What can I do with you, machine?

The article states that an increasing number of entrepreneurs entering the healthcare space are harnessing old and new technologies in the solutions they take to the marketplace. One of the most active sectors is AI, being applied to clinical, operational, or financial solutions in health settings. In this article, the authors generally describe AI as computer systems that perform tasks requiring human-like intelligence.

Current research has classified healthcare startups on the basis of the type of the problem addressed (e.g., whether they provide telemedicine services, virtual assistance, or image recognition). However, at the start of the article, the authors do not know whether startups that solve different problems share common business models. Nor do we know what value is created for stakeholders (e.g., clinicians and patients).

Hence, in this study, they plan on answering this question by examining some of the most promising AI-driven healthcare startups and developing a framework for studying and tracking their emerging business models.

The authors point out that any investigation into the implications of technological change begins with a characterization of the technology’s attributes and the potential business model that can effectively take it to the market. To start, they reviewed the deployment of AI in the healthcare sector using a data set drawn from 30 healthcare startups around the world.

The driving question of the articled investigation was: "How do entrepreneurs in healthtech develop business opportunities and capture these opportunities with innovative and technologically driven business models?"  In the order from start to end:

• Start: Review several academic and practitioner-oriented studies into healthtech (healthcare technology) in general and closely observe up to three AI-driven healthcare startups. To describe these three distinct startups:

  1. Used AI in the creation of a marketplace connecting providers and patients.

  2. Was a company operating in the digital health space.

  3. Was developing a smart pillow solution for the detection of sleeping disorders.

• Collaborative study: Studied literature on AI and discussed it with researchers in both business and engineering to entrepreneurs. Afterwards, authors identified startups in the healthcare sector to study.

• Match information: Matched/compared them to data from other types of media after collecting information about funding received by each startup.

• Interview high-level stakeholders: Interviewed several stakeholders (i.e., executives, CEOs, venture capitalists) currently utilizing AI technologies or managing AI-driven companies to finalize understanding of models.

• Presentation: Found and presented current iteration of understanding of AI healthcare sector and business models to a new set of stakeholders (this time including only entrepreneurs and executives in the field) to gauge feedback on authors' research.

In industries such as healthcare, human intelligence is both invaluable and increasingly in high demand. The debut of innovative, AI-powered technologies has been lowering costs, hastening drug discovery, and improving health outcomes.

More and more, the potential of AI to revolutionize the industry is catching the attention of key players in both healthcare and venture capital, with increasing funding allocated to the sector in recent years.

In order to better advance our understanding of such phenomena, however, we must further break down these disruptive new technologies into different categories:

1. Assisted Intelligence: helps improve what the business is already doing by amplifying the value of current activities. Often involves clearly defined, rule based, and repetitive tasks (i.e., medical image classification to improve accuracy over conventional processing techniques).

2. Augmented Intelligence: an emerging technology that provides new capabilities. Differs from assisted intelligence in that it alters the nature of an activity, which as a consequence requires changes in the business model. Stated to play a critical role in precision medicine — tailoring of medical treatment to target specific needs of individuals based on their characteristics (e.g., genetic makeup)

3. Autonomous Intelligence: the advanced stage of AI that is currently being developed. Acts on its own and chooses its action on the basis of business goals. Currently not in widespread use, but recently, has been proposed in a doctorless hospital application to both advance AI technology and build AI-human trusts in that the former acts under the latter's best interests.

To better understand how the types and subtypes of AI above are changing the healthcare technology landscape, the authors classified a selection of a set of 30 startups from private market intelligence platform CB Insights.

Among them, none currently offer any form of autonomous intelligence as this class of AI is not yet widely available in the healthcare technology market. Instead, forms of assisted and augmented AI are disrupting the current health-tech landscape through the use of limited memory, machine learning-based platforms with a local in-memory database to inform the system’s real-time decision making.

Nevertheless, these findings sufficiently told the authors that rapid advancement in autonomous intelligence technology is visible, although it will take a longer while for it to become more widespread within the overall AI sector.

