Reflections on the Business Model of Medical AI Products

[baoshi.rao]

Many people confuse business models with profit models, but the author emphasizes the distinction between the two and further analyzes the unique characteristics of medical AI products in terms of business models.

Preface:

This is my third summary on medical AI, and this time I will discuss business models.

While preparing this article, China's first AI product to pass Class III medical device approval was officially announced. This is undoubtedly a significant positive development for the medical AI industry, which is currently in a downturn. As the saying goes, the first step is always the hardest. It is believed that more similar products will gain approval in the future, and the entire medical AI industry may enter a new phase of large-scale commercialization.

Many people cannot distinguish between business models and profit models. Online, I often see articles labeled as analyzing "business models" that are actually discussing profit models.

In reality, business models and profit models are two different concepts: A business model describes the fundamental principles of how an organization creates, delivers, and captures value, while a profit model refers to the way a company generates profits.

Simply put, a profit model is only a part of a business model.

The Business Model Canvas is a unified language that can intuitively describe and evaluate a business model. It helps entrepreneurial teams generate ideas, reduce guesswork, ensure they target the right users, and solve problems effectively. The canvas consists of nine interrelated modules, as shown below:

The Business Model Canvas is an excellent tool for analyzing business models. This article also uses this tool to analyze the business model of medical AI products, taking the most mainstream product in the field—medical imaging-assisted diagnostic products—as an example. Due to my limited expertise, I welcome any feedback on any inaccuracies in the article.

Customer Segments describe the different target groups and institutions a company aims to serve, forming the core of any business model.

Generally, customers can be categorized into mass markets, niche markets, multi-sided platforms, etc. Medical imaging-assisted diagnostic products typically target a niche market, as they cater to a very specific group—radiologists. This group can be further subdivided into four types, each with different interests and demands. A detailed analysis can be found in my previous article, "The Challenges of Medical Imaging AI Implementation: The Demand Perspective", which I won’t repeat here.

Designing a good business model should start with selecting which customer segments to serve. Once determined, the business model should be tailored to meet the unique needs of these groups. Customers are the heart of the business model, the driving force behind its effective operation.

Value Propositions are the products and services offered to a customer segment that create value for them. They are the reason customers choose one company over another, addressing their problems or fulfilling their needs.

A value proposition creates value for a specific group by customizing a combination of elements tailored to their needs. Common elements of a value proposition include:

Taking medical imaging-assisted diagnostic products as an example, we analyze them based on customer segments as follows:

Channels are the ways a company communicates with and delivers its value proposition to customer segments. Their main functions include:

Horizontally, channels are typically divided into two types: owned channels and partner channels. Vertically, there are different channel stages, each serving a distinct purpose. Using medical imaging-assisted diagnostic products as an example, the analysis is as follows:

Customer Relationships are the types of relationships a company establishes with a customer segment. Customer relationships serve three main purposes:

Generally, there are six types of customer relationships:

For medical imaging AI products, the typical relationships are "dedicated personal service" and "collaborating with customers to co-create."

Dedicated Service:

Usually, the decision to purchase a product is made by a specific role. However, the procurement process in the medical industry is complex, involving multiple stakeholders, making it difficult to push through in one go. To achieve sales targets, sales personnel need to establish deeper, personal relationships with customers, which often requires long-term effort. It’s evident that almost all sales personnel for medical AI products come from hospitals, pharmaceutical companies, or medical device firms, bringing with them existing customer resources.

Research Collaboration:

China's medical system emphasizes the integration of "clinical practice, teaching, and research." Research performance determines hospital funding and doctors' career advancement, making "publishing papers" a pain point all doctors must address. The development of AI, especially deep learning, has opened new research directions for clinical problems. The "doctor provides ideas, AI companies provide solutions" model of "medical-engineering collaboration" has emerged. This collaborative customer relationship is relatively stable, but companies must evaluate the return on investment.

Revenue Streams represent the cash income generated from each customer segment. A business model's revenue streams can be divided into two categories:

Currently, medical AI companies primarily generate revenue through the following methods:

Key Resources are the most important assets required to ensure a business model operates smoothly. Different business models require different key resources, generally categorized into the following four types:

Medical AI companies are typical knowledge-intensive tech industries, so their key resources are human resources and knowledge resources.

