How is the AI model of the six major state-owned banks progressing, and what applications have been explored?

Author | Ho Yat-chan

Curated | Luo Yanshan

Against the backdrop of declining interest rates, narrowing interest margins, and slowing revenue growth, it has become an urgent need for the banking industry to find new growth points and improve operational efficiency.

According to the 2024 Report on Generative AI Applications in the Financial Industry, jointly compiled by Tsinghua University's School of Economics and Management, Du Xiaoman, and MIT Technology Review China, although the mainland financial industry has the world's largest real-time data, these financial data alone cannot bring business value simultaneously. By building a large AI model in vertical fields, we can not only give full play to these data resources, but also drive the innovation and development of financial technology.

Under this transformation demand, the logic of digital transformation of banks is gradually clarified as "data + algorithm", in which AI large models have become the key to maximizing the value of data and promoting business innovation. In the past year, the mainland's six major state-owned banks and a number of leading commercial banks have stepped into this field in order to upgrade the value chain through new digital means and remain competitive in the financial market.

How are AI models in the banking industry evolving

A few days ago, the six major banks in mainland China all disclosed the progress related to the large model in their 2023 annual reports. Among them, the Industrial and Commercial Bank of China (ICBC) took the lead in building a full-stack, autonomous and controllable 100-billion-level AI large model technology system, and its AI large model construction achievements were awarded the first place in the "10 Major Events of Financial Informatization in 2023" by the People's Bank of China's "Financial Electronics".

At the beginning of 2023, the Agricultural Bank of China released ChatABC, a large financial AI model application, and opened it for trial to internal employees in the form of multiple rounds of Q&A assistants and automated work order response assistants.

Bank of China is exploring the application of large model technology in internal knowledge services, assisted coding and other scenarios, and using artificial intelligence, big data and other information technologies to improve credit risk assessment capabilities.

China Construction Bank launched the large-scale model "Ark Project", deeply cultivating its professional capabilities in five major fields, including computer vision, intelligent speech, natural language processing, knowledge graph, and intelligent decision-making, and its self-developed AI platform has accumulated 43.3 billion service calls, and won the "Best AI Application" by The Asian Banker 2023.

Bank of Communications has deepened the application of AI in customer service, product promotion, risk prevention and control, and explored the application of large models in scenarios such as office assistants and customer service Q&A, and listed "building a generative AI framework with embedded risk control elements" as a priority for 2024.

The Postal Savings Bank of China has built a "Postal Savings Brain" that integrates large model technology, exploring the application of large model technology from the directions of text generation, code generation, text refining and multimodal understanding generation, and submitted more than 5 patent applications in the field of large models in 2023.

In addition to the six major state-owned banks, national joint-stock banks such as China Merchants Bank, Ping An Bank, and Industrial Bank, as well as many urban commercial banks such as Bank of Beijing, Bank of Shanghai, and Bank of Jiangsu, have also mentioned the R&D and application results of AI large models in their annual reports. AI models are gradually becoming the second competition point in the development of banks.

What banks do with AI models

As Gao Feng, CIO of the China Banking Association, said: "Without application scenarios, new technologies are 'rootless trees'." "The ultimate goal of the development of AI models in the financial vertical field is still to serve the operation and management scenarios. At present, the application of large banking models can be divided into two major purposes: internal operation management and reshaping external business scenarios.

(1) Oriented to internal operation management

In terms of reducing the burden on internal staff, PSBC focused on R&D, testing and incubation of "R&D assistants" to assist in the whole R&D process such as demand analysis, UI design, code generation, and system testing, so as to promote end-to-end R&D efficiency.

CCB continued to build financial image text recognition products, supporting the recognition of more than 140 types of bills, covering 75% of the bill recognition volume, helping to improve the efficiency of bill review information entry by 120 times, and won the championship of the seal text detection track in the global artificial intelligence document image analysis and recognition competition (ICDAR 2023).

In addition, CCB's "Ark" assistant and "Ark" toolbox can also realize 25 scenario functions, such as quickly generating research report summaries and comments, entering voice to automatically generate visit records, Wensheng diagrams, and automatically generating customer survey reports of listed companies, so as to comprehensively improve the professionalism and efficiency of employees.

(2) Reshape external business scenarios

Smart Marketing

Postal Savings Bank pays attention to customer acquisition capabilities, and launches emotional model conversation insights and "smart think tank" services to enhance the operational function of WeCom and improve the ability of refined customer insight at the grassroots level.

Bank of Communications uses AI technology to dig deep into the interests and preferences of individual finance customers, and uses large models to strengthen the ability to retain customers on the business side, and the cumulative customer transaction volume of various financial management model strategies is nearly 400 billion yuan, which is 16 times higher than the overall transaction rate.

CCB realizes personalized speech AI synthesis, supports 100,000-word ultra-long text speech synthesis, supports voice broadcasts such as CCB News and WeChat Work, and realizes the automatic production of marketing creative content and copywriting, helping to build brand image, improve the quality of marketing content, and improve the ability to stick to customers.

Intelligent customer service

The emergence of intelligent customer service is to make up for the shortcomings of traditional manual services, but its effect of reducing costs and increasing efficiency is far from expected, especially the intelligent customer service model under RNN (recurrent neural network) technology has great defects in understanding customer problems, locating key knowledge points, and matching knowledge base problems. In this context, the application of large model technology to intelligent customer service is like installing a "brain" on the customer service digital human, which has become the only choice for more and more banks to improve the intelligent level of customer service.