For a solution to thrive in the marketplace, it needs to begin with a clear definition of value that is to be created for a particular user. The question of value relates to the question of who is the user that the solution aims to address.

In the article's proposed case of Your.MD aka Healthily, an AI chatbot application that helps patients find the most relevant health information, the namesake company's CEO Matteo Berlucchi highlighted that he did not go through the deliberate process of opportunity identification to come up with the idea behind Your.MD. That the idea had already envisioned a clear user value aimed at addressing a critical problem.

To quote his own words: "With mobile phones becoming ubiquitous and providing you with computing power on your hands and an easy connection to centralized computing power, there must be a way to get people the health information they need for free when they need it… So mobile phones plus health equals something useful for a lot of people. The original idea was just to give information to people."

Berlucchi's original idea did not include a business model that outlined how taking information to the masses would work. Only after the original value creation idea was formed did the typical methodological approach of market analysis begin. One insight was realizing that the healthcare industry is far behind other industries when it comes to digitization and transfer of control to the end user.

In many cases, using examples from other industries to solve a problem is applied consciously or unconsciously by entrepreneurs in discovering new approaches to business model innovation, as well as strategies that can then be adapted to their specific industry. It is at this point that Berlucchi started to think about solving the problem in a very patient-centric way.

Health-related searches are among the most common searches in Google and tend to be associated with highly paid advertisements. However, search engines do not provide the most accurate information and are not a curated marketplace of healthcare information. For this reason, it is important to distinctively note that the provision of information is only one step toward an AI-enriched use of information in healthcare.

Another important consideration at the very start of building a new venture is identifying who the users are. In many instances, there are a number of individuals who benefit from an AI-driven solution (really, any topic will do) in different ways. It is thus critical to understand who the primary and secondary users are as well as who will pay for the solution.

The question of who pays is particularly important in healthcare, where insurance providers or national health systems ultimately bear some of the costs. This has already happened in relation to medical devices that are enabled by IoT technologies.

One example is a healthcare provider that detects potential issues in a prosthetic joint using data sensors to summarize the force distribution and pressure patterns. This helps deliver value to the patient by promptly alerting them to see a medical professional, as well as to the provider since unnecessary costs due to remedial treatment or prolonged recovery are avoided.

When several stakeholders concurrently benefit from the solution, monetization becomes an interesting question — are all parties involved liable to pay or should one side be subsidized (paid), as often happens in platform business models such as gaming systems?

Say you have founded a healthtech startup that collaborates with a research center to develop a noninvasive smart pillow to help diagnose and monitor sleeping conditions. You have identified broad target markets, including end users for at-home use, sleep clinics, and so on. You can see enormous value in your innovation to end users as poor sleep is a precursor to heart disease and dementia… which can lead to death.

Your task here is to decide who you are really creating value for and what that value is. Fundamentally, you need to create hypotheses to test these alternative futures, which allow you to accordingly tweak your technology, depending on the specific value-user combination you are exploring. The data is a crucial aspect in this regard.

After much research and social interactions with importnat individuals in AI-centric healthcare, the authors think that AI is bringing a wealth of value to healthcare across various stakeholders and different aspects of the healthcare journey, from patients to pharmaceutical companies. Listing the benefits in terms of stakeholder scales:

• Patients: AI can help by giving them access to personalized, validated, and actionable information and data.

• Healthcare diagnostics: AI can help by providing a faster and more precise detection of small variations within patients' health data and comparing such variations across similar patients.

• Community-wise: AI also allows identification of outbreaks and pandemics much earlier than current methods, which helps to contain their spread. Future benefits could also include the selection of patients for clinical trials.

After giving the benefits of AI technology above, the authors propose the following classification of value drivers focusing on end users (i.e., patients and their families), as well as organizational goals and priorities. Based on the original classification developed by multinational technology company IBM, while IBM their original classification focused on healthcare service providers’ goals with respect to the use of analytics, the authors further added goals related to value creation for patients. 

