Toward human+ medical professionals: navigating AI integration in healthcare to enhance human expertise

The role of digital innovations is expanding across all industries, with particularly significant impacts on the healthcare sector (Schiavone et al., 2021; Ali Mohamad et al., 2023). Among these innovations, artificial intelligence (AI) has emerged as a transformative force, reshaping medical procedures, service delivery and professional roles (Alowais et al., 2023; Guo et al., 2025). This transformation has significant implications for innovation management, requiring new approaches to govern technological integration and its organisational impact.

AI is increasingly recognised as a driver of innovation in healthcare, enabling service transformation through improved decision-making and optimised resource allocation (Huang and Rust, 2021; Pantano and Scarpi, 2022). Its adoption compels healthcare organisations to develop new capabilities and reconfigure their innovation practices (Rialti et al., 2019; Chakravorty et al., 2020). In addition, AI facilitates cultural and structural changes within healthcare organisations, creating an environment that promotes continuous learning and adaptation (Bohr and Memarzadeh, 2020).

Despite its potential, adopting AI in healthcare presents significant challenges due to the unique characteristics of the sector. The healthcare sector, indeed, is a distinctive, high-stakes, knowledge-intensive domain where full automation is both technically and ethically constrained. As a highly professionalised, knowledge-intensive industry, healthcare relies on expert judgement, individualised decision-making and non-standardised processes (von Nordenflycht, 2010). Unlike in other industries where automation enhances repetitive tasks, healthcare involves high-stakes decisions on diagnoses, treatments and surgeries that directly affect human lives. Delegating such responsibilities entirely to machines gives rise to complex ethical and practical concerns, especially around issues of trust, accountability and transparency in AI-assisted decisions. Nevertheless, when employed to support rather than replace clinicians, AI has the potential to improve healthcare delivery. It can alleviate cognitive load, uncover complex patterns and enable more accurate, personalised and timely interventions. From this perspective, AI is not a substitute for human intelligence, but rather a complement to it, capable of augmenting clinical expertise and opening new paths for innovation. The implementation and adoption of AI-driven solutions in the healthcare sector should take into account their implications from technological, organisational and ethical perspectives. Understanding how AI enhances human capabilities, in terms of professional expertise and innovation management, is therefore both necessary and timely.

The discussion is reflected in the broader literature on AI and innovation management, which highlights an ongoing debate about whether AI should be seen as a complement to or substitute for human expertise, which has different implications for innovation capabilities (Raisch and Krakowski, 2021; Krakowski, 2025). Existing frameworks often conceptualise human–AI interaction along a continuum from augmentation to automation, reflecting different degrees of technological integration and human involvement. However, much of this research emphasises AI's predictive and optimisation capabilities, providing limited insight into its contribution to creativity-, judgement- and interpretation-based innovation processes (Johnson et al., 2022; Krakowski, 2025). At the same time, the role of human professionals is evolving. Rather than performing repetitive or narrowly defined tasks, professionals are increasingly engaged in higher-order activities – such as problem framing, goal setting and system configuration – that are essential to the effective functioning of AI systems (Verganti et al., 2020). This shift requires a rethinking of expertise, as it demands new skill sets that go beyond technical proficiency, incorporating interpretive, integrative and design-oriented competencies (Krakowski et al., 2023). Therefore, the discussion on AI integration in medical practice still misses a systematic examination of how different configurations of human-AI interaction reshape medical expertise and innovation management practices.

These dynamics are giving rise to Human + professionals, intended as the co-evolution of human and AI capabilities that shifts expertise from task execution to the orchestration and interpretation of AI-enabled systems. In this perspective, the collaboration between humans and AI is no longer seen as a choice between augmentation and automation, but as complementary roles working together. As AI capabilities continue to evolve, the discussion is shifting towards a more nuanced understanding of human–AI integration and its transformative potential for healthcare professionals and organisations (Kopalle et al., 2022).

In line with this perspective, the paper moves beyond the automation versus augmentation dichotomy, adopting Human + as a more nuanced integration logic, grounded in the idea that AI can enable human–machine interaction by enhancing rather than replacing human capabilities, reshaping clinical roles and organisational routines.

This study, therefore, aims to analyse the different levels of human–AI interaction in healthcare and their impact on human expertise within innovation management practices. Specifically, the research seeks to answer the following question: How does the use of AI in healthcare transform medical practitioners into Human + professionals, and what are the implications for innovation management practices?

In order to answer the research question, a qualitative multiple case study design (Yin, 2017) was adopted. We selected four cases that represent the best practices in the integration of AI into healthcare services. Data were collected from secondary sources, including written and video interviews, public documents, industry reports and observational material. These were triangulated with insights from scientific literature to enhance the validity and robustness of the findings (Yin, 2017). All collected material was transcribed and subjected to qualitative content analysis (Strauss and Corbin, 1990). Data analysis was guided by the three-dimensional framework proposed by Bolton et al. (2018), which offers an integrated perspective on service innovation. This framework is particularly well suited to the healthcare sector (Polese and Carrubbo, 2017; Saviano et al., 2022) as it captures the dynamics and tensions involved in shifting from traditional to digitalised services (Saviano et al., 2023). Specifically, we applied the framework to examine the digital, physical and social dimensions of each case, enabling us to identify how AI reshapes service interactions across these dimensions.

The main contribution of this article is the development of a taxonomy of AI integration in healthcare, grounded in the Bolton et al. (2018) framework, using it to explain the emergence of Human + professionals and distinct innovation trajectories.

The findings contribute to the categorisation of different levels of AI integration, linking them to AI and innovation categories and assessing their impact. This approach enables the identification of key implications for innovation management and highlights areas where human expertise remains indispensable. In addition, the study explores actionable strategies to facilitate the transition to Human + practitioners, providing insights that bridge theoretical perspectives with practical applications. By contributing to the fields of innovation and organisational management (Davenport and Kalakota, 2019), this study provides valuable guidance for both academic scholars and industry professionals navigating the evolving landscape of AI in healthcare.

The article is structured as follows. Section 2 provides the theoretical background on AI in innovation management, paying particular attention to the healthcare sector and the evolving role of medical expertise. It also introduces the theoretical framework used to interpret service innovation in healthcare. Section 3 outlines the research methodology. Section 4 presents the study's findings, which are analysed and discussed further in Section 5. Section 6 offers the main conclusions and outlines the study's implications, limitations and directions for future research.

In the field of innovation management, the potential of AI to revolutionise processes by enhancing creativity, improving decision-making and optimising operational efficiency is widely recognised (Agostini et al., 2020; Füller et al., 2022; Haefner et al., 2023; Mariani et al., 2023). Innovation management involves systematically generating, selecting, implementing and sustaining ideas that create organisational value (Tidd and Bessant, 2020) and relies on idea generation (Keum and See, 2017), creative problem-solving (Proctor, 2010) and organisational innovation capabilities (Schewe, 1994).

