The 20th National Congress of the Communist Party of China pointed out that "use the construction of digital China to help Chinese-style modernization, and accelerate the construction of a network power and digital China." Interaction and virtual reality technology research. In recent years, major banks have launched their own digital employees (Digital Employees). Digital employees can provide service support and execute business processes like human employees, with higher efficiency and less error-prone. On the one hand, its promotion is to improve efficiency, improve customer experience, and strengthen risk management. On the other hand, it also follows technological trends, enhances brand image, and optimizes the talent system.
01Introduction
to digital employees
Digital employees are a way of digital transformation of enterprises. Its working force built through software, algorithms or other forms of automation tools can perform tasks performed by human employees, or provide customers with a variety of information and data retrieval. Application scenarios include information retrieval, report generation, task processing, data analysis, etc. . Digital employees can help customers improve work efficiency, reduce human errors, reduce costs, and be able to work 7*24.
Artificial Intelligence (AI) and Machine Learning (ML) technologies provide further possibilities for the development of digital employees. For example, chatbots or virtual assistants are common examples of digital workers that can handle a high volume of customer inquiries, provide answers to questions, or perform basic tasks.
In addition, Robotic Process Automation (RPA) technology is also widely used to build digital employees. This technology can automate some highly repetitive and well-defined workflows, such as information entry and report calculation.
However, while digital workers in particular have significant advantages, there are also potential risks and challenges. Data privacy and algorithmic bias are also concerns for digital workers who rely on machine learning and artificial intelligence.
02
The application of digital employees in commercial banks
In terms of commercial banks, the Industrial and Commercial Bank of China promotes digital transformation and development around the digital ICBC "D-ICBC", focusing on breakthroughs in anthropomorphic artificial intelligence technology, and endowing digital employees with independent personal identities; China Construction Bank in 2019 Since 2010, technologies such as RPA have been introduced to build a digital employee operation service platform; Industrial Bank has built smart digital robots, and the 24-hour smiling service ensures convenient and direct banking services at any time.
Although digital employees are developing rapidly in commercial banks, the technology maturity in the field of operation and maintenance is relatively low. In summary, the main problems are as follows:
1. User acceptance. Users are skeptical or resistant to digital employees in specific scenarios, and prefer to interact with human employees, hoping to express their dialogue intentions more directly in a colloquial way.
2. Business complexity. Operation and maintenance scenarios often involve multiple applications, multiple systems, or complex decision-making processes, and the business complexity of one-stop solution to user demands in the digital employee dialog box is even higher.
3. Production safety. Behind the digital employees is a large amount of user data, which involves privacy and security issues. Compared with other business systems, it needs to be matched with stricter data security and business continuity requirements.
03Agricultural
Bank of China's digital employee construction ideas in the field of operation and maintenance
Proceeding from the above problems, the Agricultural Bank of China launched an intelligent operation and maintenance robot "Operation and Maintenance Yidiantong" (hereinafter referred to as "Yidiantong") to provide knowledge consumption in the field of operation and maintenance for the technology line users of the head office and branches , information query, task handling and operational services. The following will mainly focus on the design and application of intelligent operation and maintenance robots in the field of operation and maintenance of the Agricultural Bank of China.
1. AI empowerment builds the foundation of digital employee service capabilities
The application of artificial intelligence technology enables the digital staff of the Agricultural Bank of China - Operation and Maintenance Yitong to process a large amount of data, queries and transactions more quickly and accurately, and provide users with a service experience that is close to human emotions to the greatest extent. Since 2022, the number of Yidiantong service users has reached more than 6,500, including 5,200 in the head office and 1,300 in branches, with an average monthly question-and-answer round of more than 30,000, with a hit rate of 95%.
(1) Natural language processing
Yitong's dialogue ability technology base uses the Agricultural Bank of China's intelligent dialogue robot system. The latter conducts information interaction with the application system in the industry through a unified intelligent question-and-answer dialogue service interface. After the consumer system applies for the resource environment provided by the intelligent dialogue robot system, it first needs to prepare the corpus of question-and-answer knowledge in this business field, and conduct it on the platform. Model parameter tuning ensures that the robot has a certain natural language recognition ability for user questions.
(2) Intelligent search engine
After nearly 2 months of promotion of the intelligent robot-based dialogue system, user input logs were analyzed. Contrary to expectations, users will not communicate with robots in pure natural language as they do with Siri and Xiaoai. There are two reasons. First, the target user group is the employees of the Agricultural Bank of China. This target group often uses this digital employee during office hours and when there is a need for operation and maintenance; Mainly, there is no voice-activated voice channel, and user questions are manually input by keyboard. Most of the user's questions are phrases and specific character strings, and rarely have a complete sentence structure of subject-verb-object.
