A medical company has embraced AGI as its annual revenue soars tenfold.

Text | Hai Ruojing

In 2022, in the seventh year of entrepreneurship, Xue Chong started running, using sweat to fight against depression and frustration.

During the capital winter and the sudden cooling of the market, the promised 50 million yuan in financing failed to arrive. His medical SaaS company, Quanzhen Medicine, had to shrink its front line and retreat to Zhejiang. At that time, the medical industry was generally suffering from large - scale layoffs, and Xue Chong was no exception. After a desperate struggle, he decided to squeeze out resources to keep a team of 10 people to explore the unclear AI innovation business.

Unexpectedly, it was this counter - intuitive decision that turned the tables for Quanzhen Medicine and allowed it to catch the "gift" of the large - model era:

In 2025, it received three rounds of financing from investors such as Chuangxin Medical; the new AI business outside medical SaaS saw a 12 - fold increase in signed ARR (Annual Recurring Revenue), reaching 60 to 70 million yuan, and the signed contract value is expected to reach 150 million yuan in 2026.

A medical AI practitioner said that he didn't know about this company until Quanzhen won the bids from Guang'anmen Hospital and the First People's Hospital of Changzhou in 2025. Guang'anmen Hospital attaches great importance to digitalization; and the unit prices of the large - model bids from these two hospitals are not low, indicating that the company has a good foundation and ready - made products.

In Xue Chong's view, "having ready - made products to demonstrate, rather than just existing in PPTs" is one of Quanzhen's competitive advantages. During the interview, he casually opened the "Quanzhentong" APP on his phone, said a patient's self - description, and the software instantly recognized the voice and formed a structured medical record.

Abridge, an American company that does AI intelligent medical records, had its actual ARR exceed 100 million US dollars in May 2025, and its valuation soared to 5.3 billion US dollars. In addition to benchmarking against Abridge in the 2B business, in Xue Chong's blueprint, he also wants to create an "AI intelligent assistant" that doctors can't do without, benchmarking against another unicorn, OpenEvidence.

Although the medical environments in China and the United States are very different, Xue Chong, who has a background as a doctor from Peking Union Medical College and a post - doctor from Johns Hopkins University, believes that if you find the real pain points of hospitals and doctors and use AI to reduce the moments when doctors are "frustrated", the business model can be closed - loop.

After ten years of entrepreneurship, with identities as a doctor, a businessman, and a technology believer intertwined, Xue Chong has always wanted to make machines become "doctor assistants" and has been solving the problem of "who to sell to". He is not satisfied with doing labor - intensive business and doesn't trust the business model of "the wool comes from the pig" (doctors as users and pharmaceutical companies paying the bill).

In Xue Chong's perception: Product innovation is the biggest lever in the business world. "The standard for testing innovation is whether the directly served users are willing to pay." He firmly believes that Chinese doctors are a good consumer group. If doctors don't pay, it means that the product doesn't solve the problem painfully enough.

This common - sense perception seems counter - intuitive in the medical scenario where interests are highly represented and payment methods are very complex. The explosive growth in the AGI era has made Quanzhen Medicine a worthy research sample.

How to catch the era dividends of medical intelligent agents? What are the real demands that can make users pay? How to resist the involution in the medical AI industry?

The following is a dialogue between 36Kr and Xue Chong (edited):

Staring at the "nail" and waiting for the "hammer"

36Kr: Quanzhen Medicine currently benchmarks against the American medical AI company Abridge, and one of its core businesses is to do "AI intelligent medical records". How did you find this scenario? What are the differences between AI writing medical records and traditional voice - input of doctor - patient consultation records?

Xue Chong: At the beginning of 2022, I selected a team of 10 people from the company and led them myself to explore the AI innovation business. Based on our understanding of medical scenarios, we found three directions at that time:

First, through voice and natural language processing, achieve automatic writing of medical records by AI to liberate doctors' hands;

Second, create digital humans to make patients' health data highly visual and provide intelligent warnings for physical examination institutions;

Third, combine software with wearable devices to help patients with long - term health management.

We first wanted to do automatic writing of medical records by AI because we saw a strong demand. Previously, Quanzhen provided electronic medical records and Clinical Decision Support System (CDSS) to more than 8,000 primary - level clinics. We found that what doctors used most was not the auxiliary clinical decision - making function, but the auxiliary input of medical records. They wanted to "click the mouse" to complete a medical record and didn't want to spend a lot of time typing.

When I was doing my post - doctorate in the United States, I often saw surgeons pick up the phone immediately after surgery and start talking about the operation process. At first, I was curious about who they were calling, and later I found out it was a recording, and the assistant would sort it into a text version afterwards. This scene impressed me deeply, and I always wanted to use AI to do this.