AI supports different stages of the innovation process by automating repetitive tasks, uncovering patterns in large datasets and enhancing decision-making (Haefner et al., 2021; Freisinger et al., 2024). It assists with idea generation, selection and implementation by providing data-driven insights and enabling managers to identify opportunities with greater competitive potential (Kumar et al., 2019; Haefner et al., 2021). AI also fosters collaboration, facilitating discussions and stimulating ideation (Bouschery et al., 2023). By managing routine tasks, AI allows human resources to focus on complex and strategic activities, boosting both efficiency and innovation (Mariani et al., 2023). Debates persist on AI as a complement or substitute for human expertise (Raisch and Krakowski, 2021; Krakowski, 2025).

Research on human–AI interaction has primarily been framed within two paradigms: augmentation and automation. Augmentation refers to AI augmenting human capabilities, enabling individuals to perform more complex, creative and strategic tasks (Wilson and Daugherty, 2018; Jarrahi, 2018). In this scenario, AI acts as a “complementary cognitive artefact” (Krakowski, 2025, p. 6), extending human cognition and facilitating learning through interaction and exploration. Automation, on the other hand, involves the complete delegation of tasks to AI systems, which act as “competitive artifact” (Krakowski, 2025, p. 10), minimising human involvement in order to increase efficiency and consistency (Brynjolfsson and McAfee, 2014; Davenport and Kirby, 2016). The decision to pursue automation or augmentation is largely determined by the nature of the task (Johnson et al., 2022). According to the literature, structured and narrowly defined tasks are more conducive to automation, where AI systems can efficiently execute processes with minimal human supervision (Tegmark, 2017). In contrast, complex, ambiguous and broadly defined tasks, particularly those that require creativity, judgement or adaptability, remain more difficult for AI to solve independently and are better suited to augmentation, where AI supports and extends human capabilities (Brynjolfsson and McAfee, 2014; Davenport and Kirby, 2016; Wilson and Daugherty, 2018).

Several frameworks conceptualise human–AI interaction along the augmentation–automation continuum, reflecting different levels of technological integration and human involvement. Freisinger et al. (2024) propose a five-level framework ranging from full human control to full AI automation (Garbuio and Lin, 2021; Raisch and Krakowski, 2021; Seeber et al., 2020). Gama and Magistretti (2025) present a three-part taxonomy that categorises AI applications into replace, augment and reveal. Building on the taxonomy (Gama and Magistretti, 2025), Biel et al. (2025) introduce a two-level framework that distinguishes between task-level and approach-level AI integration. Spanjol et al. (2024) propose a four-stage evolutionary model that charts the progression of AI from a tool, intended as an advanced data processor to improve efficiency, to an innovation orchestrator, where AI evolves into an autonomous, self-learning system that manages innovation initiatives, scans for emerging opportunities, develops strategies and implements innovations with minimal human oversight.

Research suggests augmentation and automation evolve cyclically (Raisch and Krakowski, 2021), with organisations initially adopting augmentation before advancing automation (Choudhary et al., 2025). The challenge lies in balancing automation's efficiency with augmentation's role in fostering creativity and strategic thinking (Verganti et al., 2020).

The adoption of AI, whether for augmentation or automation, has significant implications for innovation capabilities, often requiring the development of new skills and competencies (Correani et al., 2020; Alzoubi and Mishra, 2025; Gama and Magistretti, 2025). AI adoption influences the capability lifecycle, acting as a selection event that drives the transformation or creation of new organisational capabilities (Gama and Magistretti, 2025). In this regard, Iansiti and Lakhani (2020) distinguish between weak AI, which automates routine and repetitive tasks without significantly changing the broader system and strong AI, which supports decision-making and generates novel solutions (Shrestha et al., 2019; Garbuio and Lin, 2021). Strong AI requires new interpretive capabilities to translate AI-driven insights into actionable innovation strategies (Verganti et al., 2020). With respect to AI adoption, Gama and Magistretti (2025) categorise innovation capabilities in enabling and enhancing. Enabling capabilities support the development, implementation and adoption of AI (Brock and von Wangenheim, 2019; Bessen et al., 2022), while enhancing capabilities emerge through collective experimentation and help firms improve innovation processes and create value (Verganti et al., 2020; Bag et al., 2021; Kinkel et al., 2022). In addition, individual capabilities are increasingly recognised as a determinant of the effectiveness of AI appropriation (Cimino et al., 2024). AI acts as a cognitive extension, enhancing ideation, problem-solving and decision-making processes. This perspective frames AI not just as a tool for automation, but as an enabler of adaptive and collaborative innovation environments where human expertise and AI capabilities continuously co-evolve.

The literature review shows that AI research in innovation management remains relatively underdeveloped. While numerous studies highlight the benefits of AI in enhancing innovation processes (Haefner et al., 2021; Freisinger et al., 2024; Mariani et al., 2023; Alzoubi and Mishra, 2025), much of the existing research focuses on AI's predictive and optimisation capabilities, primarily within a performance-driven logic of efficiency, accuracy and process optimisation (Mariani and Dwivedi, 2024). This narrow perspective limits the understanding of AI's potential impact on creativity-driven and innovative activities (Johnson et al., 2022; Krakowski, 2025). In addition, the human role in the innovation process is increasingly shifting from operational tasks to higher-level tasks, including problem identification, goal setting and system configuration, which enable AI to operate autonomously (Verganti et al., 2020). This shift requires the development of new skills that differ from traditional skill sets (Krakowski et al., 2023). However, research on AI as a driver of innovation capability development remains scarce, leaving a critical gap in understanding how AI supports not only existing capabilities but also the creation of new ones (Gama and Magistretti, 2025). Furthermore, there is limited research on theoretical models and practical strategies that guide organisations in using AI to augment human potential and drive value creation (Gama and Magistretti, 2025; Krakowski, 2025). Several studies call for future research examining how organisations can strategically integrate AI, balancing augmentation and automation to ensure it serves as an enabler of innovation rather than a substitute for human ingenuity (Johnson et al., 2022; Mariani and Dwivedi, 2024; Krakowski, 2025). This necessitates expanding current theory and practice to address the cyclical dynamics between automation and augmentation (Raisch and Krakowski, 2021), with a focus on emerging capabilities associated with AI adoption (Gama and Magistretti, 2025). Additionally, a growing body of research underscores the need to explore the interdependence between AI and humans, emphasising how they co-evolve as mutually shaping systems (Spanjol et al., 2024). Understanding this interplay is critical for advancing innovation management frameworks that harness AI's synergistic potential in capability development and organisational transformation.

