An Interview with Wang Xiaohang, CTO of Ping An Group: Taking a Step Ahead – Ping An's AI "Pragmatism"

On August 20, when Ping An Group released its semi-annual report, a figure shocked the industry: Ping An Group's daily token consumption reached 120 billion. Public information at the same time showed that some leading financial institutions consumed tokens in the tens of billions of yuan per day.

However, Wang Xiaohang, Chief Technology Officer (CTO) of Ping An Group and General Manager of Ping An Technology, is not surprised by this figure: "The figure of 120 billion is as of the release of the semi-annual report (June 30). Ping An Group's daily token consumption has now reached 210 billion."

This demonstrates Ping An's speed in AI development, and it's also Wang Xiaohang's "report card" after one year with the company. The outside world is focused on what this CTO, with his internet industry background, will bring to Ping An.

The answer is gradually becoming clear: Ping An's AI strategy has been fully rolled out along three main lines; over 47 million people used AI-powered quick services in the first half of the year; the AI generation rate of new code reached 62% in the first half of the year; in AI diagnosis and treatment, especially in the diagnosis of tumor diseases, the consistency with experts reached over 90%; AI inquiry and processing covered 88% of the group's business scenarios; and AI-assisted sales reached 57.313 billion yuan...

In Wang Xiaohang's view, this is just the beginning. "The consumption and value output of AI in the financial and insurance industry are on the verge of a breakthrough, and Ping An's goal is to be half a step ahead."

Three-pronged approach: Ping An's AI panorama

Wang Xiaohang summarized Ping An's current AI strategy into three main lines: integrated finance, healthcare, and efficiency improvement for internal enterprises .

Integrated financial services are the primary focus. Ping An boasts 253 million individual customers, with a peak monthly active user base exceeding 90 million, making it "the industry's largest app traffic ecosystem." Converting and serving this traffic effectively is a crucial task for Wang Xiaohang.

What Ping An Technology's team has done is to connect various services using "AI-powered quick services": multiple entry points such as banking, insurance, securities, and healthcare are integrated into a single traffic platform, using AI technology to support customers in achieving "one-sentence consultation and processing, one-sentence claims, financing, transactions, and emergency assistance." This seamless integration of over 300 cross-entity and cross-business services provides a one-stop solution. The interim report reveals that this service has been used 47 million times cumulatively, with the number of users rapidly increasing since its launch.

In Wang Xiaohang's view, AI not only brings "the ultimate consultation and service experience", but also transforms this APP ecosystem into an "emerging channel" - the month-on-month growth rate of online incremental orders remains in double digits, and the customer acquisition cost is significantly lower than that of external channels.

The second main focus is healthcare. The goal is to "create the most professional AI doctor." Wang Xiaohang explained that Ping An's AI in general practice, specialist care, and multidisciplinary team (MDT) consultations for serious illnesses has "already reached the level of top-tier hospitals" in terms of accuracy in diagnosis, treatment, and consultation. In the field of oncology, the company has collaborated with Peking University and Ping An Good Doctor to develop specialized AI, with expert consistency exceeding 90%.

Online diagnosis is just the starting point for medical treatment. Ping An also connects online and offline resources through the "four-to" service system (family doctors at the hospital, hospital network at the hospital, corporate benefits at the enterprise, and door-to-door service at home) to form a closed loop for medical treatment.

The third main focus is on internal operations . Ping An has over 250,000 internal staff, and the three major enterprise AI projects—AI Coding, AI Operations, and AI Office—are rapidly progressing internally, yielding considerable results.

In terms of AI coding alone, Ping An Group's new code AI generation rate reached 62% in the first half of the year—double the 30% level at the beginning of the year. The promotion of AI operations has also enabled double-digit improvements in working hours for relevant repetitive and transactional benchmark positions.

"Whether it's AI services in integrated finance or AI services in healthcare, the future applications will be on a massive scale, reaching hundreds of millions of times per year."

Wang Xiaohang predicts that the combination of ultra-large-scale external applications and ultra-high-frequency internal applications will constitute the key area for Ping An AI to "truly generate commercial and customer value".

Stay focused: "Expert-level AI for finance and healthcare"

Wang Xiaohang proposed early on that Ping An's AI model construction should "not focus on general-purpose large models, but rather on expert-level AI in finance and healthcare." After more than a year of exploration, his view has become increasingly firm.

