The financial challenges of artificial intelligence integration in healthcare: lessons from the past and considerations for the future

Introduction

Healthcare spending in the United States (U.S.) continues to rise at an accelerating pace. In 2023, national health expenditures were projected to have grown by 7.5%, surpassing the nominal gross domestic product (GDP) growth rate of 6.1% (1). As a result, healthcare spending as a share of the nation’s economy increased to 17.6% (1) (Figure 1A). According to the annual National Health Expenditure Accounts reports, total U.S. health care spending in 2023 reached approximately $4.9 trillion, equating to $14,570 per person (3) (Figure 1B). Over the full projection period from 2023 to 2032, health spending is expected to grow at an average annual rate of 5.6%, consistently outpacing the nominal GDP growth rate of 4.3% (1). As health care costs continue to rise, policymakers and industry leaders are exploring innovative solutions, including the integration of artificial intelligence (AI), to improve efficiency and reduce expenses (4,5).

Figure 1 An overview of AI applications in healthcare, trends in health expenditures as a percentage of GDP, and health expenditures per capita. (A) Health expenditures per capita in the U.S. and comparable high-income countries [2022]. This figure compares health expenditures per capita (in U.S. dollars) across several high-income countries, highlighting the significantly higher spending in the U.S. ($12,555) relative to the average of comparable countries ($6,651). The data is sourced from OECD. The U.S. outpaces all other nations in healthcare spending per capita, with expenditures nearly double the comparable country average (2). (B) Trends in health expenditures as a percentage of GDP in the U.S. and comparable countries [1970–2022]. This figure illustrates the rising trajectory of healthcare expenditures as a percentage of GDP in the U.S. (green) compared to the average of comparable high-income countries (blue) from 1970 to 2022. The data is sourced from the OECD. The U.S. consistently spends a higher proportion of its GDP on healthcare relative to peer nations, with a sharp increase observed around 2020 (2). (C) Generalized overview of AI applications in healthcare. AI, artificial intelligence; GDP, gross domestic product; OECD, Organization for Economic Co-operation and Development; U.S., United States.

AI has the potential to introduce significant cost-effective measures in health care as evidenced by efforts to develop applications that streamline insurance claim processing for more efficient reimbursement, automate aspects of healthcare delivery to optimize staffing requirements, and enable earlier detection of diseases to prevent complications and reduce long-term treatment costs (4) (Figure 1C). AI-driven tools have already begun transforming the industry, with nearly 875 AI applications approved by the Food and Drug Administration (FDA) since 2019, nearly 700 of which are designed for radiology (6). These advancements suggest that AI could play a crucial role in improving health care efficiency by improving turnaround times and addressing some of the financial burdens within the system. However, despite these promising developments, integrating AI into health care presents substantial financial and structural challenges that call into question its overall cost-effectiveness (4,5). The true financial impact of integrating AI into healthcare should be measured based on the costs related to developing and investing in infrastructure necessary to maintain AI-based systems along with the money saved by the increased efficiency and improved health outcomes.

While AI offers promising opportunities to enhance health care delivery, its implementation comes with significant upfront and recurring costs, complex regulatory hurdles, and additional administrative expenses (5). These hidden and direct costs must be carefully considered to determine whether AI will ultimately contribute to reducing overall health care expenditures. By examining the financial implications of AI adoption, drawing comparisons to electronic health record (EHR) implementation, and discussing systemic barriers to cost-effectiveness, this paper seeks to provide a critical analysis of AI’s true financial impact on the healthcare industry.

Administrative costs and the lack of billable codes for AI services

Administrative expenses constitute a significant portion of healthcare spending in the U.S., driven primarily by billing and coding, physician administrative activities, and insurance administration (7). According to 2019 estimates, administrative costs account for 15% to 25% of total national healthcare expenditures, amounting to $600 billion to $1 trillion annually (7). Compared to other countries, the U.S. healthcare system is uniquely burdened by market-driven complexities, such as multiple insurers and plan variations, which contribute to higher costs (7). In contrast, other nations standardize and centralize administrative functions, including claims processing and pricing regulations, resulting in lower administrative expenses (7). Structural barriers, such as pre-negotiated insurance payment rates and a lack of consumer price transparency, further complicate cost-saving strategies and limit the potential for efficiency-driven reductions in healthcare spending (4).