In the proposed classification of value creation, the authors suggest that healthcare startups clearly identify the target user — patient and healthcare provider focus — as well as the area of value they aim to create. Value creation with patient focus includes:

• Patient healthcare accessibility, disease predisposition, and lifestyle management.

• Clinical effectiveness and patient outcome/satisfaction.

• Patient safety.

Value creation for healthcare providers and payers may focus on:

• Operational effectiveness and efficiency.

• Financial and administrative performance.

Another analysis done by the authors highlights seven business model archetypes in AI-driven healthcare startups that can be useful to future entrepreneurs and managers. These archetypes differed in their target user, area of value creation, and value capture mechanism, the last one which they based off of to further classify all seven archetypes into two groups: information providers and connectors.

Additionally, they also briefly highlighted three delivery models that are of particular interest in the healthtech space dominated by AI. They exact details can be read in the graphs below.

In the information provider archetype are sub-archetypes that describe how they provide value, to whom do they prioritize in healthcare, and their general aims interpreted by the authors:

• Specialized diagnostic: Traditional (non-platform) business model that transforms specific inputs in order to solve a problem for a user further down the value chain. Provides analysis focused on one specific type of data (e.g., images). Value is in offering more accurate insights to clinicians and payers through AI technology to assist them in efficiently delivering high-quality services to patients. Is more commonly deployed to support than prevent and treat in case of healthcare.

  • Example: Healthtech startup Imagen Technologies focuses on applying a state-of-the-art AI system to medical image analysis in order to detect pathologies and make early disease identification within medical images. It aims to deliver value to healthcare providers by reducing diagnostic errors and improving patient outcomes, the company's long-term goal being to identify the next set of breakthroughs at the intersection of AI and medicine to transform early disease identification and management.

• Aggregator: Focuses on consolidating data from disparate sources (e.g., electronic health records or EHRs, images, even social media), analyzing them, and offering insights to a variety of stakeholders. Value creation lies in facilitating the clinical effectiveness for clinicians by providing more insightful information to help them make more accurate decisions in diagnosis and treatment. Also involves enhancing operational effectiveness for healthcare providers and payers by providing more accurate information on things such as risk profiles of the patient population.

  • Example: San Francisco-based pioneer in medical deep learning Enlitic leverages AI to analyze a wide variety of healthcare data sources (e.g., readiology iamges, lab tests, and clinical cases) to create a clinical decision support product that improves clinicians' speed and accuracy of their diagnosis, highlights patients at risk of a specific disease based on suspicious findings, and accelerates pharmaceutical research and drug trials. The company creates value for healthcare providers through its data aggregation and proprietary deep learning algorithm that allows clinicians to fully utilize the richness of data from various sources and make fast and accurate decisions for their diagnosis and treatment.

• Personal health companion: Applies AI technology to assist patients in conducting a preliminary (preparatory) diagnosis on their own. Works on reducing information asymmetry — where one party in a transaction has more or better information than the other — between clinicians and patients by providing patients with evidence-based and customizable explanations for a variety of questions. Aims to provide an accurate and quick diagnosis than implement preventive or curative measures.

  • Example: Your.MD is pursuing this archetype. Some challenge lies in identifying a way to monetize that goes beyond simply asking customers to pay for answers to their questions. One avenue is to connect them directly with a doctor for a consultation through a booking system.

• Smart prevention companion: Applies AI technology (e.g., deep learning) to give patients a nudge/attempt to change their behavioral patterns based on the analysis of big data. Its emphasis is on prevention rather than diagnosis or remedial intervention. By virtue of this archetype, patients are empowered to become more responsible for their health and wellness. Aims to create value through increasing patient healthcare accessibility and clinical effectiveness.