Wang Xiaohang believes that over the past year, the capabilities of general models have been continuously improving. In general fields such as routine health consultations and general practice consultations, the capabilities of cutting-edge models have also improved significantly, and "the models themselves are no longer significantly different."

For insurance institutions, future competition will focus more on product experience and traffic advantages, which is also where Ping An's strength lies.

Therefore, online AI medical consultation is only the starting point for insurance institutions. The "real challenge" is how to help customers solve serious medical problems and form a service loop .

Wang Xiaohang thus broke down Ping An AI's positioning in the medical field into two layers: first, to be the "most professional AI doctor" in the serious medical field; second, to connect medical resources and connect "four-to" services.

In his view, Ping An's advantages and barriers to entry are rooted in its "integrated finance + healthcare and elderly care" ecosystem, which is reflected in three aspects: First, real and accurate medical needs - Ping An covers nearly 200,000 corporate clients every year, enabling it to accurately understand their health status and serious medical needs; Second, data accumulation - long-term accumulation of insurance claims, online medical, and offline medical network data, in collaboration with Peking University and Good Doctor, has formed "the industry's largest and highest quality medical and health records and medical knowledge project"; Third, service closed loop - AI doctors serve as a unified entry point, proactively identifying and reaching customers, connecting the "four-to" service system, and solving pain points in online and offline medical treatment.

All of this points to one conclusion: Ping An needs to continue to invest in the research and development of professional large-scale models, and at the same time, integrate them deeply and thoroughly with front-line services.

Daily token investment: from 30 billion to 210 billion

The use of tokens is a hot topic that all financial institutions cannot avoid. Ping An's latest interim report also disclosed that the group's average daily token consumption jumped from 30 billion yuan in December 2025 to over 120 billion yuan in June 2026. Wang Xiaohang provided an updated figure: "Our annual report cutoff date is June 30th, and the data at that time was 120 billion yuan. Now the daily token consumption has reached 210 billion yuan."

However, Wang Xiaohang emphasized that token consumption "is not about maximizing it," but rather about whether a "positive cycle" of tokens is achieved in the three stages:

Firstly, regarding the supply side, can a stable and efficient computing infrastructure be formed? Ping An, through its long-term deep cultivation of intelligent computing and AI infrastructure, has built a domestically produced multi-GPU heterogeneous computing power cluster.

Secondly, regarding efficiency, can the supply cost of tokens be continuously reduced? Wang Xiaohang revealed that in the past six months, the prices of computing hardware facilities such as memory and GPUs have more than doubled, while Ping An's unit token cost has been "reduced by more than half." This is due to the progress of full-stack technologies, from context optimization, model optimization, and inference optimization to computing power scheduling.

Thirdly, regarding value, does the use and consumption of tokens reflect genuine customer needs? Within Ping An, token consumption primarily stems from ultra-large-scale applications serving 253 million customers and ultra-high-frequency applications targeting key internal positions.

"The next step in the development of AI applications depends on the development of the above three links," Wang Xiaohang predicted. With the simultaneous optimization of supply, efficiency, and application, the application and value of AI in the insurance industry will "continue to explode," and Ping An's "development space" in AI will continue to expand.

AI Applications: Security and Boundaries

Finance and healthcare are two industries with the highest requirements for professionalism and rigor. AI technology itself has certain "illusion" problems, so it is particularly important to do a good job in security management and control of application boundaries.

Wang Xiaohang introduced that Ping An is continuously advancing three things regarding the controllability of AI.

First, there is professional quality control. Ping An is one of the very few institutions in the industry that conducts AI medical quality control in the healthcare field "according to the National Health Commission's doctor quality control requirements". To date, more than 130,000 AI medical consultation cases have undergone professional quality control, and the hallucination rate, error rate and red line rate in specialist and general practice are all "less than three per thousand", far exceeding the quality control requirements.

Second, professional verification is conducted through medical knowledge engineering, case databases, and evidence-based document databases to ensure that responses are traceable and controllable.

Third, safety safeguards and expert review are implemented, clearly defining boundaries in key decisions such as transactions and formal medical treatment recommendations, and entrusting them to expert teams.