AI has the potential to reduce administrative inefficiencies by automating billing, claims processing, and utilization management (7). AI-driven interoperability of medical records could help streamline administrative tasks across different payers and providers, reducing redundancy and improving efficiency (7). However, AI implementation itself comes with significant costs, and its ability to replace complex administrative negotiations, such as insurer-provider price setting, remains uncertain (4,8). Pre-negotiated insurance payment rates may also limit AI-driven cost reductions. Furthermore, the financial landscape of AI integration remains uncertain due to the absence of billable codes for AI services, which prevents direct reimbursement and makes it challenging for providers to justify investments in AI technologies (9). Although new Current Procedural Terminology (CPT) codes have been introduced for automated AI applications in radiographic studies, they are typically classified as category III codes, where reimbursement is not guaranteed and payment decisions are left to Medicare contractors and private insurers (9). This lack of standardized reimbursement pathways adds to the financial risks of AI adoption. Additionally, AI algorithms often function as supportive tools rather than standalone diagnostic services, making it difficult to establish them as distinct billable services separate from existing radiology CPT codes (9). Without clear reimbursement structures, healthcare institutions face barriers in leveraging AI’s full potential to reduce administrative burdens while maintaining financial viability.

Upfront and recurring costs of AI implementation

Developing and implementing AI in healthcare comes with substantial upfront and recurring costs, which vary depending on the complexity of the AI model and deployment strategy (Table 1). Basic AI integration into an existing healthcare application can cost around $40,000, whereas a custom deep learning solution may exceed $100,000 (8). For specific use cases, a decision tree-based classifier for patient readmission prediction costs between $35,000–$45,000, while a deep learning model for cancer diagnosis ranges from $60,000–$100,000 (8). More advanced models, such as generative adversarial networks (GANs) used for medical image synthesis, require specialized expertise and high computational resources, with development costs surpassing $200,000 (8).

Beyond development, the choice between cloud-based and on-premises AI deployment influences operational expenses. Cloud-based AI models have recurring costs starting at $430–$650 per month for simple models, but GAN-based models running on tensor processing units (TPUs) can incur cloud costs of $5,000–$15,000 per month (8). While on-premises deployment avoids continuous cloud-related expenses, it requires significant investment in hardware, with costs starting at $5,000 for simple AI models, $20,000–$50,000 for deep learning models, and exceeding $100,000 for high-performance GAN applications (8). Additionally, integrating AI into existing EHRs and electronic medical records (EMRs) systems costs between $7,800 and $10,400, with middleware/application programming interface (API) development adding at least $10,000 (8). Customizing user interfaces and updating legacy systems further increase expenditures, requiring at least $10,000 and up to $35,000 for analysis alone. Off-the-shelf AI solutions may offer lower upfront costs, typically $10,000–$50,000 for integration, but come with ongoing licensing fees (8). Retraining AI models is another cost consideration, with expenses starting at $10,000 for classic machine learning models and significantly higher costs for fine-tuning large language models (LLMs) (8).

Data acquisition and management are also significant cost drivers. Collecting high-quality medical data is expensive due to privacy concerns and data scarcity, often requiring organizations to purchase commercial datasets, which can cost tens of thousands of dollars. Data-sharing agreements introduce additional legal and administrative expenses (8). Data labeling for supervised learning starts at $10,000, even when leveraging generative AI for annotation, as human review remains necessary (8). Data cleaning and preprocessing costs begin at $10,000, depending on dataset size (8). Ensuring compliance with regulations like Health Insurance Portability and Accountability Act (HIPAA) introduces further financial burdens, with certification expenses ranging from $10,000 to over $150,000, based on an organization’s specific compliance requirements (8). While AI offers significant potential benefits in healthcare, these costs present considerable financial and logistical barriers that must be addressed before widespread adoption can be achieved.

