On August 20th, when Ping An Group released its interim report, a figure stunned the industry: the company's internal daily token consumption had reached 120 billion. For comparison, publicly available information at the time showed that some leading financial institutions were consuming only tens of billions of tokens per day.
However, for Wang Xiaohang, Chief Technology Officer of Ping An Group and General Manager of Ping An Technology, this number was hardly surprising. "The 120 billion figure was as of June 30th, the interim report cutoff date. Our daily token consumption has now climbed to 210 billion," he stated.
This is the pace at which Ping An is advancing in AI, and it serves as Wang Xiaohang's "report card" one year after joining the company. The question on everyone's mind is: what will this CTO, who comes from an internet industry background, bring to Ping An?
The answer is gradually becoming clear. Ping An's AI strategy, built on "three main pillars," is now fully deployed. In the first half of the year, over 47 million people used AI-powered quick services. The AI generation rate for new code reached 62%. In AI diagnostics, particularly for oncology, consistency with expert opinions surpassed 90%. AI-powered inquiry and processing now covers 88% of the group's business scenarios, and AI agents facilitated sales totaling RMB 57.313 billion.
In Wang Xiaohang's view, this is only the beginning. "AI consumption and value creation in the financial and insurance industry are on the eve of a breakthrough. Ping An's goal is to stay half a step ahead."
Three-Pronged Strategy: The Full AI Landscape
Wang Xiaohang summarizes Ping An's current AI deployment into three main pillars: integrated finance, healthcare, and internal efficiency enhancement.
The first pillar is integrated finance. Ping An serves 253 million individual customers with a peak of over 90 million monthly active users, making it the "largest APP traffic ecosystem in the industry." Converting this traffic into quality service is one of Wang Xiaohang's key responsibilities.
The Ping An Technology team's approach is to connect various services through "AI quick services." Banking, insurance, securities, and medical entry points are integrated into a unified traffic platform, using AI to support customers in "one-sentence inquiries and transactions, one-sentence claims, financing, trading, and emergency assistance." This pathway integrates over 300 cross-entity, cross-business services in one seamless flow. According to the interim report, this service has been used over 47 million times, with user numbers rapidly climbing since its launch.
In Wang Xiaohang's view, AI not only delivers an "ultimate consultation and transaction experience" but also transforms the APP ecosystem into a "new channel." Online incremental orders are maintaining double-digit month-over-month growth, with customer acquisition costs significantly lower than external channels.
The second pillar is healthcare, with the goal of "building the most professional AI doctor." Wang Xiaohang notes that in general practice, specialty care, and multi-disciplinary consultation for major diseases, AI diagnosis and consultation accuracy is "already aligned with the level of top-tier hospitals." In oncology, the company has partnered with Peking University and Ping An Healthcare and Technology to develop specialized disease AI, achieving over 90% consistency with expert opinions.
Online diagnosis is just the starting point of the medical journey. Ping An also connects online and offline resources through its "Four-Reach" service system—reaching family doctors online, hospital networks, corporate benefits, and home-based services—to form a closed-loop healthcare experience.
The third pillar targets internal operations. With more than 250,000 internal support staff, Ping An is rapidly advancing three enterprise AI initiatives: AI Coding, AI Operations, and AI Office, delivering substantial results.
In AI Coding alone, the AI generation rate for new code in the first half reached 62%, double the 30% level at the beginning of the year. The rollout of AI Operations has also led to double-digit efficiency gains in repetitive, transactional benchmark positions.
"Whether it's AI services for integrated finance or healthcare, these will become ultra-large-scale customer-facing applications in the future. We expect them to reach over a billion uses per year," Wang Xiaohang predicts.
He believes that ultra-large-scale external applications combined with ultra-high-frequency internal applications will constitute the "key areas where Ping An's AI truly generates commercial and customer value."
Staying Focused: Expert-Level AI for Finance and Healthcare
Wang Xiaohang has long advocated that Ping An's AI model development should "not pursue general-purpose large models, but rather focus on expert-level AI for finance and healthcare." After more than a year of exploration, his conviction has only strengthened.