  • Example: Intendu helps people with brain impairment train their cognitive skills at home. Through the use of real-life, interactive scenarios and a motion-controlled video camera, it helps patients train their thinking, memory, and attention skills. The system is adaptive and can personalize the training program in real time to a patient’s performance, biofeedback, and rehabilitation needs, then nudge patients to change their behavioral patterns to achieve better outcomes in their training program.

As for details about the connector's sub-archetypes:

• Promotor: Offers patients a preliminary diagnosis and then matches them with the relevant healthcare practitioner, with the application of AI technology mainly in the diagnosis phase. Aims to create value for patients in diagnosis and in directing them to appropriate practitioners, thereby enhancing accessibility to information, treatment, and doctors, as well as increasing efficiency in targeted advertising for healthcare providers. Similar value delivery in diagnosis phase is similar to personal health companion but further adds value by timely and effective channeling of patients to practitioners.

  • Example: Your.MD CEO Matteo Berlucchi noticed specialized clinics' demand for paid advertising to target specific customers seeking clinical consultations, so OneStop Health was born as a curated network of trustworthy healthcare providers. It was the first one-stop shop — a business that provides a variety of products or services in one place — 

• Discriminator: Platform composed of online communities. Creates value for both patients and clinicians by allowing individual patients and their families to share their experiences with the healthcare system as well as allowing clinicians to share their medical insights on therapies. Aims to apply deep learning AI technology to increase patients’ access to healthcare information and enhance clinical effectiveness, through the use of big data from sources of health data.

  • Example: Patientslikeme describes itself as a clinical research free website platform that can provide real-world, real-time insight into thousands of diseases and conditions. Patients can share their health data on it to track their progress, help others, and offer data to researchers for advancements in the medical field. The website hence creates value for patients by providing them with information on how to live with specific conditions and improve treatment outcomes, at the same time creating value for clinicians by allowing them to learn directly from patients’ real-world experience in real time on thousands of diseases and conditions.

• Trusted brokers: Focuses on utilizing AI technology to diagnose, manage, or treat patients through the aggregation of several forms of data (e.g., text, images) from IoT-enabled devices (i.e., sensors), social media, and other sources. One application of this sub-archetype is remote monitoring of patients (e.g., aging individuals) by aggregating (grouping) relevant data from a disparate set of IoT-enabled devices, which will alert patients and clinicians to any sign of health deterioration and consequent need for prompt intervention. Ultimately creates value for patients and their families by equipping them with real-time information on their health condition and allowing clinicians to reactively generate personalized treatment plans, further increasing healthcare providers' operational effectiveness and efficiency.

  • Example: Real-world data (RWD) company Prognos’s healthtech startup, DxCloud, adopts this sub-archetype. It provides an end-to-end solution with historical and ongoing clinical insights for healthcare providers to interpret and identify patients' conditions for better management and to optimize risk adjustment. Ongoing monitoring and early detection of health symptoms are important for better care and reduced medical costs for patients, as well as for increasing operational speed and treatment adaptability for clinicians.

Finally, onto the sub-models of the delivery model:

• Platform: Market model that creates value by facilitating exchanges between two or more interdependent parties (e.g., demand and supply). Not to be confused with the piece of technology of the same name. Facilitates exchanges by reducing transaction costs and/or enabling externalized innovation. Is able to scale in ways that traditional businesses cannot with advent of connected technology and network effects (see below). Creates value by bringing together the two sides of a market who otherwise would use alternative places to meet, often at higher costs (e.g., eBay and Uber).

  • Network effect: Way in which a platform's customer base impacts the value of the platform itself. The more people engage with a platform, the more useful and valuable it becomes. There are two types of network effects that can be enabled:

    • Direct network effect: Happens when the greater number of members (e.g., patients) on one side of the market leads to a direct increase in value (e.g., from stockpiling incentive to connect with doctors) for other members (e.g., doctors).

    • Indirect network effect: Occurs when the greater number of members on one side of the market attracts an increased number of members on the other side of the market (e.g., more patients joining the platform means greater chance to attract more doctors).