Germany offers a compelling model for addressing many of the logistical and regulatory challenges associated with data storage and administration. Through its National Health Data Lab, Germany has established a centralized system that pseudonymizes the health data of approximately 90% of its population enrolled in the government health system (10). This approach not only streamlines data interoperability and improves communication across care settings, but also facilitates research by making large-scale, standardized datasets more accessible (10). Importantly, by operating within a unified infrastructure, hospitals are better equipped to comply with medical confidentiality and data privacy laws, illustrating how centralized governance can reduce institutional burden while promoting ethical AI development (10).

Germany’s progress in digital health integration is further exemplified by the success of Hamburg-based startup Elea, which developed an AI agent tailored to modernize legacy pathology systems that often struggle to interface with EMRs (11). By embedding top-down functionality into its platform, Elea has replaced siloed software with a voice-driven, AI-powered workflow system that significantly reduces diagnostic turnaround times. This vertically integrated “pathology operating system” streamlines reporting, automates routine lab tasks, and maintains compliance with strict data privacy laws through pseudonymization and secure cloud infrastructure (11). While Elea’s success thus far has been limited to Germany’s nationalized healthcare system, where centralized oversight and interoperability enable such innovation, it illustrates the potential of AI to transform administrative and diagnostic efficiency when supported by unified digital infrastructure and clear regulatory pathways.

In the U.S., AI applications are commonly used as separate tools within EHR environments, with medical scribe AI tools serving as a prime example. ScribeAI has published cost information for widely used AI-powered scribe solutions, including Abridge and DAX/Nuance. Abridge is priced at $99 per user per month, while DAX/Nuance costs significantly more at $700 per user per month (12,13). These costs highlight the ongoing financial burden associated with AI adoption in clinical settings. Overall, AI implementation in healthcare requires significant financial investment, with costs influenced by model complexity, integration needs, infrastructure, and data processing expenses.

The cost implications of AI-driven early detection

AI has been widely touted as a tool for reducing the overall cost of care by enabling earlier disease detection. A study highlighted the ability of well-trained recurrent neural networks to diagnose breast cancer in a timely manner (14). However, earlier diagnosis is not universally beneficial. Historical examples underscore the potential pitfalls of early detection without sufficient clinical evidence. In 1985, Japan introduced mass screening of infants for neuroblastoma at 6 months of age without the benefit of unbiased clinical trials (14). The campaign was flawed due to lead-time bias, where survival was measured from the time of diagnosis rather than from birth (15). As a result, although infants were diagnosed at an earlier stage, their overall survival outcomes remained unchanged. Instead, this led to an increase in the length of time patients lived with a disease classification rather than a meaningful extension of life expectancy.

Additionally, length-time bias played a role, as screening was more effective at detecting slow-growing tumors while missing fast-growing, aggressive cases that would manifest clinically (15). Consequently, many infants underwent intensive interventions such as surgery and chemotherapy without significant medical benefit (15). This example underscores that earlier diagnosis does not always equate to better outcomes and can, in some cases, introduce unnecessary costs and patient burdens.

Building off the historical lesson from Japan, another study further highlighted the potential dangers of large scale screening programs. For example, the National Lung Screening Trial (NLST) demonstrated that while standardized screening reduced lung cancer mortality, it also resulted in high false-positive rates and greater use of follow-up diagnostics. Moreover, the trial’s favorable outcomes were partly due to factors not easily replicated in community settings, such as a 1% surgical mortality rate, significantly lower than the 4% seen in the general population, raising concerns about the feasibility and safety of widespread implementation (16). While AI-driven early detection holds promise, its financial and clinical impact must be carefully evaluated to ensure that the benefits outweigh the risks and additional expenditures on healthcare systems. A balanced approach is necessary to maximize the advantages of AI while minimizing inefficiencies and unintended harms.