In terms of online intelligent services, the "Postal Savings Brain" integrates large model technology to build new generative AI capabilities and accelerate the reshaping of the digital financial service model. In addition, the PSBC App integrates AI space, digital staff, and video customer service to create an immersive companion service.

ICBC accelerated the application of new technologies such as digital humans and large models in the operation field, and officially launched the first intelligent assistant for branch employees based on large models to improve the efficiency of its branches, with 320 million intelligent processing transactions in the operation field throughout the year, an increase of 14% over the previous year, and built 13 comprehensive digital employees and more than 1,000 digital employees for process automation, increasing the efficiency of intelligence by more than 30,000 people.

In terms of intelligent telephone customer service, CCB has independently developed end-to-end speech recognition and speech synthesis capabilities, realizing speaker identity voiceprint recognition, speech recognition in dialects such as Sichuan dialect, and audio quality detection capabilities, and supporting scenarios such as intelligent outbound calls.

ICBC uses a general AI model in the financial industry to support intelligent customer service to answer customer calls, significantly improving the accuracy of identifying customers' call demands and emotions, and greatly reducing maintenance costs.

ChatABC, a financial AI large-scale model product launched by ABC, can also use large-scale model technology to improve the financial knowledge understanding, content generation and security Q&A capabilities of intelligent customer service.

In addition, CCB also noticed that the traditional customer service manual filling and proofreading work orders were time-consuming, labor-intensive, and inefficient, as well as the problem of non-standard work order filling that led to repeated communication and affected the customer experience, so CCB launched the intelligent customer service work order generation function in its basic application of the financial model, saving customer service working time by an average of 15-20 seconds per order, with an availability rate of 82% and a consistency of 80%, and the project won the championship of other tracks in the 2023 Customer Service and Remote Banking Innovation Application Competition of the China Banking Association.

Customer complaint management and consumer rights protection

Whether it is the physical industry or the financial industry, its pre-sales and after-sales services are ultimately to serve the customer experience, and customer complaints are an important feedback path for the customer experience. Managing customer complaints is not only to protect the rights and interests of consumers, but also to manage the bank's own operational risks.

ICBC promoted the intelligent governance model for customer complaints, comprehensively applied technologies such as robotic process automation, natural language processing, and generative artificial intelligence (AIGC) in the main aspects of complaint handling and management, and successfully implemented the first bank-wide AIGC scenario to realize the automatic writing of complaint handling reports by large models.

PSBC has developed an intelligent complaint classification model based on a large model to realize automatic statistical analysis and intelligent monitoring of consumer protection complaint management, and launched and promoted a text analysis model for consumer protection complaints and an intelligent auxiliary tool for consumer protection review, effectively improving the ability of pre-council review and post-event analysis of consumer protection trustees.

CCB launched the AI intelligent review function for consumer protection, which generates AI review results through intelligent information recognition and processing, assists reviewers, improves review efficiency, and enhances the standardization and professional ability of consumer protection review.

How the "Bank of the Future" uses AI

The application and development of large models in the financial industry in mainland China is in the policy dividend period, but most of the current bank large models are only used to empower employees to reduce the burden and improve customer experience, and most banks are still in the stage of technical reserves and shallow experiments, and it is difficult for AI large models to really detach themselves from "people" to play the effect of "AI+". With the in-depth development of vertical large model technology in the financial field, the application of large bank models will touch more of the core business of banks in the future. Many pioneers have explored risk management and investment decision-making.

There are three main effects that a large model may have to help build an intelligent risk control system. First, improve the degree of operation standardization and improve the efficiency of approval with automated processes. Second, it has built a large-scale model intelligent analysis system to quickly process massive financial information and improve risk assessment capabilities in the business process, such as the anti-fraud integrated platform built by the Bank of Communications, which accurately intercepted more than 70,000 suspicious transactions, involving an amount of more than 1.4 billion yuan. Third, it reduces the operational risk of manual operation and strengthens the level of compliance management, such as the Postal Savings Bank's use of intelligent risk control "intelligent review assistant" to assist in legal review compliance. The prospect of large models to help banks in risk control is promising, but the compliance risks and data security risks inherent in large model technology are not clear, and the corresponding regulatory framework and industry standards have yet to be established.

The application of the bank model in the investment scenario has just taken the first step, due to the high professional requirements and experience requirements of this business scenario, the intelligent investment research assistant is currently more used for repetitive work such as sorting out investment research reports and processing transaction data, and it is not yet possible to generate customized professional investment suggestions for different objects. Before the comprehensive and large-scale development of large-scale models in the financial field, investment research assistants are unlikely to replace professionals such as wealth planners, and it is even difficult to form pressure on them to "drive out good money".

KPMG's 2024 China Banking Outlook Report believes that the emergence of large models will catalyze the iterative development of "future banks", and Agent-based productivity tools are indispensable atomic modules in the next generation of large model application systems.

Original link:

How is the AI model of the six major state-owned banks progressing, and what applications have been explored?_Enterprise News_He Yican_InfoQ Selected articles