Ping An Scales AI Strategy to Drive Operational Efficiency

The group’s AI-powered Express Service has reached a milestone of 100 million user interactions since its debut four months ago. By utilizing natural language processing for transactions and claims, the platform achieves a 92% end-to-end resolution rate, effectively cutting traditional processing times in half. Simultaneously, the company’s AI Doctor now handles a quarter of all in-hospital initial contacts, with diagnostic assessments for breast and gastric cancer matching expert conclusions in 91% of observed cases.

To support this expansion, Ping An doubled its computing capacity to 1,500 PF, facilitating a tenfold increase in token consumption. Through full-stack optimization of inference engines and resource scheduling, the group improved token output per unit of computing power by 368% and reduced costs per token by 76%. Internal productivity has also seen a sharp rise, as AI-generated code now accounts for roughly 80% of total software output, while AI-driven fraud detection contributed RMB 7.11 billion in claims savings for the property and casualty division during the first half of the year.