Summary of Key Points
Postal Savings Bank of China (PSBC) has accelerated its adoption of large-scale AI models: In just over half a year, the number of scenarios where these models are being used has increased from 260 to 370, with the daily average number of calls to the models rising from 6 million to over 10 million. The daily volume of data processed (in terms of tokens, the smallest units of language that AI can recognize) has also tripled. By collaborating with Huawei on the “Youzhi” financial large-scale AI model, the bank has significantly reduced the cost and time required to implement AI applications. The “AI2ALL” ecosystem has been established to improve operational efficiency internally and to empower retail and corporate services externally, leading to actual business growth. Additionally, the bank’s financial reports show strong performance, with both revenue and profit increasing, with non-interest income growing at a much faster rate than interest income.
Detailed Analysis
How Fast Are the Large-Scale AI Models Being Adopted?
The expansion of PSBC’s large-scale AI models is evident: in its 2025 annual report, only 260 scenarios were using these models, but now that number has risen to 370 (an increase of 110 in just over half a year). The daily average number of calls to the models has increased by more than 60%, and the daily volume of processed tokens has increased by threefold, from “over tens of billions” to “over 30 billion.” This indicates that AI has penetrated into an increasingly large number of business processes within the bank, with the amount of information and tasks processed each day growing exponentially.
How to Make AI Both Inexpensive and Effective?
The “Youzhi” large-scale AI model, developed in collaboration with Huawei, has reduced costs through three key technical improvements:
- Reconstructing the model architecture: Making the model more efficient, reducing the time required to deploy it from two weeks to just one day.
- Compressing the parameter size: Reducing the amount of storage space the model occupies, with cache usage decreasing by 72% (saving nearly three-quarters of the memory).
- Optimizing the inference framework: Making AI more resource-efficient, with inference costs reduced by up to 87%. For example, what used to cost 100 yuan to process 100 tasks now only costs 13 yuan. These improvements have made AI more affordable for the bank, allowing it to deploy it in more scenarios with confidence.
How Does AI Help the Bank Attract Customers and Generate Business?
PSBC’s “AI2ALL” ecosystem provides mutual benefits:
- Retail segment: The “Retail Marketing Intelligent Agent” provides personalized recommendations to customers, reaching 54 million high-potential clients in the first half of the year (for example, suggesting suitable financial products or loans based on their spending habits).
- Corporate segment: The “Intelligent Industrial Chain Business Development Assistant” helps the bank identify corporate clients and conduct business analysis, assisting in opening over 6,000 corporate accounts and granting credit worth nearly 310 billion yuan. AI acts as a highly effective and precise “business agent.”
How Does AI Improve Internal Operations?
AI has been deeply integrated into various internal processes:
- Document processing: The intelligent recognition of documents in the operations center has replaced manual entry in over 90% of cases, eliminating the need for manual data entry.
- Intelligent Q&A: The “Little Post Assistant” can answer 98% of internal and customer inquiries, reducing the burden on human customer service staff.
- Risk control: AI models for preventing fraud have improved the efficiency of identifying risky accounts by nearly 30%, helping to detect fraudulent accounts more quickly.
- Credit due diligence: The efficiency of credit investigations has increased by over 50% due to the intelligent system, completing what used to take two days in just one day. These changes have streamlined internal operations, saving manpower and increasing productivity.
Is There Evidence of AI’s Impact in the Financial Reports?
PSBC’s semi-annual report shows positive results: revenue increased by 7.26% to 192.482 billion yuan, and net profit increased by 4.57% to 51.671 billion yuan. Notably, non-interest income (such as service fees and handling charges) grew by 12.25%, more than twice the growth rate of interest income (5.82%). This growth may be attributed to AI-driven services, such as intelligent marketing that has led to more financial products and loans, as well as AI-assisted services that have reduced costs, ultimately driving the increase in non-interest income.
Overall, PSBC is using large-scale AI models to transform its business model. By making internal operations more efficient and providing precise services to customers, the bank is also controlling costs through technological improvements. These factors could become new drivers of future growth.