He believes that over the past year, general-purpose models have seen continuous capability improvements. In fields like daily health consultations and general practice, frontier models have benefited from the rising tide, and "the gap between models has largely narrowed."
For insurance institutions, future competition will center more on product experience and traffic advantages—areas where Ping An already excels.
Online AI medical consultations, therefore, are just the starting point for insurance companies. The "real challenge" lies in helping customers address serious medical issues and forming a closed service loop.
Wang Xiaohang thus breaks down Ping An's AI positioning in healthcare into two layers: first, becoming the "most professional AI doctor" in serious medical fields, and second, connecting medical resources and the "Four-Reach" services.
In his view, Ping An's advantages and moats are rooted in its "integrated finance + healthcare and elderly care" ecosystem, reflected in three aspects. First, authentic and precise medical needs: Ping An covers nearly 200,000 corporate clients annually, enabling precise understanding of customer health status and serious medical needs. Second, data accumulation: long-term accumulation of insurance claims, online medical, and offline medical network data, combined with partnerships with Peking University and Ping An Healthcare and Technology, has built "the industry's largest, highest-quality medical health records and medical knowledge engineering system." Third, closed-loop services: the AI doctor serves as a unified entry point, proactively identifying and reaching customers, connecting the "Four-Reach" service system to address online and offline medical pain points.
All of these elements point to one conclusion: Ping An must continue investing in specialized large model R&D while deeply integrating these models with frontline services.
Token Investment: From 30 Billion to 210 Billion Daily
Token usage is a hot topic that no financial institution can avoid. Ping An's latest interim report disclosed that daily internal token consumption jumped from 30 billion in December 2025 to over 120 billion by June 2026. Wang Xiaohang provides an even more recent figure: "Our reporting period ended on June 30th, when the number was 120 billion. Our current daily token consumption has reached 210 billion."
However, Wang Xiaohang emphasizes that token consumption is "not for the sake of volume." The key is whether a "positive cycle" of token usage is achieved across three aspects.
First, on the supply side, the question is whether stable and efficient computing infrastructure can be established. Ping An has built a domestic multi-GPU heterogeneous computing cluster through long-term investment in intelligent computing and AI infrastructure.
Second, on the efficiency side, the question is whether token supply costs can be continuously reduced. Wang Xiaohang reveals that over the past six months, hardware costs for memory and GPUs have more than doubled, yet Ping An's unit token cost has "dropped by more than half," driven by full-stack technological advances from context optimization, model optimization, and inference optimization to computing scheduling.
Third, on the value side, the question is whether token usage reflects genuine customer demand. Ping An's internal token consumption primarily comes from ultra-large-scale applications serving 253 million customers and ultra-high-frequency applications for key internal positions.
"The next stage of AI application development depends on these three aspects," Wang Xiaohang judges. With simultaneous optimization across supply, efficiency, and applications, AI applications and value in the insurance industry will "continue to explode," and Ping An's "development space" in AI will keep expanding.
AI Applications: Safety and Boundaries
Finance and healthcare are the two industries with the highest demands for professionalism and rigor, yet AI technology inherently carries "hallucination" risks. Managing safety and application boundaries is therefore particularly critical.
Wang Xiaohang explains that Ping An is advancing three initiatives for AI controllability.
The first is professional quality control. Ping An is one of the few institutions in the industry conducting AI medical quality control "in accordance with the National Health Commission's physician quality control requirements." To date, over 130,000 AI medical consultation cases have undergone professional quality control, with hallucination rates, error rates, and red-line rates "all below 0.3%," far exceeding quality control requirements.
The second is professional validation. Through medical knowledge engineering, case databases, and evidence-based documentation libraries, Ping An ensures that responses are traceable and controllable.
The third is safety guardrails and expert review. For critical decisions such as transactions and formal medical treatment recommendations, clear boundaries are defined and delegated to expert teams.