AI is widely recognised as a transformative force in healthcare, enhancing decision-making and patient outcomes (Alowais et al., 2023; Guo et al., 2025). Applications include medical imaging, predictive analytics, robotic surgery and intelligent virtual assistants (Tomašev et al., 2019). AI goes beyond automation to improve care quality, reduce costs and expand healthcare access (Leone et al., 2021; Cannavale et al., 2022). Hospitals using AI report fewer diagnostic and treatment delays, leading to better clinical outcomes (Hong and Lee, 2018). AI-driven predictive analytics enhance disease diagnosis, streamline workflows and enable personalised treatment (Jiang et al., 2017). AI also improves surgical precision, reduces complications and optimises treatment strategies in population health management (Alowais et al., 2023). Advances in telemedicine, mental health and drug discovery further enhance healthcare efficiency (Bohr and Memarzadeh, 2020). AI also facilitates healthcare management, supporting data integration and strategic decision-making (Secundo et al., 2021; Pietronudo et al., 2022). Healthcare professionals and managers are increasingly relying on AI to manage vast datasets derived from electronic health records (EHRs), clinical findings, prescriptions, imaging procedures and mobile health technologies (El Morr and Ali-Hassan, 2019; Ilangakoon et al., 2022). AI-driven tools enable more informed decision-making, improving diagnostic accuracy, patient management and resource allocation (Secundo et al., 2021; Pietronudo et al., 2022; Tortorella et al., 2022). AI's ability to structure and analyse large volumes of heterogeneous healthcare data supports not only individual patient care, but also broader strategic and policy-level decision-making within healthcare organisations.

AI is also recognised as a key enabler of innovation in healthcare, facilitating service transformation through enhanced decision-making processes and optimised resource allocation (Huang and Rust, 2021; Pantano and Scarpi, 2022). At the organisational level, AI facilitates predictive maintenance of medical equipment, improves inventory management and enhances strategic planning (Flynn, 2019). The implementation of AI-driven solutions requires the development of new competencies within healthcare institutions, promotes digital transformation and reshapes innovation management practices (Rialti et al., 2019; Chakravorty et al., 2020). In addition, AI facilitates cultural and structural changes within healthcare organisations, creating an environment conducive to continuous learning and adaptation (Bohr and Memarzadeh, 2020). Given the growing challenges facing healthcare systems, such as ageing populations, resource constraints and disparities in access to care, AI-driven innovations offer promising solutions to improve the efficiency, equity and overall quality of healthcare services (Court et al., 2023; Li et al., 2023; Zahlan et al., 2023).

Despite its transformative potential, the adoption of AI in healthcare faces significant challenges due to the sector's strong reliance on clinical expertise and individualised decision-making. Unlike industries characterised by standardised and routine processes, healthcare requires context-sensitive and tacit knowledge, which limits the feasibility of full automation (von Nordenflycht, 2010). These limitations are particularly evident in diagnostics, where AI systems often struggle to replicate the nuanced reasoning that clinicians develop through years of experience (Roppelt et al., 2025). There are also concerns about over-reliance on AI, which could undermine professional autonomy, particularly in contexts where mandatory AI consultation creates ethical dilemmas (Witkowski et al., 2024). Furthermore, the insufficient availability of training data for rare or complex conditions increases the risk of diagnostic inaccuracies and thereby constrains large-scale implementation (Hallowell et al., 2022).

Beyond these technical constraints, broader governance and ethical issues further complicate AI integration. A persistent challenge lies in ensuring that AI-generated recommendations are transparent, interpretable and ethically justifiable (Lysaght et al., 2019). Bias and error remain significant barriers to reliability (Jussupow et al., 2021), as current AI technologies primarily rely on statistical pattern recognition rather than human-like reasoning and contextual awareness (Wiljer and Hakim, 2019). Consequently, AI-driven diagnostic tools may produce ambiguous or inconsistent outcomes, increasing the risk of overdiagnosis or underdiagnosis compared to traditional clinical evaluations (Wong et al., 2019). These concerns are exacerbated by questions of trust and interpretability, as clinicians must reconcile AI-generated insights with their own expertise, often in high-stakes, life-threatening situations where opaque algorithms present ethical and practical challenges (Jussupow et al., 2021; Singh et al., 2024).

In light of these complexities, scholars have increasingly adopted the automation–augmentation paradigm to conceptualise the evolving relationship between humans and AI systems in healthcare. Garbuio and Lin (2019) propose a three-level model of human–AI collaboration, which illustrates the progressive integration of AI into healthcare practices. At the most basic level, assisted intelligence enhances the efficiency of repetitive, rule-based tasks such as data verification and medical image classification by helping clinicians reduce errors and improve diagnostic accuracy. Moving beyond this, augmented intelligence introduces new capabilities that transform clinical workflows and promote a more personalised, preventive approach to care. An example of this is precision medicine, where AI enables treatments to be customised according to patients' genetic and clinical profiles. The final level, autonomous intelligence, sees AI systems capable of operating independently within predefined parameters and managing aspects of patient care with minimal human intervention. This concept is illustrated by the emerging notion of the ‘doctorless hospital'.

However, recent literature emphasises that the heavy reliance of healthcare on human judgement, ethical sensitivity and contextual complexity makes full automation both impractical and undesirable (Spring et al., 2022; Jussupow et al., 2021). Consequently, researchers advocate an augmented intelligence approach in which AI complements, rather than replaces, clinical expertise (Guo et al., 2025). This model enhances workflow efficiency, enabling physicians to focus on complex diagnoses and patient-centred care, while also expanding knowledge creation through data-driven insights that foster innovation and continuous learning (Leyer and Schneider, 2021; Raisch and Krakowski, 2021). Building on this, recent studies propose that a balanced and complementary integration of automation and augmentation can maximise clinical effectiveness and innovation potential (Guo et al., 2025). A balanced approach applies both paradigms at comparable levels, whereas a complementary approach leverages their respective strengths synergistically, allowing automation to enhance efficiency and augmentation to deepen expertise and foster creativity.

Despite these insights, the literature calls for further research on human–AI interactions in healthcare, particularly regarding the role of medical professional expertise in AI-integrated service delivery. There is a need for research that further explores how AI intersects with human expertise to support continuous improvement practices within healthcare organisations, promote effective innovation management and ensure the delivery of high-quality patient care.

A widely acknowledged framework for analysing service innovation is the three-dimensional model proposed by Bolton et al. (2018). This framework conceptualises service experiences as emerging at the intersection of three key dimensions – digital, physical and social-each defined along a continuum from low to high. The interaction of these dimensions creates a three-dimensional space subdivided into eight octants, each representing a distinct configuration of service innovation (Figure 1). This structure enables a nuanced analysis of how innovation unfolds across varying degrees of digital density, physical complexity and social presence.

Building on Bolton et al. (2018), digital density refers to the extent to which digital technologies are embedded in the service environment. Low digital density indicates minimal technological integration, where human interactions and physical processes dominate. Conversely, high digital density characterises environments where technology plays a central role. Physical complexity represents the sophistication and tangibility of the physical infrastructure. Low physical complexity reflects simplified environments with minimal reliance on physical resources, while high complexity includes advanced, resource-intensive facilities. Social presence captures the level of human interaction within the service experience. Low social presence is associated with automation and self-service technologies, while high social presence emphasises interpersonal relationships.