  • Example: Healthtech company Wellframe adopts this model. Same as other SaaS (software as a service) company, it uses a remote monitoring system to collect data but fundamentally relies on recruiting both patients and care teams to improve care management through a tech-enabled, data-driven, and patient-centered approach. This model can increase productivity among care managers, improve engagement and retention in programs among members, and ultimately maximize medical cost savings.

• Software as a service (SaaS): Way of delivering applications over the Internet as a service. Instead of installing and maintaining software on the client device, startups of this model license software online on a subscription basis, freeing clients from complex software and hardware management tasks. Model provider manages access to application, including security, privacy availability, and performance (e.g., speed). Aims to create value for healthcare service providers by increasing their operational efficiency, reducing costs, and increasing value-based revenue. Reliant on monetization through subscription services, including purchase of additional or premium features/services.

  • Example: Enlitic utilizes machine learning, NLP, and other AI applications in a service platform to streamline the process of data-driven patient population management, risk assessment, and monitoring. The focus on continuous risk assessment and patient population management reflects a shift within the healthcare market, across providers and payers, toward preventive care and thus an increase in demand for intelligent patient data analytics and risk management.

• Platform as a service (PaaS): Form of cloud computing platform that allows firms to develop (i.e., customize) and manage applications and services without the cost and complexity of buying and managing software licenses, infrastructure, and development tools and resources. In short, it provides a framework for firms to manage and customize applications while having the platform provider manage the storage, server, and networking. Compared to SaaS, this model gives startups more flexibility and control in customization and management of the application.

  • Example: Cloud-based digital healthcare startup Datica offers compliance tools for healthcare providers to integrate their EHR and other records (e.g., radiology information systems or RIS). The Datica platform provides a secure, compliant service managed by Kubernetes — an open-source container orchestration platform — that integrates with cloud services. It delivers value to healthcare providers by reducing operating costs and by boosting operational and administrative efficiency.

The authors' discussion of AI's application in healthcare startups offers several implications for both management theory and practice. While the application is the focus, they acknowledge the importance of the availability of data and its structure and quality as a key ingredient for applying AI into healthcare; despite the topic being outside their scope for their article on application of AI for decision making. The authors' focus, as of the section, is on how data facilitates and enables decision making for patients and clinicians (i.e., stakeholders).

The authors' investigation raised three critical questions, regarding the implications for theory development, for scholars working at the intersection of business and technology:

1. Entrepreneurial top management teams: Study is stated to contribute to the top management team literature by providing a taxonomy/classification of different business model archetypes emerging in healthcare entrepreneurship.

   • First expectations: Some team compositions are better suited to the development of some archetypes than others, not limited to in terms of their ability to raise funding.

   • After preliminary analysis: Teams raising the greatest amount consist of individuals with at most two different backgrounds (e.g., business and IT). As number of backgrounds in team composition grows, amount of funding appears to decrease (controlling for other company characteristics). May be due to conflicts among funding team or loss of focus in terms of new venture’s mission and vision. An avenue for future research.

2. Entrepreneurial opportunity creation: A wealth of research has gone into identifying preconditions of opportunity recognition, including the importance of prior knowledge and external conditions, as well as the thought processes that transform knowledge and observations of the environment into opportunities and the impetus to act upon them.

   • Predictions: Authors predict that AI will transform research in opportunity creation as it opens up the possibility to identify opportunities more easily and to a greater number of individuals.

   • Argument: Whereas in the past doctors have had access to some data points on patients, AI enhances data accessibility by providing a stream of information available to relevant decision makers more often or in real time and also more structured because of greater data points from health records and previous conditions.

   • Availability: These will also be available not only to doctors and nurses, but also to stakeholders from other medical fields such as pharmacists and administrators, with implications for the effectiveness of drugs and therapies.

3. Open innovation: Openness of data is critical for the application of AI in healthcare as it allows extraction of more accurate and insightful information from AI technology. However, models of data openness in healthcare are still in its emerging stage.

