoing concerns about the accuracy, reliability, and contextual coherence of AI-generated content, particularly in documentation where clinical nuance and narrative detail remain essential.
Administrative functions—including claims processing, prior authorization, and provider credentialing—emerge as particularly well-suited to AI-enabled automation. These high-cost, low-clinical-risk tasks present opportunities for efficiency, standardization, and transparency, with relatively lower ethical and safety concerns. As such, they may serve as practical entry points for broader AI integration at the system level.
Limitations of Existing Studies
While the research on AI in healthcare is growing fast, there are still some major gaps we need to talk about. First, a lot of the studies are short-term, so we don’t really know how AI tools perform over the long run or how they change actual patient outcomes. Also, many of them rely on very specific patient groups or are based in just one or two hospitals, which means the findings might not apply to everyone. Another big issue is that most studies don’t break down their results by race, income level, or social background, so we’re missing a clear picture of whether AI is helping or hurting when it comes to health equity. On top of that, some studies struggle with confounding factors—like whether certain types of patients are more likely to get AI-assisted care—which can muddy the results. And finally, while we know that introducing AI can shake up how people work, very few papers look into the real-world barriers to getting these tools adopted, such as whether clinicians trust the tech or how it fits into their daily routines. So even though the potential is huge, there’s still a lot we don’t fully understand.
Despite the potential of artificial intelligence to enhance efficiency and reduce costs in healthcare, its unintended consequences warrant careful consideration. Overreliance on AI may erode clinicians’ diagnostic judgment, leading to skill degradation, increased risk of misdiagnosis, and the resulting costs of medical errors and retraining39,40. If training data used for AI models is biased, the technology may deliver inequitable outcomes, disproportionately impacting minority groups and further exacerbating health disparities, with increased long-term costs due to preventable disease progression41,42. While AI can improve diagnostic sensitivity, it may also result in incidental findings and overtreatment, contributing to the waste of limited healthcare resources. Furthermore, the vast amount of health data required by AI systems poses significant privacy and cybersecurity risks; data breaches may lead to legal liabilities and high infrastructure costs, especially burdensome for smaller institutions43,44. Although AI can alleviate workforce burdens, it may disrupt labor markets by displacing certain roles and creating reskilling needs. Highly specialized algorithms may also fragment care delivery, reducing coordination and integration. To mitigate these risks, the implementation of AI must be grounded in careful governance and equitable model design to ensure that it strengthens— rather than undermines—high-quality, patient-centered care45,46.
Research and Policy Implications
Artificial intelligence is speeding image reading, documentation, and administrative workflows, but supporting evidence remains scattered across small pilots, heterogeneous designs, and simulations, hindering comparison and generalization47 48. Future studies should adopt shared efficiency metrics—time saved per case, full cost-utility ratios that include upkeep, and indicators such as user trust and alert overrides—while embedding prospective mixed-methods evaluations in routine care to link data quality, culture, and interface design with short-term performance and long-term sustainability49. Economic work must count real deployment costs—retraining, integration, workforce preparation—so decision-makers see net value, not idealized savings47,50.
It’s worth to note that efficiency gains alone will not necessarily ease shortages: while in nursing, AI may enhance nurses’ work-life balance51, some physicians are worried that they may have to see more patients in the case52. Large-scale deployments therefore need labor-market impact assessments that shape staffing benchmarks, reskilling programs, and wage structures, keeping progress humane38. Because data distributions shift—dramatically so during events like COVID-19—models need continual calibration and monitoring53; governance should mandate drift-detection dashboards, external validation, and agile updates48.
On top of that, early studies already flag automation bias. Prinster□let□lal54. showed feature-based explanations improve accuracy yet foster faster, less critical acceptance, while Patil□let□lal55. warn that unquestioned trust may heighten burnout and liability. Both were short term; longitudinal work must test whether routine exposure blunts clinicians ‘ability to spot hallucinations or edge-case errors.
Before nationwide deployment, small and representative pilots should precede national roll-outs, revealing how diverse users weigh AI advice and guiding interface refinements. Budgets must fund these cycles of testing, training, recalibration, and monitoring50, acknowledging hidden implementation costs and AI’s lifelong upkeep. To close evidence gaps, stepped-wedge cluster trials56, difference-in-differences analyses57, and sequential mixed-methods studies—supplemented by long-term drift-monitoring cohorts—may yield causal, scalable insights while capturing behavioral adaptations and safety signals. Paired with transparent cost accounting and workforce modules, such designs will turn scattered demonstrations into a robust evidence base.