The relevance of Bolton et al.’s (2018) framework is particularly evident in healthcare, a sector inherently service-oriented and heavily reliant on human expertise (Polese and Carrubbo, 2017; Saviano et al., 2022). The model proves especially useful for studying healthcare service innovation, as it facilitates the analysis of how different degrees of digitalisation, physical infrastructure and social interaction reshape the patient experience. This is especially pertinent in the context of the rapid adoption of AI technologies in healthcare (Alowais et al., 2023; Guo et al., 2025), where the reconfiguration of service delivery demands a comprehensive understanding of how technological, material and relational dimensions interact.

Within this context, Bolton et al. 's (2018) model allows researchers to map the evolving interplay between AI-based technology, the physical setting of care delivery and social dynamics. Bolton et al. (2018) stress the importance of people and organisations shaping the role of technology in designing and delivering customer experiences. This insight is particularly relevant in healthcare, where gains in efficiency, accuracy and accessibility through AI must be carefully balanced with risks related to depersonalisation, ethical concerns and potential systemic vulnerabilities.

Overall, Bolton et al.'s (2018) framework provides a structured approach for examining how AI adoption is transforming healthcare services. By analysing the interplay of digital, physical and social dimensions, it provides valuable insights into how human experiences are both preserved and reconfigured across varying degrees of human–AI integration.

To explore how the adoption of AI in healthcare affects medical professionals' expertise and innovation management practices, we adopted a qualitative research design based on multiple case studies (Yin, 2017). This approach is particularly appropriate for the inductive exploration of emerging and complex phenomena (Fawcett et al., 2014) and is well-suited for addressing “how” research questions (Yin, 2017), especially in understudied contexts such as AI integration in healthcare. Case studies are also widely recognised in innovation management literature as a robust method for examining innovation dynamics in real-world settings (Goffin et al., 2019). We conducted exploratory case studies, a research approach appropriate when limited empirical evidence exists on the phenomenon under investigation (Yin, 2017). This type of case study is characterised by open-ended inquiry, flexible data collection methods and the absence of predefined hypotheses (Eisenhardt, 1989; Yin, 2017). It allows for a holistic and in-depth understanding of complex dynamics, enabling researchers to describe, interpret and explain relevant mechanisms (Baxter and Jack, 2008; Mills et al., 2009).

Our research focuses specifically on human–AI interaction in medical practices. Accordingly, we applied an information-oriented selection strategy, identifying four case studies that represent best practices in the integration of AI into healthcare services. We deliberately excluded cases still in the development phase, often characteristic of emerging technologies and focused instead on mature AI applications currently in use. These cases were selected based on their capacity to offer rich insights into the service dimensions of AI-enabled healthcare. To ensure the robustness of the research in terms of the selection criteria for the analysed case studies, we followed Tisdell et al. (2025) guidelines, applying the maximum variance criterion between cases in terms of the research field's relevant dimensions. In line with this, the selected cases demonstrate diversity in terms of AI integration models, human–AI interaction configurations and the implications for professional roles and service delivery. This diversity enabled us to examine various AI adoption configurations, identify conditions under which human expertise is crucial and evaluate the impact on innovation management practices.

To ensure triangulation and strengthen the robustness and validity of our constructs, we used multiple sources of secondary data. Specifically, we collected data through: (1) direct observation; (2) online documentation and official archival materials; and (3) publicly available online interviews. Direct observations involved downloading the AI applications from digital platforms (e.g., app stores), where possible and attending webinars that demonstrated system functionalities. These activities enabled us to gain a better understanding of the technological components and operational dynamics of the AI systems under study. We also analysed a broad range of online and archival documents. These included official reports, white papers, clinical studies, company publications, professional magazines, news articles, institutional websites and publicly available multimedia content, such as YouTube videos (Cucari et al., 2023). This enabled us to cross-validate information and gain a comprehensive overview of each case. Following Yin (2017), we integrated academic and industry sources to enhance the depth and reliability of the analysis.

To further enrich our understanding of the phenomena under investigation, we analysed a series of interviews with the founders and developers of the selected AI systems. The interviews analysed were not conducted by the authors but obtained from publicly accessible sources such as the official websites of selected AI applications, online magazines and video platforms (e.g., YouTube). These interviews provided insight into the original intentions behind the technologies, the perceived benefits for healthcare professionals, and the potential for innovation within clinical settings. Additionally, we reviewed interviews with healthcare professionals and hospital managers who are currently using the systems. These sources were useful for capturing the human dimension of AI integration, particularly with regard to the reshaping of medical roles, work practices and professional perceptions. Finally, we examined user-generated content, such as blogs and app reviews, to incorporate patient and practitioner feedback and better understand the real-world experience of these technologies.

The collected data were transcribed and analysed using qualitative content analysis (Strauss and Corbin, 1990). The analysis was structured around the three-dimensional framework developed by Bolton et al. (2018), which is particularly well-suited to the healthcare context as it captures the associated opportunities and challenges of transitioning from traditional to digitalised services (Bolton et al., 2018; Saviano et al., 2022, 2023). Specifically, the framework guided the examination of the digital, physical and social dimensions of the selected case studies. By analysing the intersection of these dimensions, we were able to position each case within one of the eight octants proposed by the model, thereby identifying the various configurations of AI–human interaction.

We conducted a qualitative content analysis to inductively extract a set of categories from the textual data, identifying key themes according to semantic criteria defined by the researchers (Krippendorff, 2004). We used the framework of Bolton et al. (2018) as an analytical lens to formulate guiding questions and interpret the data within each of the three dimensions.

To ensure the validity and reliability of the analysis, we adopted a double coding procedure. The two co-authors coded the same set of qualitative data independently and then compared and discussed their results (Patton, 2014). This iterative process involved multiple rounds of individual coding, followed by joint discussions to reconcile differences and reach a consensus on categorising AI-human interaction levels. The final categorisation was determined through a process of saturation, whereby no new categories emerged and a shared interpretation was reached. To further enhance reliability, we assessed agreement between coders, consistent with Krippendorff's (2004) recommendation to measure inter-coder reliability as an indicator of the extent to which conclusions drawn from imperfect data can be considered valid beyond chance. Ultimately, classifying each case within the Bolton et al. (2018) framework enabled us to generalise about the impact of AI on the medical professionals's expertise and innovation management practices across various human–AI interaction configurations.

The following sections present the four case studies analysed, along with the main findings that emerged from the qualitative content analysis. To enhance the study's practical relevance and support managerial interpretation, the names of the AI applications and systems examined are explicitly included.

The first case study focuses on ADA, an AI-powered digital health platform designed to support patients and healthcare professionals in the diagnostic process. Originally developed as a clinical decision support system in 2011, ADA was designed to improve diagnostic accuracy, particularly for rare diseases. The initial ADA product was tailored for clinicians and utilised a Bayesian probabilistic reasoning system to provide a ranked list of potential diagnoses based on patient symptoms, signs and medical findings entered into the system. The platform visually indicates the contribution of each data point to the final diagnostic output, mirroring the differential diagnosis process used in clinical medicine.