Administrative and workforce costs of AI implementation

Administrative expenses represent a significant portion of healthcare costs, accounting for 15–25% of total national healthcare expenditures, amounting to $600 billion to $1 trillion annually, according to 2019 estimates (7). Major cost drivers include billing and coding, physician administrative activities, and insurance administration (7). A Forbes article highlighted how a large portion of insurance claims get denied, creating administrative bottlenecks for all stakeholders involved (17). Errors such as filing duplicate claims, entering incorrect insurance ID numbers, or providing incomplete patient information can directly result in claim denial or delayed reimbursements (17). Additionally, the involvement of multiple stakeholders, each with unique documentation and regulatory requirements, can lead to further filing errors and claim denials (17). Human errors in data entry, document handling, or verification can exacerbate inefficiencies, leading to unnecessary costs and delays (17). AI can be a useful tool during data entry to screen documents and ensure that all necessary information has been included for the sake of improving administrative turnaround time.

AI has the potential to streamline claims processing by automating administrative tasks such as billing, claims processing, and utilization management (7). AI-driven interoperability of medical records could enhance administrative workflows across different payers and providers, reducing inefficiencies (7). However, implementing AI in these processes comes with additional costs. Insurance companies would need to invest in AI services which, while improving claims processing efficiency, would likely reduce their profit margins, thereby removing the incentive to adopt AI systems in their workflows. Furthermore, AI applications require a dedicated workforce for implementation, monitoring, and maintenance. This would require retraining and upskilling of current information technology (IT) technological support staff, further limiting the ability to offset costs (7). As a result, insurance companies may pass these expenses onto clients in the form of higher premiums, which could counteract the intended cost-saving benefits of AI integration in healthcare.

Learning from EHRs: a case study in costly implementation

EHR implementation itself presents substantial costs that can serve as a predictor for AI application integration in healthcare. According to an analysis by SPRY, an EHR software company, EHR costs range from $150–$500 per user per month, translating to $1,800–$6,000 per user annually (17). Hardware investments vary widely, from $10,000 to over $100,000, depending on the scale and existing infrastructure (18). First-year maintenance costs range from $10,000 to $50,000, with decreasing support costs in subsequent years (18). Initial training costs typically range from $1,000 to $5,000 per staff member, covering hands-on sessions, webinars, and training materials, while ongoing training for system updates costs an additional $500 to $2,000 per staff member annually (18). On-premises deployment requires a substantial upfront investment between $1,000 and $500,000, along with ongoing hardware and support expenses (18).

Like EHRs, AI applications may introduce administrative burdens and inefficiencies before their long-term benefits are realized. A study on EMR system adoption in a Korean hospital system, including both outpatient and inpatient centers, demonstrated that the implementation process was net value negative for approximately 4 to 5 years before turning net value positive after the 5-year mark (19). This “lag” effect is likely to be a characteristic of AI model adoption as well, as healthcare organizations navigate initial workflow disruptions, staff training, and integration challenges before realizing efficiency gains (19).

The improved efficiencies and productivity achieved with EHR systems in the latter part of the implementation period significantly contributed to the positive return on investment. Similarly, the long-term success of AI integration in healthcare will depend on how well software vendors provide timely updates and how regulatory requirements evolve to minimize disruptions. Ensuring that vendor software upgrades align with user needs and that compliance regulations do not impose excessive operational burdens will be critical in sustaining AI’s long-term value in clinical practice (20). These insights are supported by a cost-benefit analysis conducted on an ophthalmology practice during EHR implementation, which highlighted the necessity of strategic vendor support and regulatory adaptability in achieving long-term financial and operational success (20).

There were also some notable issues with EHR implementation that should be carefully considered when planning for the implementation of AI applications. A major limitation of EHR systems has been the lack of standardization, which directly contributes to poor interoperability across healthcare settings (21). Inconsistencies in terminology, coding systems, data formats, and entry practices not only complicate communication between different systems but also undermine the accuracy of clinical decision-making and research outcomes (21). Errors in data entry, coding, or transcription can compromise patient safety and impair analysis. Even within a single hospital, providers may struggle to access patient information if components of the EHR system are not fully integrated (21). These challenges have opened the door for AI-powered solutions, such as digital scribes, which can aid in clinical documentation by transcribing audio notes, standardizing language, and even recommending accurate International Classification of Diseases (ICD) codes to replace vague or surrogate diagnoses (22). By reducing variability and promoting consistency, such tools have the potential to improve interoperability and enhance the administrative efficiency of clinical workflows.