When exploring the first two businesses, we encountered a problem: structuring the data of a large amount of messy and non - standard information would consume a lot of manpower. At the end of 2022, ChatGPT emerged, and the natural language processing ability of large models was very strong, and the problem of data alignment was easily solved.

AI intelligent medical records, more precisely called "invisible automatic writing of medical records" (ADS, Ambient Digital Scribing), allows AI to "listen" to the natural conversation between doctors and patients in the environment like an assistant. It not only transcribes voices and records, but also understands and reasons. For example, it can also read patients' examination data and automatically generate structured medical records in combination with the consultation conversation.

After the product was launched, we first upgraded all primary - level electronic medical record users. About more than 1,000 doctors gave good feedback and said it was very smooth to use. At this time, we thought it was time to promote it to hospitals.

36Kr: So, was it the business pain points that forced the technology choice, rather than looking for nails with a hammer?

Xue Chong: Yes, the large model appeared at the right time, giving us a "hammer". If it had come later, it would have been quite difficult.

36Kr: Does AI writing medical records mainly save doctors' manpower? From the hospital's perspective, what are the motivations for purchasing such products?

Xue Chong: It's not only about improving doctors' efficiency, but also includes medical record quality control, risk warning, etc.

For example, when we first cooperated with a top - level tertiary hospital in Zhejiang, it was to solve the problem of medical insurance deduction. Previously, doctors often missed some consumables in the surgical records, and when the medical insurance was verified, there would be a mismatch, resulting in deductions. We let AI read the consumable material data and listen to the doctors' oral surgical records. By comparing the two, it automatically generated matching surgical records. This not only improved efficiency but also helped the hospital solve the pain point of medical insurance verification.

36Kr: At the beginning of 2025, many hospitals actively embraced AI and deployed open - source models such as DeepSeek. How did this affect Quanzhen's business expansion?

Xue Chong: This was both a setback and an opportunity.

At first, many hospitals thought that deploying DeepSeek would be enough and there was no need to purchase medical AI applications, which briefly affected our business. But after the hospitals' trial operation, they found that the inference cost of the 671B large - parameter model was very high and the thinking time was long; the base model couldn't be directly combined with the hospital's business process, and there were some difficulties in implementation.

At this time, Quanzhen had done a lot of "post - training", compressed the parameters of the medical vertical large model to 7B, and the generation accuracy and speed of medical texts were very high. Moreover, the hospital had a certain understanding of the ability boundary of the large model and was in need of professional people to implement AI. It was much easier for us to undertake the business.

36Kr: There should be many companies that can use AI for medical interpretation, such as large - scale voice recognition companies and listed companies that do electronic medical records and CDSS. In your opinion, who are the strong competitors in this niche direction?

Xue Chong: When developing products and entering the market, it's easy to have "imaginary enemies", which are not the same as the actual competitors.

The product logic of large - scale voice technology companies stays in the "perceptual AI" stage in the early stage, just hearing and transcribing. But the medical scenario requires more understanding, reasoning, and structured processing. So we created a dual - architecture of voice + large - model reasoning, and the accuracy of extracting key points from medical records is very high, ensuring good word - of - mouth.

Traditional medical informatization companies, especially electronic medical record manufacturers, do have opportunities because they control the entrance to the doctor's workstation. But in fact, the real investment of many companies in this area is still relatively low, more like a defensive move. In addition, the Chinese electronic medical record market is different from that in the United States. In the United States, Epic dominates, while in China, there is no monopoly yet.

For doctors in large hospitals, they are very familiar with the diagnosis and treatment of common and frequently - occurring diseases and usually don't need CDSS for auxiliary decision - making. They need tools that can write medical records and record data, as well as auxiliary tools for treating difficult and complicated diseases.

When new things emerge, the competitive players in the market are often not the old forces with strong inertia. In the decision - making process of listed companies, they need to consider the "chicken - and - egg" problem. If they don't see a clear business model for the time being, it's difficult for them to invest a lot of manpower and funds to develop new businesses. I think the real opportunities in the future belong to startup companies focusing on the application layer.

Understanding human nature and creating a Chinese version of OpenEvidence

36Kr: In addition to the B2B large - model and intelligent agent development business, you also launched the doctor - version "Quanzhentong" APP. Recently, OpenEvidence, an American company aiming to be the "AI co - pilot" for doctors, has seen its valuation soar to 12 billion US dollars. How do you understand such products?

Xue Chong: In essence, OpenEvidence is an AI - enhanced version of PubMed. Just like ChatGPT is the AI version of Google, OpenEvidence solves the problems at both ends of the medical search process.