   • Security concerns: Healthcare data is traditionally restricted to relevant parties (e.g., patients and clinicians) due to its confidential nature. Patients have privacy and security concerns, making the enrollment in electronic health records a concern for most governments.

   • Lack of trust: Lack of reliability and standards of data across multiple sources poses a challenge in the wide acceptance of data openness in healthcare.

   • Data management: The concept of open data in healthcare also suffers from the difference in the level of sophistication in the way data are collected, organized, and managed across stakeholders (e.g., data from health medical records), which make a seamless integration of healthcare data across sources much more difficult.

   • Stakeholder wants: The motivations of stakeholders need to be considered in order to promote data openness in healthcare.

Further questions arise regarding management of these models and incentives to be provided in order to maintain stakeholder engagement with the platform, both for stakeholders within organizational silos and stakeholders from outside. The authors believe that this area of research is very much connected to questions of emerging models of partnerships between companies providing models and algorithms (e.g., IBM Watson) and startups providing industry solutions (like the examples in the article). They believe that future research can examine how data availability may affect the survival of resource-constrained startups.

Onto the implications for practice, the authors' research pointed to three major recommendations for entrepreneurs and managers. Their ultimate question is not whether AI in healthcare will create profitable business models or benefit a business incorporating it into existing operations — rather, it is how to best start and manage a transition from 'analog' to AI-powered digital solutions.

• How to get started: The first two key decisions in getting started are to identify the healthcare issue a startup attempts to address and strategically evaluate how the AI application may enable the solution in an existing business by analyzing the landscape of applications (see Table 2).

  • Know your limits: For new businesses, look at 'low-hanging fruit' such as improving operational effectiveness and solving inefficiencies. It is exciting to solve big problems, such as optimizing patient treatments, but they also carry important baggage (i.e., regulatory hurdles to data cleaning).

  • Progressive approach: Take robots, for example. have been shown to provide substantial benefits in delivering supplies to different parts of hospitals or in assisting elderly patients. These provide a learning opportunity, as well as a 'quick win', that can accelerate development of solutions aimed at solving bigger problems.

  • Cannot improve upon nothing: Pointing out the authors' discussion with Your.MD, getting started may be more important than focusing on monetization, which usually comes later.

• How to select AI technology development strategy: Startups need to evaluate whether the most appropriate approach to their value creation goal is to build solutions by leveraging existing AI platforms or develop their own AI technology.

  • Critical questions: The authors proposed four questions that ultimately decide which technology path to follow:

    • 1st question: What stage is the company at? If a company is at the prototyping stage and trying to understand how AI can solve its problem and whether there might be a market for its solution, there is little reason to spend extensive resources in developing its own AI.

      • Quick solution: Instead, there is much better value in quickly prototyping a solution and testing it with potential users to gain valuable feedback for iteration and improvement.

      • Understanding your position: Hence, at the prototyping stage, the emphasis on value creation is understanding the time needed to identify the problem to solve and how value is created to potential users.

    • 2nd question: Relates to technology and whether the effectiveness of insights from AI relies more on the available data or on the algorithm that can be built. Solving a very narrow problem may require an incredible amount of data to generate an effective solution. An effective pilot solution may be developed in a very short period of time, such as during a hackathon.

      • Pick a target: A case in point is Arsenal Health, originally called Smart Scheduling. Smart Scheduling was started in 2012 by a team who met at the Massachusetts Institute of Technology hackathon. The startup utilized more than 700 variables and machine learning to investigate the question, "What if you could use data science to determine which patients are likely to show up and which ones will be no-shows and manage office appointments around those tendencies?", and it quickly achieved accuracy of over 70%.

      • New breed of problem solving: The above is an example of a new breed of healthcare startups that leverage big data to offer a solution to a specific area of customer need, rather than attempting to solve big issues in healthcare such as population health, which is better suited to larger organizations (e.g., IBM Watson). In doing so, this new breed works to streamline processes and increase efficiencies.