Regarding the three-tiered defense mechanism proposed in the "Administrative Measures for Cybersecurity in the Banking and Insurance Industries (Draft for Comments)," Wang Xiaohang stated that Ping An has established a corresponding mechanism: the first line of defense is the technology department (Group Technology and Information Department) responsible for overall security management; the second line of defense is the Group Risk Management Department responsible for risk identification, empowerment, assessment, and monitoring; and the third line of defense is the audit department's independent perspective in examining, holding accountable, and promoting rectification, with member companies forming a closed-loop protection system as the primary responsible parties.

Faced with the challenge of scaling up intelligent agents, Ping An's management logic is based on the principle of "tiered access and minimum necessary permissions": on the one hand, in scenarios involving customer rights, funds, and key medical decisions, AI and intelligent agents are "not authorized unless necessary";

On the other hand, all key decisions are reviewed by qualified personnel; the entire process is monitorable, traceable, and auditable.

"AI will become increasingly autonomous, but at the same time, its boundaries should become increasingly clear."

Reshaping Business: Supporting and Empowering Frontline Workers

Speaking about the profound impact of AI on the insurance industry, Wang Xiaohang believes that the health insurance sector is where the effectiveness of AI has been most fully unleashed in this wave of development. As AI improves its ability to detect, predict, and manage diseases, insurance is expected to "cover populations that were previously inaccessible, providing greater accessibility."

In July of this year, Ping An launched a matrix of AI products for cancer patients at the World Artificial Intelligence Conference (WAIC), covering scenarios such as disease identification, early screening of high-risk groups, cancer recurrence, and Alzheimer's disease. After an incident, a medical concierge will provide full-process services.

Meanwhile, increasingly sophisticated AI agents are also empowering frontline sales teams. In the life insurance sector, Ping An has an AI agent platform called "Marketing Askbob," which provides individual insurance agents with services such as policy inquiries and strategy interpretations. Currently, 39% of agents use it daily, and it is being upgraded from a "Q&A consultation tool" to an assistant covering the entire process of customer acquisition, outreach, and conversion.

According to the interim report, Ping An's AI-assisted sales reached RMB 57.313 billion in the first half of the year. In the auto insurance sector, due to more standardized processes and more mature AI applications, 94% of policies issued through the auto agency channel were completed intelligently within one minute.

"Even the most powerful AI cannot replace the connection and trust provided by agents." Wang Xiaohang pointed out that the current focus of AI is to free frontline sales teams from more transactional and repetitive tasks.

Resource allocation: How will the investment of over 10 billion yuan be spent?

Ping An invests over 10 billion yuan in technology annually, and how to make better use of these funds is a focus of external attention.Wang Xiaohang summarized the group's technology investment allocation logic in two points: first, structural optimization, and second, value orientation.Structurally, the approach is "save where necessary and spend where appropriate." Ping An divides its technology investment into three categories: basic research and operations, large-scale innovation, and forward-looking incubation and strategic planning. Resources are "structurally and proactively shifting from basic research to innovation."In terms of value, Ping An has a target management and production tracking mechanism called the "K System," which covers long-term, medium-term, and short-term goals, tracks production and evaluates results, and assigns responsibility to individual managers. Wang Xiaohang said that the key is that "if the value can be clearly explained and implemented, we are willing to invest manpower and computing power," focusing on the "five intelligences"—intelligent operation, management, marketing, and service.Wang Xiaohang revealed that Ping An is pursuing a pragmatic "industrial AI" approach: "We will not invest a large amount of resources in problems that can be solved by general AI in the next one to two years."Ping An's industrial AI primarily revolves around four key elements: data, algorithms, computing power, and application scenarios. In terms of data, the "Nine Data Banks" project, spanning five to six years, has accumulated data assets in the financial and healthcare verticals. Regarding algorithms, its financial big data model maintains its top position on the CNFinBench public leaderboard, while its healthcare big data model 3.5 achieved the highest score globally in the HealthBench Hard benchmark. In terms of computing power, the emphasis is on both supply and efficiency. Wang Xiaohang considers this "the path that industrial AI should take.""Ping An has provided a large stage and space for technology, which can help Ping An create greater value within the industry," said Wang Xiaohang. "Currently, the consumption and value output of AI in the insurance industry are on the verge of a breakthrough, and Ping An's goal is to be half a step ahead."Perhaps this is the most pragmatic and intelligent stance taken by this trillion-dollar financial giant in the wave of AI.Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.