Based on the above characteristics, the intelligent search ability is greatly improved on the basis of AI dialogue. First, the search results are sorted by relevance by analyzing query words, context and historical data; second, when users make continuous queries, the search engine remembers the previous The content of a query, with contextualized personalization of results.
(3) Intelligent recommendation system
In order to better promote products, improve users' personalized experience and office efficiency, building an intelligent recommendation system can effectively enhance the value and reputation of digital employees. According to information such as the user's position, organization, query history, behavior, and context, Yidiantong regularly recommends functions to technology line users of different functions or organizations through the Agricultural Bank of China unified communication software client point-to-point; secondly, regularly analyzes user logs, Make targeted recommendations to a user group that uses less common functions; the third is to recommend specific functions to specific users during special periods such as change windows and regular work meetings.
2. Scenario-driven and sound digital employee service capability system
This operation and maintenance digital employee is not limited by time and place, and can provide customers with timely support and solutions regardless of working hours or non-working hours. This makes its positioning different from intelligent question and answer and intelligent search engines, not limited to "checking", but can focus on "executing tasks", greatly improving work efficiency and user experience for users.
(1) Smart emergency
Digital employees can effectively play the value of information dissemination and decision-making assistance in emergency consultation scenarios. When an emergency event occurs in the industry, Yidiantong, as the medium of information transmission between the emergency workstation and Changchat, can broadcast and privately broadcast emergency information to Changchat emergency conference groups and related individual users, including emergency progress, diagnosis results, and related information. Changes, associated system performance indicators, etc. The broadcast of the conference group will regularly notify the relevant system of the progress of the disposal, and the point-to-point private broadcast can be targeted to remind the person in charge of the system, the shift leader, and the shift manager of relevant information, which can not only avoid missing important information, but also facilitate users to inquire about emergencies at any time. Related Information.
According to the emergency management requirements of the Agricultural Bank of China, in the current emergency consultation scenario, digital employees can provide real-time data analysis and decision support, and have been well connected and cooperated with the emergency personnel involved in the development process.
(2) Smart reminder
Smart reminder means that digital employees provide users with timely work reminders, tips or warnings in an intelligent way. These reminders can be personalized and precise based on information such as user behavior, preferences, context or system status.
Based on the work characteristics of the operation and maintenance field, Yidiantong takes the change progress reminder as a pilot, pushes the change window change implementation progress to the management users of the development and operation and maintenance departments in the industry in real time, regularly broadcasts the change implementation progress of each department, and provides relevant information on failure and overdue changes . At the same time, combined with the RPA automation process arrangement, the full-screen screenshot of the change implementation view on the webpage is pushed and pushed to the user, so that the user can directly see the detailed change implementation view screenshot on the webpage through the mobile APP.
(3) Smart Approval
Digital employee smart approval refers to the use of intelligent technology and digital tools to optimize the approval process of traditional processes or tasks. Relying on the characteristics of Yidiantong, the pilot project of the transactional process of the operation and maintenance platform of the Agricultural Bank of China has realized the conversational approval process. For some processes with simple approval process and high processing frequency, conversational approval is a service that does not interrupt the work rhythm of others and improves work efficiency.
In order to ensure that a single person can approve multiple processes on the client terminal, support users to "quote and reply" in one click (see the picture above), which can achieve the effect of approving multiple processes on the communication software client. Working hours can not only improve user experience, but also improve work efficiency.
Since the interface of conversational approval is a chat process with a robot, it is currently not suitable for dealing with complex processes. If the process requires the user to fill in multiple parameters, or the process needs to upload files, it is still recommended to give priority to the processing of the web page.
(4) Smart operation
With the continuous development and improvement of the instrumentalization capability of the operation and maintenance platform of the Agricultural Bank of China, the access of various operation and maintenance automation tools and platform use support, the initial human support model has become unsustainable. Yidiantong transfers this part of the workload to digital employees as much as possible. Users only need to send some simple instruction requests to digital employees, and then they can receive tool access guides, required menu entries, and core indicator detailed reports in the dialog box; You can also receive sensitive operation and maintenance operation information such as sensitive log analysis results that need to be paid attention to, self-service query frequency, etc.
3. Strictly control digital employee service application risks in safe production
It has been more than two years since Yidiantong was launched. In terms of non-functional aspects, it follows the design principles of "authority isolation, data desensitization, continuous monitoring, and security audit" to carry out digital employee construction to protect the operation and maintenance data of the Agricultural Bank of China to the maximum extent and avoid production safety risks.
In terms of data security, firstly, in terms of authority isolation, by classifying user groups and distinguishing service scenarios, authorized user groups can obtain services, and users without authority will enter the conventional question-and-answer logic when entering the same command; secondly, for digital employees The search results returned across multiple platforms have been desensitized, such as user personal information, production account information, and log sensitive information.