Early on, there was a discipline called information search science. For medical retrieval, we used to have to rack our brains to combine keywords, otherwise we might not be able to find the article. After finding the article, we still had to read and analyze it word by word, consuming a lot of manpower. After the emergence of large AI models, accurate retrieval can be achieved through natural language, and AI can also do reading and analysis in seconds and communicate naturally with doctors. This convenience has brought about a huge improvement in efficiency, and doctors are naturally willing to use it.

In China, the core of making a similar product lies in the copyright of academic databases, especially international core databases. Quanzhen has subscribed to multiple relevant databases and reached reasonable application agreements with them, both giving play to the value of AI and protecting the original copyright.

36Kr: It sounds more inclined to the medical research scenario. Will Chinese doctors use such products frequently in the clinical diagnosis and treatment process?

Xue Chong: Actually, it's all about "diseases".

There are two types of diagnosis and treatment scenarios in hospitals. The first type is for out - patient patients. The pace is fast and there are many mild cases. At this time, doctors can't stare at the screen and communicate with patients while looking at the AI prompts. Moreover, doctors are very familiar with the diagnosis and treatment processes of common and frequently - occurring diseases and usually don't need AI assistance.

How do top - level doctors usually work? There is usually a student or assistant sitting beside them, helping them type and input medical records. Now we use AI to replace the assistant and do "invisible" medical record keeping, allowing doctors to focus on interacting with patients. This is similar to Abridge in the United States.

The second type is for patients with difficult and complicated diseases, and many of them are admitted to the ward for treatment. Facing such cases, doctors need to conduct long - term research, thinking, and even discuss with colleagues. The in - depth thinking scenario for difficult and complicated diseases is where medical decision - making AI can best play its value.

36Kr: Is it difficult to promote AI products for doctors? Does it require relatively high marketing costs?

Xue Chong: I don't think promotion is a problem.

Many recently popular AI products don't rely on promotion, such as DeepSeek, ChatGPT, Gemini 3.0, etc. As long as the product is strong enough, users will actively choose it. The doctor - version APP of Quanzhentong has not been promoted at all at this stage, but the number of doctor downloads has been growing naturally.

My logic is that we must make the product to the point where "doctors are willing to pay for it" before starting to promote it vigorously. Only when users are willing to pay themselves does it mean that the product is really useful to them and can solve the core pain points.

As for the promotion method, it's not difficult. We will promote it precisely at academic conferences with a high density of doctors, and we can also use new media channels such as Douyin, Xiaohongshu, and Video Accounts. During the Spring Festival in 2026, Quanzhentong will launch an upgraded "international version" APP for a trial run. Doctors are relatively free during the Spring Festival and are more willing to try new products, which is also easier to generate word - of - mouth.

36Kr: Will you choose a business model like OpenEvidence, providing free services for doctors and having pharmaceutical companies pay, making money from medical digital marketing?

Xue Chong: This won't be our core business model.

Pharmaceutical companies can be one of the payers, but I think a reasonable business logic must face the customers directly and achieve direct conversion. Whether the directly served customers are willing to pay is the most important standard for testing the product's value.

If it's to solve problems for hospitals, then the hospitals should pay; if it's to solve pain points for doctors, then the product should be made to the point where doctors are willing to pay. For example, for the clinic SaaS system developed by Quanzhen Medicine, I require that clinics must be willing to pay for the software service, rather than relying on commissions from guiding patients to physical examination institutions.

If there is a "middleman" in the payment process, the transmission of the product's value will be distorted.

Like many AI products, we also offer free trials, but the independent charging direction in the future is clear. If a product can't impress doctors and they think it has no value, they won't use it even if it's given for free. On the contrary, if the product can really solve the pain points, doctors will find ways to pay.

In the Internet industry, the business model of "the wool comes from the pig" has misled many people. There are many companies that first trapped users for free and then tried to find a business model, but ended up burning through their capital chain.

36Kr: There is a view in the industry that Chinese doctors have low willingness and ability to pay, and it's difficult to tap commercial value from the C - end. What do you think?

Xue Chong: The doctor group definitely has strong commercial value.

Whether in terms of income level, payment ability, or consumption willingness, doctors are a very high - quality customer group. If doctors are not willing to pay, it's because the "pain points" that can make them pay haven't been found. For example, when I was a doctor, sometimes when I urgently needed a paper, I would post a help - seeking message on the doctor forum. Even if the other party asked for a few hundred yuan each time, I would pay without hesitation.

Many people think that the income of Chinese doctors is low, mainly compared with doctors in the United States. If we consider the domestic purchasing power parity and the average urban income, doctors are actually a group with strong payment ability.