Knowledge resources include accumulated data, algorithm patents, software copyrights, product copyrights, etc. One particularly important resource is the medical device sales license.

Human resources include algorithm teams, annotation teams, sales teams, R&D teams, and operations teams. Algorithm teams are the core of all AI companies. Sales teams are also crucial in the medical industry, as it is a high-barrier, relatively closed field. Building a sales team with rich medical resources is essential.

Key Activities are the most important tasks required to ensure a business model operates normally. Every business model has a set of key activities—the most critical actions necessary for successful operation. Key activities can be divided into three categories:

Medical AI companies primarily focus on software products for B2B clients, with little involvement in production. They are typical examples of business models where providing solutions is the key activity.

After identifying the needs of various customer segments and defining the value propositions our products can offer, we still need to bundle these scattered value propositions into tailored solutions to drive sales. For example, most medical AI companies develop a series of products based on various needs and integrate them into solutions for specific departments, diseases, or workflows.

Key Partnerships are the network of suppliers and partners required to ensure a business model operates smoothly.

There are generally three motivations for forming partnerships:

Key partnerships can be categorized into the following types:

Most medical AI companies are emerging internet startups. As their businesses expand, they may extend into upstream and downstream areas such as medical equipment and information systems. It’s impossible to develop all aspects in-house, so establishing partnerships for resource complementarity is the best choice. Another important partner is medical institutions, including hospitals and medical associations or organizations.

Cost structure refers to all costs incurred by a business model. Ensuring the smooth operation of the aforementioned modules comes at a cost, and only when these modules are clearly defined can costs be accurately calculated. Based on the focus on costs, business models can be divided into two categories:

Medical AI products should adopt a value-driven business model, with the following three key costs:

1. Value Proposition: Depending on the细分客户 (segmented customers), medical AI products offer more direct and significant value propositions to frontline doctors, while providing only indirect and weaker value propositions to decision-makers with influence. In product design, expanding the functional boundaries to offer compelling value propositions for decision-makers is a critical consideration for every designer.

2. Revenue Sources (RS): Although several existing and potential business models for medical AI are listed, the current primary model involves hospitals making one-time purchases through tenders, typically by large tertiary hospitals, as they are the only ones with sufficient demand and purchasing power at this stage. However, with only a few thousand tertiary hospitals in China, the sustainability of so many companies competing in this limited market is questionable.

3. Cost Structure (CS): The main cost components for medical AI companies include high R&D expenses, as internet and software development are inherently costly industries. AI products, being software or internet-based, also rely on massive data for algorithm training, leading to additional costs for data acquisition and annotation. Medical AI faces even higher barriers and costs, as medical data must be sourced from professional institutions and annotated by qualified doctors.

4. Key Resources (KS): The医疗器械销售许可证 (Medical Device Sales License) is highlighted as a critical resource, as recognized by all players in the medical industry. While the recent approval of the first Class III-certified AI product by China's NMPA (formerly CFDA) is encouraging, it's important to note that similar products, like the CTFFR, had already received FDA and CE certifications years earlier. Many popular products currently on the market lack such credentials.

5. Financial Resources: Another challenge in the医疗AI industry is securing financial resources. The medical industry is inherently long-cycle, with profits often taking years or longer to materialize. Medical AI products, requiring extensive algorithm training, exacerbate this timeline. Analysis of revenue models and cost structures reveals that without substantial cash flow, such business models are unsustainable in the short term. Thus, founders must either have significant personal resources or secure continuous external funding—the latter being the mainstream approach. Unfortunately, with China's economy in a downturn, "funding difficulties" and "capital winter" have become buzzwords in 2019. The医疗AI industry's financing拐点 (turning point) began in 2018, and under these conditions, companies unable to secure funding will gradually exit the market.

Business models are complex concepts. Here, we’ve used the Business Model Canvas as a tool to briefly analyze the商业模式 of medical AI products, hoping to provide some insights. Finally, a standard Business Model Canvas is included as a summary.