AI in drug development: faster, but not necessarily cheaper

AI has revolutionized drug development by enabling pharmaceutical companies to analyze vast datasets, significantly accelerating the drug discovery process (23). AI-driven virtual screening techniques have hastened the identification of promising drug candidates, while AI-powered toxicity prediction models help identify potential safety concerns early in development (23). These advancements theoretically should expedite drug development and reduce costs for pharmaceutical companies.

However, despite AI-driven efficiencies, drug prices remain high due to industry practices (4). Pharmaceutical companies prioritize first-in-class drugs to maintain market exclusivity, allowing them to charge premium prices without competition (4). Additionally, pharmacy benefit managers contribute to price inflation through opaque pricing negotiations with insurance companies, limiting open-market competition (4,24). According to the United States Federal Trade Commission (US FTC), UnitedHealth’s Optum, CVS Health’s Caremark, and Cigna’s Express Scripts significantly inflated drug prices, sometimes by hundreds or even thousands of percent, generating $7.3 billion in revenue beyond the original acquisition costs (24). These entrenched industry dynamics suggest that, even with AI, drug prices are unlikely to decrease substantially, as companies continue to maximize profit margins within the existing healthcare framework.

Conclusions

AI offers significant opportunities to enhance efficiency in healthcare, from improving diagnostic accuracy to streamlining administrative processes. However, its financial benefits are not guaranteed. The high costs of AI implementation, coupled with the administrative burden and the complexities of the U.S. healthcare system, pose significant challenges that must be carefully evaluated.

Despite these financial hurdles, the value of AI applications in medicine remains evident. Over time, AI can contribute to a more cost-effective and resource-efficient healthcare system, provided that its deployment is guided by strategic planning. To fully realize AI’s potential, specific areas of focus for policymakers, hospital administrators, healthcare providers, and insurers should include developing standardized methodologies for evaluating the cost-effectiveness of AI interventions, incorporating metrics for both financial and clinical outcomes. It would also be helpful to explore new reimbursement models that incentivize the adoption of cost-effective AI solutions and reward value-based care. Navigating these financial challenges will allow for a more honest assessment of the true cost-effectiveness of AI in medicine.

Acknowledgments

None.

Footnote

Provenance and Peer Review: This article was a standard submission to the journal. The article has undergone external peer review.

