1. China's financial industry is moving beyond experimental AI pilot projects, with banks now deploying the technology to automate routine operations, accelerate research, and build systems capable of executing complex tasks. [para. 1]
2. However, broader adoption presents difficult questions for financial institutions, including whether productivity gains justify costs, how to integrate AI with fragmented data systems, and how to defend against increasingly sophisticated cyberattacks. [para. 2]
3. Reported usage figures illustrate the shift: Bank of Jiangsu said AI applications have cumulatively replaced 1.2 million working hours; Shanghai Pudong Development Bank said its digital workforce handled workloads equivalent to more than 2,500 people annually; and Bank of China's systems can recognize 305 types of documents across 3,800 business scenarios. [para. 3]
4. The most effective current applications involve repetitive, rules-based tasks or those dependent on processing large volumes of information, including document handling, customer service, payment-fraud detection, and preliminary credit assessments. Bank of Jiangsu reported that AI reduced the time required to approve credit for small and midsize enterprises by 30%. [para. 4][para. 5]
5. Investment research presents a more complicated challenge. While AI can collect information, analyze financial statements, screen securities, and produce reports—reducing time analysts spend on preliminary work—it cannot necessarily make reliable investment judgments. Financial insiders said AI-generated research may serve as a reference but is not automatically suitable for external publication or decisions involving significant financial risk. Human analysts remain responsible for interpreting results and exercising judgment. [para. 6][para. 7]
6. The next phase of adoption involves AI agents capable of performing several connected tasks rather than responding to a single prompt. For instance, an agent could retrieve data, compare companies, screen investment products, and assemble an initial report. Citic Securities has developed proprietary AI for ETF screening, stock analysis, and financial statement analysis. Moonshot's Kimi has been deployed at institutions including ICBC, CICC, Citic Securities, and E Fund Management. [para. 8][para. 9][para. 10]
7. These applications suggest AI could become embedded in workflows, but agents are only as dependable as the information and processes available to them. Institutions need accurate, current, and traceable data, plus clearly defined rules governing what an agent can access and do. [para. 10][para. 11]
8. Building such capabilities is expensive, particularly for institutions wanting to keep sensitive data within their own systems. According to a source at Tencent's WorkBuddy platform, a single private AI project can cost several million to tens of millions of yuan, excluding ongoing maintenance. Private systems offer greater control but may take around three months to update, whereas SaaS solutions can be updated in roughly two days, though with security and data-management implications. [para. 12][para. 13]
9. A large mutual fund executive said his institution deployed a customized agent "mainly for data security considerations, but currently it's not as user-friendly as the SaaS version." [para. 14]
10. Fragmented data systems and poor coordination between technical teams and business units hinder integration. An AI technology executive at a joint-stock bank noted the lack of a bridge between business and technology teams: business staff don't understand the technology, while technical staff don't fully understand business needs. [para. 15][para. 16]
11. AI capabilities also empower criminals, automating reconnaissance, identifying vulnerabilities, and generating convincing phishing messages. The Bank for International Settlements estimated AI could reduce the cost of simple cyberattacks to around $50–$100 and more sophisticated operations to $5,000–$10,000. [para. 17][para. 18]
12. This creates a widening challenge: as AI agents gain permissions to retrieve data and execute tasks, mistakes or compromised credentials could have consequences beyond inaccurate answers. Fan Yuan, chairman of cybersecurity firm DBAPPSecurity, said future cybersecurity battles will be fought increasingly between AI systems, requiring the industry to rethink security governance for the AI era. [para. 19][para. 20]
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新