In 2016, ADA pivoted towards supporting patients directly by launching a symptom checker available as a browser-based and mobile app. This application enables users to input demographic information and medical history and engage in a dynamic conversation with an AI-powered chatbot. The chatbot adapts its questions based on user responses, minimising the number of questions asked to reduce user fatigue while maintaining diagnostic accuracy. Upon completing the assessment, users receive a personalised 'triage' recommendation outlining the urgency of required medical attention and suggesting next steps, ranging from self-care at home to emergency intervention. The tool also provides a list of possible causes that may explain the user's symptoms.

The reasoning engine underlying ADA is built upon a comprehensive and regularly updated medical knowledge base. This base is curated by physicians and informed by peer-reviewed literature, textbooks, case reports, epidemiological data and disease models. The system is designed to handle a broad spectrum of user profiles, including children, pregnant people, older adults and individuals with mental health concerns. A retrospective study conducted in 2019 evaluated the performance of ADA in diagnosing rare diseases. The results showed that ADA's top suggestion matched the confirmed diagnosis in 89% of cases. In more than half of these cases, the platform provided accurate diagnostic suggestions earlier than a conventional clinical diagnosis would have. In over one-third of cases, ADA could potentially have led to a correct diagnosis as early as the patient's first recorded clinical encounter.

ADA's evolution from a clinical decision support tool to a patient-facing digital health assistant reflects the growing trend of integrating AI to empower users and improve early access to medical insights. The platform offers scalable symptom assessment and valuable diagnostic support to clinicians and healthcare systems worldwide.

The second case study looks at Aidoc, a clinical AI platform developed by the eponymous company, which was founded in 2016. Aidoc positions itself as a pioneering force in medical AI, focusing on empowering healthcare teams to improve patient care by optimising treatment pathways and achieving better clinical and economic outcomes. The company's flagship platform integrates artificial intelligence across various clinical departments, particularly in diagnostic imaging. Initially developed to support radiologists in reducing turnaround times and improving diagnostic accuracy, the company's solutions have since expanded to cover a broader range of specialties, including cardiology, neurovascular care and vascular care. The platform is currently used in over 1,000 medical centres worldwide, including 7 of the top 10 hospitals in the USA and has been the subject of over 100 clinical studies. Aidoc's platform is designed to integrate seamlessly into existing hospital workflows, providing real-time analysis and interpretation of medical images. Its core strength lies in its ability to detect and flag acute anomalies through advanced algorithms, enabling radiologists and clinicians to respond more quickly and effectively. The system supports coordinated, connected patient care across departments and facilities, placing radiology at the centre of the patient journey.

In the context of oncology specifically, Aidoc has developed a dedicated AI solution for diagnosing and staging rectal cancer using ultrasound imaging. This application consists of three coordinated subsystems: (1) an acquisition and annotation module for capturing ultrasound images; (2) a continuous training engine for developing diagnostic models; and (3) a physician support system that assists with diagnosing and staging rectal tumours through automated risk clustering and prognosis. The platform also features a digital assistant that enhances clinical decision-making throughout the care pathway.

The third case concerns the Da Vinci system, a robotic surgical system developed by Intuitive Surgical. Introduced in the late 1990s, it was approved by the US FDA in 2000 for general laparoscopic procedures. It is now one of the most advanced technologies for minimally invasive surgery. As of 2024, over 7,500 Da Vinci systems had been installed in more than 70 countries and used in millions of surgical procedures worldwide.

Da Vinci enables surgeons to perform complex operations with enhanced precision, flexibility and control via a minimally invasive approach. The system comprises four robotic arms: three are equipped with surgical instruments (e.g., scalpels, scissors and electrocautery tools) and one carries a dual-lens endoscopic camera which provides a high-definition, 3D stereoscopic view of the operative field. The surgeon operates remotely from a dedicated console, which is physically separated from the operating table. Using hand controls and foot pedals, the surgeon manipulates the robotic arms while viewing a real-time 3D projection of the surgical site. The system's advanced interface supports a full range of motion, tremor filtration and the intuitive translation of the surgeon's movements into the micro-movements of the instruments. Additionally, the system performs over a million safety checks per second and provides the surgical team with real-time audio-visual feedback, ensuring maximum operational reliability. The Da Vinci system is equipped with features designed to improve team communication and training. A large touchscreen monitor enables surgical annotations during procedures and dual-console configurations with integrated virtual simulation provide comprehensive training for novice surgeons, facilitating a more effective learning process.

Da Vinci has numerous clinically proven benefits, including reduced blood loss and transfusion rates, fewer surgical complications, shorter hospital stays, decreased postoperative pain, smaller incisions with improved cosmetic outcomes and faster functional recovery, particularly in procedures such as prostatectomy, valve repair and gynaecological surgeries. Furthermore, by enabling system sharing between hospitals and facilitating the transfer of surgical knowledge, the Da Vinci system offers a sustainable economic model that maintains high clinical standards, even during the early stages of surgeon training.

The fourth case concerns Dave, the world's first real-time conversational AI oncology mentor. It was developed by Belong.Life, a leading digital health company. Designed with the mission of improving global access to quality cancer care, Dave provides personalised, round-the-clock support to cancer patients and their caregivers. Integrated into the Belong – Beating Cancer Together mobile platform, Dave functions as a virtual oncology assistant, offering continuous guidance, education and emotional support throughout the patient journey. The system has been trained using deep learning techniques on billions of data points derived from seven years of anonymised patient journeys shared on the Belong platform. These data include interactions between patients and physicians, as well as between patients themselves. Dave provides accurate, oncology-specific information based on a vast, medically validated knowledge base including reputable clinical sources such as scientific journals, clinical trials and standard treatment guidelines (e.g., NCCN). Its underlying architecture combines large language models (LLMs) with Belong. AI's proprietary conversational AI technology, enabling interactive, context-sensitive dialogue with users. The system retains long-term conversational memory, enabling continuity across multiple sessions. Dave is free to use via the app and provides patients with a user-friendly channel through which they can ask questions, receive reliable answers and access a peer support community. In addition to supporting patients directly, Belong. Life actively collects user feedback to refine and improve the system over time. Dave is a mature, operational AI application that has been widely adopted and medically validated. Recognitions include being selected as one of Newsweek's “World's Best Digital Health Companies” in 2024.

Building on the conceptual framework outlined above, this section presents the findings derived from the case studies' analysis. The findings demonstrate how diverse social, physical and digital intensity manifests differently in AI-based healthcare services, leading to distinct patterns of innovation. The application of Bolton et al.’s (2018) framework to the above-mentioned cases extends the framework's application to AI-specific contexts, establishing a taxonomy of AI integration levels. Furthermore, by identifying different AI configurations in healthcare (defined by human–AI interaction and AI integration levels), the analysis determines where human expertise remains crucial.