Current evidence on economics of AI is constrained by methodological heterogeneity, limited use of real-world implementation data, and a predominant reliance on theoretical and algorithmic models. While projections suggest substantial savings and operational gains, these benefits remain contingent on how AI is integrated within complex clinical environments that allow widespread adoption of the technology. Policy strategies that align regulatory safeguards with practical implementation challenges—while addressing trust, data governance, and equity—will be central to realizing the sustainable value of AI in healthcare.
METHODS
This study employed a systematic review methodology to synthesize existing literature on the macro-level impact of AI on healthcare systems, with a specific focus on health system costs and spending, workforce productivity, and administrative efficiency. Unlike systematic reviews, the systematic review approach allows for the integration of diverse types of evidence, making it particularly well-suited to emerging and multidisciplinary topics such as AI-driven health system reform.illustrates the conceptual framework guiding this review. The framework positions AI as a system-xlevel catalyst capable of generating gains across three interrelated domains: health economy, workforce productivity, and administrative efficiency. The overlapping nature of these gears reflects the interconnected effects of AI interventions, in which improvements in one area (e.g., administrative automation) may drive ripple effects in overall cost savings and labor optimization.
A comprehensive literature search was conducted using four major databases: PubMed, and Google Scholar. The search covered literature published between January 2020 and July 28 2025. Keywords and Medical Subject Headings (MeSH) were selected to reflect the core themes of the study, including terms related to artificial intelligence (such as “AI,” “machine learning,” “natural language processing,” “Neural Networks,” “Computer-Assisted Diagnosis,” “Decision Support Systems, Clinical,” “digital health,” “artificial intelligence-assisted,” “artificial intelligence-based,” “artificial intelligence-aided,” “deep learning,” “neural networks,” “AI intervention s,” “clinical decision support,” “computer aided,” and “predictive analytics”), health system economics (including “health expenditure,” “healthcare cost,” “Cost-Benefit Analysis,” “Cost Savings,” “cost analysis,” “economic impact,” and “health care cost reductions”), workforce and productivity (such as “workforce,” “workload,” “productivity,” “workload reduction,” “labor productivity,” “health systems,” “health systems,” interpretation times and ““clinician workload”), and administrative efficiency (including “Efficiency, Organizational,” “administrative efficiency,” “Workflow,” “documentation time,” “claims processing,” “reporting times,” “administrative burden,” and “workflow efficiency”). Boolean operators were used to structure the search and connect concept clusters, allowing for the inclusion of studies that explored intersections such as AI and health spending or AI and administrative efficiency.
Studies were eligible for inclusion if they examined the macro-level impact of AI on healthcare systems and reported outcomes related to total cost, national or regional health expenditure, workforce productivity, or the efficiency of administrative processes. Eligible studies employed empirical methods, simulation models, structured reviews, or policy analysis and were published in peer-reviewed journals or as policy reports from reputable health or economic institutions. Studies were excluded if they focused solely on patient- or clinic-level cost-effectiveness or cost-utility analyses; if they discussed only clinical outcomes without broader system-level implications; if they presented algorithm development without application to system performance; if they lacked clear data or methodological frameworks; or if they were non–peer-reviewed publications such as commentaries, editorials, or opinion pieces.
Given the systematic structure of this review, the synthesis of findings was conducted thematically, guided by the three pre-specified domains: health system costs and macro-level spending, workforce productivity and labor restructuring, and administrative efficiency and system process optimization. The analysis also aimed to identify recurring themes, gaps in the evidence, methodological challenges, and insights specific to regional or health system contexts. For each included study, relevant information was extracted, including the country or region of study, the domain of AI application, study design, outcome measures, and key findings related to cost, productivity, or administrative efficiency. A flow diagram of records found, screened, selected, and excluded with corresponding exclusion criteria is shown in Figure 2. The remaining 24 studies were read full text.
Data Availability
No data is used or generated in this systematic review
Acknowledgements
We thank the Yushan Fellow Program by the Ministry of Education (MOE), Taiwan for the financial support. (MOE-112-YSFMN-0003-002-P1)