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-104/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-104/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Fiore JA, Madison AJ, Poisal JA, et al. National Health Expenditure Projections, 2023-32: Payer Trends Diverge As Pandemic-Related Policies Fade. Health Aff (Millwood) 2024;43:910-21. [Crossref] [PubMed]
2. Turner A, Miller G, Lowry E. High U.S. Health Care Spending: Where Is It All Going? Commonwealth Fund 2023. Available online: https://www.commonwealthfund.org/publications/issue-briefs/2023/oct/high-us-health-care-spending-where-is-it-all-going#:~:text=A%202021%20study%20by%20McKinsey,NHE%20spending%20in%20these%20settings
3. Centers for Medicare & Medicaid Services. National Health Expenditures 2023 Highlights. 2024. Retrieved February 8, 2025, Available online: https://www.cms.gov/data-research/statistics-trends-and-reports/national-health-expenditure-data/historical
4. Coleman K. Lowering health care costs through AI: The possibilities and Barriers. Paragon Health Institute 2024. Available online: https://paragoninstitute.org/private-health/lowering-health-care-costs-through-ai-the-possibilities-and-barriers/
5. Khanna NN, Maindarkar MA, Viswanathan V, et al. Economics of Artificial Intelligence in Healthcare: Diagnosis vs. Treatment. Healthcare (Basel) 2022;10:2493. [Crossref] [PubMed]
6. U.S. Food and Drug Administration. Artificial Intelligence-Enabled Medical Devices. 2024. Available online: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices
7. Chernew M, Mintz H. Administrative Expenses in the US Health Care System: Why So High? JAMA 2021;326:1679-80. [Crossref] [PubMed]
8. Alkhaldi N. Assessing the cost of implementing AI in healthcare. ITRex 2024. Available online: https://itrexgroup.com/blog/assessing-the-costs-of-implementing-ai-in-healthcare/
9. Smetherman D, Golding L, Moy L, et al. The Economic Impact of AI on Breast Imaging. J Breast Imaging 2022;4:302-8. [Crossref] [PubMed]
10. Baumgart DC, Kvedar JC. Germany and Europe lead digital innovation and AI with collaborative health data use at continental level. NPJ Digit Med 2025;8:215. [Crossref] [PubMed]
11. Lomas N. Elea AI is chasing the healthcare productivity opportunity by targeting pathology labs’ legacy systems. TechCrunch 2025. Available online: https://techcrunch.com/2025/03/12/elea-ai-is-chasing-the-healthcare-productivity-opportunity-by-targeting-pathology-labs-legacy-systems/
12. Trang B. Hospitals want to buy doctors’ happiness with Nuance’s AI scribe. What they’re paying varies. STAT 2023. Available online: https://www.statnews.com/2023/06/01/hospitals-nuance-dax-scribe-artificial-intelligence/
13. Wang E. AI Medical Scribe Tools: A Comprehensive Price comparison. ScribeAI 2024. Available online: https://scribeai.us/zh/blog/ai-medical-scribe-tools-a-comprehensive-price-comparison
14. Darbandi MR, Darbandi M, Darbandi S, et al. Artificial intelligence breakthroughs in pioneering early diagnosis and precision treatment of breast cancer: A multimethod study. Eur J Cancer 2024;209:114227. [Crossref] [PubMed]
15. Evans I, Thornton H, Chalmers I, et al. Earlier is not necessarily better. In: Testing Treatments: Better Research for Better Healthcare. 2nd edition. London: Pinter & Martin; 2011.
16. National Lung Screening Trial Research Team. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011;365:395-409. [Crossref] [PubMed]
17. Agarwal S. The AI revolution in medical claims processing. Forbes 2024. Available online: https://www.forbes.com/sites/shashankagarwal/2024/03/28/the-ai-revolution-in-medical-claims-processing/
18. Carter A. Guide to the Cost of EHR Implementation for healthcare providers. SPRY 2024. Available online: https://www.sprypt.com/blog/guide-to-the-cost-of-ehr-implementation-for-healthcare-providers
19. Choi JS, Lee WB, Rhee PL. Cost-benefit analysis of electronic medical record system at a tertiary care hospital. Healthc Inform Res 2013;19:205-14. [Crossref] [PubMed]
20. Wiggins RE Jr, Fridl DC. Analysis of the Financial Return of Electronic Health Records. Ophthalmology 2016;123:214-6.e2. [Crossref] [PubMed]
21. Henry TA. 7 EHR usability, safety challenges—and how to overcome them. AMA 2023. Available online: https://www.ama-assn.org/practice-management/digital-health/7-ehr-usability-safety-challenges-and-how-overcome-them
22. Sasseville M, Yousefi F, Ouellet S, et al. The Impact of AI Scribes on Streamlining Clinical Documentation: A Systematic Review. Healthcare (Basel) 2025;13:1447. [Crossref] [PubMed]
23. Bhatia N, Khan M, Arora S. The Role of Artificial Intelligence in Revolutionizing Pharmacological Research. Current Pharmacology Reports 2024;10:323-9.
24. FTC finds major pharmacy benefit managers inflated drug prices for $7.3 billion gain. Reuters 2024. Available online: https://www.cnbc.com/2025/01/14/unitedhealth-group-cvs-health-cigna-inflated-drug-prices-ftc-says.html?msockid=37cd6e3a0181605034ba781800af6190