In the case of ADA technology, informed patient decisions increase the patient's confidence, according to the company. “After using Ada, 66% of patients are more certain of what care to seek, 40% report reduced anxiety, and 80% feel more prepared for their consultation, leading to better patient engagement and informed discussions” (ADA website). This AI-based solution offers a first touchpoint in the patient journey via CUF's patient-facing app. The users are already 120.000, with 84% of these completing the assessment. After the assessment, the patient is effectively guided to clinical services. The integration with CUF's EHR allows clinicians to view patients' clinical Handover reports, semi-automating the clinical history-taking process: “It increases efficiency, and clinicians really can save a lot of time” (Monteiro, M.S., CUF's Chief Medical Officer for Digital Transformation at ADA). ADA operates independently of medical practitioners, analysing symptoms and recommending interventions without direct human involvement. Its high digital density stems from its role as a purely digital triage system, integrating EHRs and telemedicine while processing large volumes of patient data. With no physical infrastructure required, it has low physical complexity. Limited human interaction results in low social complexity, as AI automates triage, enabling patients to act on recommendations independently. Thus, ADA falls into Bolton et al.'s (2018) octant of low physical complexity, high digital density and low social presence (Figure 2).

Other well-established cases of AI applications in healthcare include AI-powered decision support systems, such as AI applications in radiology, such as Aidoc. AI is widely used for detecting brain tumors, assessing neurodegenerative diseases and diagnosing conditions such as intracranial haemorrhage and strokes (Bhandari, 2024). Furthermore, recent studies in the medical field suggest that AI-assisted radiology can outperform traditional diagnostic methods in assessing tumor grades, facilitating early and precise treatment decisions (Pitarch et al., 2025). More specifically, Aidoc supports doctors in delivering healthcare services while preserving the social dimension of healthcare provision – the decision-making process remains doctor-led, with AI providing only supportive insights. This results in a medium-to-high social presence accompanied with a high digital density. Additionally, a well-equipped physical infrastructure is required to carry out the diagnostics. Thus, Aidoc falls into Bolton et al.'s (2018) octant characterised by high physical complexity, high digital density and medium-to-high social presence (Figure 2).

Chatbots, on the other hand, are commonly applied in healthcare for different purposes. According to the National Institutes of Health (NIH), AI-powered chatbots improve accessibility by providing 24/7 responses, reducing wait times for healthcare consultations. A key example is represented by Dave, designed to offer 24/7 support and mentoring to oncology patients. According to the co-founder and CEO, “Dave provides smart, personalized and accessible information instantaneously, which can greatly improve the quality of care and life for millions of patients worldwide” (Eliran Malki, Co-founder and CEO of Belong.Life). Dave provides constant patient interaction, even if it is virtual and not human-to-human and the human intervention is still needed for clinical medical directions, justifying a medium-to-high social complexity. Furthermore, given the amount of data managed and used by the technology and being based on machine learning (ML), large language models (LLM) and natural language processing (NLP), the digital complexity is medium. This solution also presents a low physical complexity. The chatbot, indeed, operates in a digital space, not needing a physical infrastructure, and there is no direct physical interaction with the patient. Thus, Dave falls into Bolton et al.'s (2018) octant characterised by low physical complexity, medium digital density and middle-to-high social presence (Figure 2).

In AI-driven robotic surgery, the Da Vinci system translates a surgeon's hand movements into precise robotic actions, improving accuracy and control. This surgical approach is minimally invasive, leveraging intelligent algorithm-based digital imaging, robotics, AI and advanced equipment to enhance precision and control. Both cases present a human-machine collaboration, through which AI enhances – but does not replace – human decision-making. Indeed, while the surgeon remains in control, AI improves the capabilities of the doctor, assuring precision, effectiveness and safety in the execution. Thus, the physical complexity of Da Vinci remains high, as its hospital-based service delivery retains the inherent physical complexity of this care provision (Saviano et al., 2022). On the other hand, the real-time integration of AI in the surgical process, optimising efficiency and accuracy, demonstrates a high digital density. While direct doctor–patient interaction is reduced, human oversight remains crucial and the surgical team still plays an essential role in decision-making and patient care, resulting in a medium-to-high social complexity. Thus, Da Vinci robot falls into Bolton et al.'s (2018) octant characterised by high physical complexity, high digital density and medium-to-high social presence (Figure 2).

Considering the service dimensions of each case study and the varying degrees of interaction between AI and medical procedures, it is possible to map the different cases to specific AI categories within the framework of Bolton et al. (2018).

The case of the ADA platform presents very high digital density, low social complexity and low physical complexity. In this case, AI takes over a series of medical procedures that include not only collecting patient data and updating medical history but also initially analysing the information and providing preliminary guidance to the patient. As a result, the patient becomes central in the interaction, actively engaging with and understanding the information provided by the AI. This process enhances the patient's awareness and comprehension, reduces anxiety and better prepares them for the next steps in their care pathway, which are suggested directly by the platform. Efficiency increases, allowing doctors to save significant time, which can then be dedicated to tasks that cannot be delegated to AI. This type of integration falls under AI-automation, which is characterised by very high digital complexity, low social complexity and low physical complexity.

Dave, on the other hand, exhibits low physical complexity, moderate AI integration and medium-to-high social complexity, representing a case where AI assists medical practitioners. It automates repetitive tasks and enhances access to information. Thanks to this application, patients can receive 24/7 support regarding lifestyle guidance and treatment adherence. However, Dave does not replace the physician; the responsibility for patient interaction and decision-making remains entirely with the doctor. Thus, this case aligns with AI-assisted healthcare, positioned in the upper section of the octant characterised by high physical complexity, medium digital complexity and high social presence.

Aidoc and Da Vinci robot, although the technologies are very different from each other, are both characterised by high digital complexity, medium-to-high social complexity and high physical complexity, represent cases in which the AI-based technology supports in real time the medical practitioners. These two healthcare AI-integrated solutions do not replace the role of the medical equipers rather enhance and augment their capabilities, assuring effectiveness, safety and accuracy. These cases align with AI-augmented healthcare, positioned in the octant of high digital complexity, medium-to-high social complexity and high physical complexity.

As a result, the Bolton et al. (2018) cube enables the identification of different levels of AI integration in medical practices, categorising them into three main configurations: AI-assisted, AI-augmented and AI-automated (Garbuio and Lin, 2019), which are summarised and represented in Table 1 and Figure 3.

As a result, it is possible to summarise the main service dimensions of each case study in Table 1, highlighting the complexity dimensions of each analysed AI healthcare innovation.

While Aidoc and the Da Vinci surgical system leverage technological advancements to enhance current medical practices while relying on human expertise, with AI engineers and technicians working alongside radiologists and surgeons to ensure the optimal, effective and safe use of AI-driven solutions, in the Dave case the impact of AI innovation lies in improving service delivery by expanding both the range and availability of healthcare services. To do so, the primary focus is on training and medical literacy for the medical practitioners, the patients and the caregivers. In the case of Aidoc, the system continuously searches for clinical insights, which, once discovered, activate the care pathway, making medical practitioners “faster and more efficient, allowing them to practice at the top of their license and improve patient care” (Aidoc CEO, Elad Walach, Aidoc website). Implementing such technology facilitates a cultural shift toward data-driven decision-making, ultimately empowering healthcare professionals, patients and caregivers. In the ADA case, doctors are empowered by the technology through a complete reconfiguration of the medical practice.