    • 3rd question: What does the company sell? There is little value in investing resources in developing an algorithm from scratch if a clinical trial company intends to use the insights from AI to better select patients and profit by selling a report with a key actionable insight.

      • Less means more: Instead, to sell a medical AI technology to facilitate faster and more efficient diagnosis (e.g., analyzing highly specific digital images), it makes sense to allocate resources to developing an ad hoc algorithm — a specifically developed, non-general program.

    • 4th question: Involves a company’s available resources. Developing AI solutions can be extremely time consuming and resource intensive. Again, if AI is only part of the solution, then there is little reason to devote extensive resources to developing an algorithm from scratch. Given an undersupply of data scientists, most startups should consider standing on the shoulders of the giants.

• Governance and decision-making responsibility: Another contribution to practice relates to responsibility in terms of decision making for AI-driven healthcare startups that are on the augmented or autonomous intelligence spectrum.

  • Who is to blame: In the case of decision-making errors, it is foresightful to preemptively consider if an error is the responsibility of the clinician ordering the test and allocating the task to an AI solution. Is it the fault of the clinician looking at the results (who may be the same person)? The machine? Or of the startups that created the AI solution?

  • Authors' stance: Consider an image analysis tool that looks at X-Rays. If a radiologist looks at an X-ray and misses evidence of cancer, the fault lies with the individual, but if a computer looks at it and misses evidence, is it the fault of the computer, the physician who ordered the test, or the company that built the computer? This will ultimately be worked out in a courtroom, as with accidents involving self-driving cars, and blame will likely be apportioned to the startups providing the AI solution.

  • Counterpoint: However, a startup’s counterpoint may be in highlighting the number of positive cases missed by a radiologist versus the number of positive cases missed by the AI solution, just as an autonomous car has fewer accidents than a human driver. These are complex legal areas that will play out in multiple industries. Recent pronouncements by the U.S. Food and Drug Administration in permitting AI to make medical decisions on its own (e.g., interpreting medical imagery for detecting eye disease diabetic retinopathy) further emphasize the importance of the governance and decision-making responsibility that are being and will be discussed in court.

• Two business model archetypes in the same entity: Another critical question raised by startups such as Your.MD is should the startup play with two business models at once or should it 'spin-out' (reject?) one of these models? No right or wrong answer has been identified in the academic or practitioner literature — not even for larger corporations.

  • Narrow and focused: For startups, the common advice is to focus on, develop, and improve one business model at a time. Ultimately, it is a question of balance between integration and separation and remains an open question in the AI context.

  • Feeling your wallet size: The question of duality of business models is also strictly related to funding. If a startup is attacking a big problem (e.g., population health), it requires a variety of data, so funding constraints may impede collection and analysis of sufficient data for meaningful insights. In that case, a more effective path to follow might be focusing on specific data (e.g., image recognition) and taking advantage of partnerships with hospitals and other providers.

• Get stakeholders to parallelly embrace AI: Innovation in healthcare is sometimes only discussed at the level of digitalizing what we are already doing rather than reinventing for what will do in the future. AI requires a major change in the way a doctor or administrator thinks about work, their role, and the processes of the organization.

  • Long road ahead: Just as it took years for keyhole surgery to replace open surgery, the adoption of AI will take place in parallel with other approaches. It is critical, therefore, to clearly identify those who can lead the transformation while making sure that others do not fail to keep up with the amount of information that is constantly being generated, and which can only be dealt with by an AI-driven system.

The authors no doubt emphasize that AI in healthcare will reduce information asymmetry between healthcare providers, payers, and patients as it redefines the healthcare landscape. While an incredible amount of work is indeed done by technology providers, they alternatively believe that the real power of AI is in opening up opportunities for startups to solve specific problems with applications and verticals.

Currently, patients are the objects of a value chain system; AI empowers them to become more responsible for their health. AI has the potential to revolutionize the way clinical staff access information and how administrative staffs manage resources and financial outcomes.