At the same time, from the perspective of business continuity, firstly, all channels of the digital employee Yitong are equipped with self-monitoring, such as web page access monitoring, log keyword monitoring, and the application client is deployed on the robot-customized client of the Agricultural Bank of China communication software , monitor by monitoring the heartbeat of the client; the second is to save all the dialogue logs of digital employees in various channels in the local database, and carry out regular self-examination, immediate investigation and immediate correction.
4. The practical effect of the application of the Agricultural Bank of China
Since the operation and maintenance Yidiantong was promoted at the end of 2021, various channels have provided dialogue services for more than 140,000 times. From 22 to 23 years, the cumulative consumption of the head office users was 130,000 times, the cumulative number of users was 6,500, and the monthly active users were 3,000; in the second quarter of 2023, three rounds of promotions were carried out for branch technology line users. Currently, the cumulative consumption is 10,000 times, and the cumulative number of users 600 people. In terms of channels, a total of 50,000 consumptions have been made through the web page of the integrated production operation and maintenance platform, and 70,000 consumptions have been accumulated through the application client.
In terms of query preferences, querying operation and maintenance tool configuration information, querying application configuration information, and querying partition information are the three most popular scenarios for users, and the cumulative consumption times are 8000+, 7000+, and 5000+ times respectively. In terms of scene services, it has provided auxiliary decision-making functions for nearly 30 emergency incidents; since the change reminder was launched in the second quarter, it has provided a total of 330 private broadcasts of change progress and abnormal change queries for more than 10 change windows, a total of 29 development departments Services; conversational process approval Since it went online in April this year, it has implemented a pilot promotion in the application process of application system operation and maintenance tools without access, and has now approved 300 rounds of this process through dialogue; Operational services such as operation and maintenance briefings and departmental alarm details have been launched for more than a year, and 3,000 services have been provided to specific users such as the exclusive operation and maintenance teams of various departments.
In terms of application effects, from the initial pure NLP knowledge search to the later combination of "AI + search engine", the accuracy rate of digital employee dialogue has increased from 61% to about 95% now. At the same time, in order to prevent users from being too simplistic in use, Yidiantong background will accurately deliver promotional copy to some users according to their preferences, so as to ensure that the target user group will use various core functions. At present, the active users of Yidiantong’s various channels account for about 15% of the users of the R&D center. In the future, it will further carry out precise marketing in batches according to the user’s department and user rank, and continue to improve the service efficiency of target users without disturbing the user’s work.
04
Summary and Prospect
As a digital employee serving the Agricultural Bank of China's technology line, Yidiantong has high accuracy, high response rate and high efficiency as the overall goal of serving employees. Using the NLP engine as the basic ability of the robot, it can identify the basic problems raised in the field of operation and maintenance, and at the same time increase the agricultural bank's R&D support knowledge base and work order database to expand the corpus to ensure that it can understand the professional knowledge of the agricultural bank's operation and maintenance instrumentation field; Continuously optimize the search engine, regularly review user logs, analyze the user's demand for operation and maintenance information in the office process, and realize various search modes; operating services. During the construction process, great attention was paid to the production and audit issues that may be caused by the promotion of innovative products, and the non-functional construction was continuously improved based on the design principles of authority isolation, data desensitization, continuous monitoring, and security audit.
With the technological breakthrough brought by the current big language model in the field of artificial intelligence, the development of digital employees is bound to rise to a new stage in the near future. First of all, the large language model provides digital employees with more powerful natural language processing capabilities. It has better semantic understanding and contextual grasp, and can more accurately understand users' intentions and needs. For example, commonly used operation and maintenance knowledge query, including querying production system operation instructions, fault diagnosis, etc.
Second, the powerful content generation capabilities of large language models bring more creativity and possibilities to digital employees, such as providing advice, giving solutions, or explaining complex concepts. Common operation and maintenance scenarios include automatic operation and maintenance document creation, fault solution generation, etc. In addition, the large language model also has the ability of knowledge fusion and transfer, and can learn and summarize valuable knowledge from massive data.
To sum up, with the promotion and application of large language models, digital employees will have stronger capabilities and broader application prospects in the future.
about the author
Li Xinyue, senior commissioner of the public service group of the operation and maintenance platform R&D team of the Shanghai R&D Department, leads the integrated production operation and maintenance platform-technical operation construction, and is responsible for coordinating platform technical operation services and intelligent operation and maintenance robots.
Qiu Liyuan, a commissioner of the public service group of the operation and maintenance platform R&D team of the Shanghai R&D Department, is responsible for the development and maintenance of intelligent operation and maintenance robots.
Xu Jiaqi, functional manager of the operation and maintenance platform team of Shanghai R&D Department, a veteran of integrated production operation and maintenance construction.