The case studies, thus, present different levels of complexity associated with each dimension and different general complexity, contributing to the literature and existing theories by leading to the identification of different levels of AI integration in healthcare, further defined in the discussion. The identification of different levels of AI integration highlights the transformative impact of AI innovation on medical practices.

The application of Bolton et al.’s framework allows for the identification of the social, physical and digital attributes of AI-based healthcare services by providing a multidimensional understanding of how innovation processes unfold and evolve under complexity conditions. The different configurations of AI integration in medical practice are derived by analysing the complexity of its service dimensions.

The transformative impact of AI innovation in healthcare manifests across the three service dimensions of physical, digital and social complexity, affecting the medical practices with different intensities. The level of complexity on these three dimensions and, therefore, the global complexity of the different AI-based healthcare innovations, leads to the definition of three main configurations of AI integration. This framework supports the classification of AI-enabled healthcare into three distinct configurations: AI-assisted, AI-augmented and AI-automated (Garbuio and Lin, 2019).

Each configuration shows a unique balance of human involvement and algorithmic capabilities. Table 2 and Figure 3 summarise and illustrate these differences, offering a structured way to understand how AI influences innovation paths in healthcare delivery.

The Bolton et al. (2018) cube provides a useful framework for linking the service dimensions' complexity to describe different configurations of AI integration in healthcare: (1) AI-assisted healthcare requires low physical complexity (human-driven tasks), medium-to-high digital density (relative to the other categories)and medium-to-high social presence; (2) AI-augmented healthcare involves high physical complexity (where AI and humans share tasks), high digital density and medium-to-high social presence; (3) AI-automated healthcare is characterised by low physical complexity (AI-driven processes), very high digital density and low social presence.

The configurations illustrate the various levels of interaction between humans and machines, emphasising the distinct complexity profiles of AI applications.

Mapping them within the Bolton et al. (2018) cube facilitates the identification of their core service attributes, understanding which are the AI-integration categories and providing insight into how each shapes innovation trajectories. While AI significantly enhances operational efficiency in healthcare, research has primarily focused on its technological advancements. However, to fully grasp AI's impact on human-driven innovation in healthcare management, it is crucial to determine where human expertise remains essential and irreplaceable in AI integration.

Furthermore, the categorisation of AI-based healthcare across its physical, digital and social dimensions also reveals how different AI configurations affect the roles of healthcare professionals.

The focus of this study shifts from concerns about a diminished human agency (den Hond and Moser, 2023; Lindebaum et al., 2024) to the view that AI can be an enabler of Human+, adding to human capabilities instead of replacing them. As AI adoption advances, interactions between humans and machines change, leading to a reshaping of clinical roles and organisational practices.

Each human–AI configuration has its own unique innovation dynamics. AI-assisted healthcare enhances efficiency without altering practitioners' roles (Pradhan et al., 2023). AI-augmented healthcare fosters collaboration, expanding medical expertise (Crigger et al., 2022), aligning with augmented intelligence, where AI enhances clinical decision-making (American Medical Association, n.d.). AI-automated healthcare replaces specific processes, reducing direct human involvement and prompting structural transformations (Park et al., 2019). Understanding these changes requires linking AI integration levels to innovation types and their effects on innovation management (Shestakov and Poliarush, 2019).

In AI-assisted healthcare, the technology supports human professionals without replacing them and fosters continuous improvement. The innovation in AI-assisted healthcare is incremental (Table 3). The primary focus is on enhancing the effectiveness and efficiency of healthcare services for different users, including patients and caregivers. AI-augmented healthcare is driven by sustaining innovation (Shestakov and Poliarush, 2019) (Table 3), through which companies develop better-performing products while maintaining existing market structures. AI collaborates with and supports medical teams, enhancing procedures, accuracy and efficiency. Implementing these technologies leads to redefining job roles, requiring healthcare professionals to be trained in managing new technologies and fostering interdisciplinary collaboration. These technologies improve precision, decision-making and operational efficiency without fundamentally altering the core structure of medical practice.

In AI-automated healthcare, innovation is radical/disruptive, leading to business model disruption (Schiavone et al., 2021). AI takes over certain functions entirely, leading to structural changes within healthcare organisations. Tasks and responsibilities are redistributed, allowing medical professionals to focus on areas where human expertise adds the most value. Medical practices are redefined, and innovation challenges traditional healthcare roles, potentially replacing specific jobs while reshaping the way healthcare is delivered. Furthermore, this shift signifies a redefinition of human expertise, emphasising the strategic, ethical and empathetic aspects of medical practice that AI cannot replicate.

Table 3 summarises the category of AI role, linking it with the respective types of innovation and the impacts generated by the innovation.

Analysing the impacts of AI integration in medical practices, AI-assisted healthcare demonstrates positive effects on workflows by accelerating processes and increasing the efficiency of medical services (Ali Mohamad et al., 2023). Additionally, AI-assisted solutions enhance accessibility by allowing patients to access medical services – such as information and appointments booking – anytime and from anywhere, 24/7. This reduces the burden on medical staff by automating repetitive tasks where human expertise adds minimal value, while ensuring that professionals remain central in patients' interactions. The adoption of AI-assisted healthcare solutions requires AI literacy for both patients and medical professionals, enabling them to interpret AI-generated data and foster a data-driven culture. In this AI configuration, human expertise remains crucial in building and maintaining patient trust. While AI supports repetitive tasks, doctors and other medical professionals continue to play a leading role in decision-making.

AI-augmented healthcare, on the other hand, does not alter the physical complexity of medical procedures but enhances their effectiveness, safety and efficiency. In this case, human expertise is strengthened through AI adoption, as medical skills and capabilities are upgraded through collaboration with technology and multidisciplinary teams. AI augmentation leads to a redefinition of job roles, as well as process and business model innovation (Schiavone et al., 2021). The human–AI collaboration evolves, shifting the role of medical professionals toward leadership and management of this interaction, ensuring optimal integration of AI within healthcare practices.

When considering AI automation, the implications for organisational structure become more significant. In this configuration, AI takes over tasks previously performed by doctors or hospital staff, leading to a radical shift in workflows, resource allocation and governance. Medical professionals are no longer responsible for certain activities, as these are fully delegated to AI, even in the decision-making process. This transformation redefines the medical profession, shifting human involvement toward strategic oversight, ethical considerations and empathetic patient care.

Understanding where human expertise remains central is key to identifying the impact of AI adoption in healthcare and leveraging it to facilitate the transition from conventional to AI-integrated medical practices (Table 4).

From an innovation management perspective, these three AI configurations represent distinct strategic and operational challenges. AI-assisted healthcare supports incremental innovation, optimising existing processes without fundamentally changing the organisational structure. Thus, innovation management aims to incorporate AI tools to enhance efficiency while preserving human control, ensuring adoption through digital training and fostering a data-driven culture.