Regarding the "three lines of defense" collaborative governance proposed in the "Banking and Insurance Network Security Management Measures (Draft for Comments)," Wang Xiaohang states that Ping An has established corresponding internal mechanisms. The first line of defense is the technology function (Group Technology and Information Department) responsible for overall security management. The second line is the Group Risk Management Department, responsible for risk identification, empowerment, evaluation, and monitoring. The third line is the audit department, which independently verifies, holds accountable, and drives remediation, with member companies serving as primary responsible parties to form a complete defense loop.
Facing the challenge of scaling intelligent agents, Ping An's governance logic is based on "tiered access approval and least-privilege principles." On one hand, for scenarios involving customer rights, funds, and critical medical decisions, AI and agents are "not authorized unless necessary." On the other hand, all critical decisions require review by professionally qualified personnel, with full monitoring, traceability, and auditability throughout.
"AI will become increasingly autonomous, but AI's boundaries should also become increasingly clear."
Transforming Business: Supporting and Empowering the Frontline
Discussing AI's profound impact on the insurance industry, Wang Xiaohang believes that health insurance is where this wave of AI efficiency gains will be most fully realized. As AI improves in disease prediction and management, insurance will be able to "cover populations that were previously uninsurable, providing greater inclusivity."
In July, at the World Artificial Intelligence Conference, Ping An launched a specialized disease AI product matrix for oncology patients, covering specialized disease identification, high-risk population early screening, cancer recurrence, and Alzheimer's disease scenarios, with medical case managers providing full-process services after claims.
Meanwhile, increasingly capable AI agents are empowering frontline sales teams. In the life insurance sector, Ping An has an intelligent agent platform called "Marketing AskBob" that provides agents with policy inquiries and strategy interpretation services. Currently, 39% of agents use it daily, and it is evolving from a "Q&A tool" into an assistant covering the entire process of customer acquisition, outreach, and conversion.
Interim report data shows that AI agents facilitated sales of RMB 57.313 billion in the first half. In auto insurance, where processes are more standardized and AI applications more mature, 94% of policies through the auto channel are issued intelligently within one minute.
"No matter how powerful AI becomes, it cannot replace the connection and trust that agents provide," Wang Xiaohang points out. The current focus of AI is to free frontline sales teams from transactional and repetitive tasks.
Resource Allocation: How Is the Over Ten Billion Investment Spent?
Ping An invests over RMB 10 billion annually in technology. How this capital is best utilized is also a focus of external attention.
Wang Xiaohang summarizes the group's technology investment allocation logic in two points: structural optimization and value orientation.
Structurally, it's about "saving where appropriate and spending where appropriate." Ping An divides technology investment into three buckets: foundational R&D and operations, scaled innovation, and forward-looking incubation and deployment. Resources are being "structurally and proactively shifted from foundational R&D toward innovation."
On value, Ping An has a target management and investment tracking mechanism known as the "K System," covering long-term, medium-term, and short-term goals. It tracks investment outcomes and evaluates results, with accountability assigned to individual managers. Wang Xiaohang says the key is that "if value can be clearly articulated and delivered, we are willing to invest manpower and computing resources." The focus is on the "Five Intelligences"—intelligent operations, management, administration, marketing, and services.
Wang Xiaohang reveals that Ping An follows a pragmatic "industrial AI" path: "We will not invest significant resources in problems that general-purpose AI can solve within the next one to two years."
Ping An's industrial AI revolves around four elements: data, algorithms, computing power, and scenarios. On data, the "Nine Libraries" project, sustained for five to six years, has accumulated data assets in finance and healthcare verticals. On algorithms, the financial large model ranks first on the CNFinBench public leaderboard, while the medical large model 3.5 achieved the highest global score on the HealthBench Hard evaluation. On computing, the focus is on supply and efficiency. This is what Wang Xiaohang considers "the path industrial AI should follow."
"Ping An has given technology a great stage and space, and technology can help Ping An create greater value in the industry," Wang Xiaohang says. "The AI consumption and value creation in the insurance industry are on the eve of a breakthrough, and Ping An's goal is to stay half a step ahead."
Perhaps this is also the most pragmatic and intelligent posture this trillion-yuan financial giant can adopt in the AI wave.
Comments