AI-augmented healthcare drives sustaining innovation, leading to the necessity to reconfigure workflows, foster interdisciplinary collaboration and invest in upskilling personnel to enable seamless human–AI synergy. Supporting organisational learning and change management processes is necessary for innovation management, while also fostering a culture that embraces technological augmentation.

AI-automated healthcare triggers radical and disruptive innovation, necessitating a comprehensive rethinking of governance models, job design and value creation logic. To ensure patient-centered values remain preserved, innovation management strategies must lead structural transformation, address ethical concerns, balance human–machine responsibilities and lead structural transformation. Innovation management must navigate the changing boundaries between human and machine roles and tailor strategies to align AI capabilities with organisational goals and healthcare values across all configurations. Designing adaptive, future-oriented innovation pathways in AI-integrated healthcare systems requires understanding these differentiated impacts.

The expanding role of AI, proper of society 5.0 (Troisi et al., 2024; Gravili et al., 2023; Bartoloni et al., 2022), is advancing rapidly and cannot be reversed, making it essential to manage human–AI interaction across different levels of AI integration. This interaction, which varies depending on the degree of AI integration in medical practices, drives the necessary evolution of medical professionals. This evolution is built upon the irreplaceable aspects of human expertise and the potential for innovation unlocked by AI adoption. Since AI is here to stay, human expertise must evolve to remain relevant. This leads to the emergence of Human+, an enhanced form of expertise that combines human intelligence with AI capabilities. The rise of Human + medical practitioners represents an evolutionary response to AI integration in healthcare. Rather than replacing professionals, AI acts as a catalyst, reshaping their roles and expanding their skill sets.

This study provides new insights into human–AI interaction in healthcare. Identifying and understanding different AI solutions and configurations is essential for determining where human expertise remains crucial and how AI adoption impacts innovation management. The multiple case studies illustrate the spectrum of AI integration, highlighting how AI transforms healthcare service provision and redefines the role of medical professionals.

The findings and discussion offer guidance on managing human–AI interaction and the necessary evolution toward Human + by leveraging human expertise within innovation management strategies. These strategies vary based on the degree of AI integration. In AI-assisted healthcare, investments should focus on AI training programs for patients, caregivers and medical professionals, alongside clinical decision-support governance models. This ensures AI enhances, rather than disrupts, human-led decision-making. Considering AI-augmented, healthcare organisations should redesign training, leadership strategies and innovation policies to promote effective collaboration between professionals and AI rather than resistance. In AI-automated healthcare, the organisational structures and governance models must be redefined to ensure proper task reallocation between professionals and AI, shifting the role of medical practitioners toward strategic oversight, ethics and empathetic care.

The adoption of AI necessitates a shift toward AI literacy and data-driven decision-making, the expansion of medical expertise, the development of multidisciplinary and hybrid skill sets and the reorientation of professionals toward leadership and empathy. AI blurs the boundaries between physical, digital and social realms, affecting social presence and reshaping human–AI interaction. This perspective moves beyond the traditional dichotomy of automation vs augmentation, recognising AI-assisted, AI-augmented and AI-automated healthcare as distinct yet interconnected configurations, each involving human expertise in different ways. Ultimately, human–AI interaction places medical professionals on an evolutionary path in which, through AI, they transition toward Human+, where human intelligence and AI capabilities merge to enhance expertise, made of knowledge, skills and capabilities, to redefine the future of healthcare.

This study has theoretical and practical implications. From a theoretical perspective, by demonstrating the role of AI integration – assisted, augmented and automated intelligence – in generating incremental, sustainable and potentially radical forms of innovation in healthcare service delivery, it advances the literature on AI as a driver of innovation capabilities. Furthermore, it enhances research on healthcare service innovation by implementing the Bolton et al. (2018) framework in AI-rich environments. Finally, it enhances the new trend of human–AI interactions by conceptualising the evolution of medical professionals towards a Human+, where human expertise is not replaced but rather reconfigured and expanded through AI-enabled capabilities.

From a practical perspective, this study provides guidance for healthcare organisations on how different AI-integration configurations shape capability development, role redefinition and governance. In AI-assisted healthcare, aligned with incremental innovation, organisations should focus on AI literacy programs for professionals, patients and caregivers to facilitate seamless AI adoption. Basic data literacy and familiarity with AI-enabled tools should be given priority in capabilities development, while governance efforts should ensure safe integration into existing workflows without compromising professional boundaries. In AI-augmented healthcare, aligned with sustaining innovation, AI expands human capabilities and reshapes clinical decision-making. Healthcare organisations should prioritise redesigning medical practitioner training programs to foster multidisciplinarity and leadership skills, ensuring AI enhances rather than replaces human expertise while maintaining autonomy. In AI-automated healthcare, aligned with radical and disruptive innovation, AI increasingly performs independent tasks that require structural and governance changes, the redefinition of roles and responsibilities and the creation of new accountability structures. Ethical oversight, relational leadership and systems-level coordination must be included in capability development to manage the redistribution of tasks between humans and AI.

To answer the RQ, these configurations generate different but converging pathways toward Human + medical professionals, requiring new skills in data literacy and interpretation, orchestration of human–AI collaboration, ethical oversight, relational leadership and innovation management in healthcare. To enhance professional judgment and support the continuous improvement of care delivery, healthcare innovation management should be intentionally directed toward nurturing these capabilities.

This study has certain limitations that pave the way for future research agenda. One key limitation is that the findings are based on only four case studies, representing a small selection of successful AI adoption cases in healthcare. The limited number of cases, the qualitative research approach and the restricted dataset may not fully capture the entire spectrum of AI integration in healthcare. Additionally, while the analysis focuses on the service dimension of healthcare, the role of human expertise and the implications for innovation management, it does not incorporate the perspectives of different actors, such as patients, caregivers, medical practitioners and healthcare organisations' management, which could provide valuable additional insights. These limits can be addressed by expanding the number of cases and data collection, conducting in-depth interviews and organising focus groups to compare different technologies. This comparison will highlight the varying levels of human–AI interaction in each technology's configuration and the implications for innovation management in the evolutionary path toward Human+.

Due to the explorative nature of this study, future research should expand the scope of case studies and incorporate quantitative methods to measure the impact of AI adoption in healthcare more precisely. This work contributes to the existing body of literature, providing a service science valuable framework for analysing AI implementations in healthcare. Furthermore, the classification of AI–integration, together with the identification of human–AI integration levels and their innovation management impacts results in a useful and effective perspective to analyse AI integration into the medical practice and the evolutionary path toward Human+. Further research should examine a broader range of AI healthcare technologies, collecting primary data at different stages – before, during and after implementation – to better understand the strategies employed and their implications for innovation management. Additionally, further studies focusing on the evolving role of medical practitioners in innovation processes could contribute to a deeper understanding of AI in healthcare and the necessary transformation of professionals into Human + medical practitioners.

Statement: During the preparation of this work, the authors used ChatGPT and Grammarly in order to proofread the text. After using this tool, the authors reviewed and edited the content as needed and took full responsibility for the content of the published article.