Guide: An engineer who does not understand AI is not a good customer service
The AI large model has shown unprecedented creativity in the field of AIGC. It is considered to be an effective path for artificial intelligence to break through "cognitive intelligence". The development of the "intelligent brain".
Now, Inspur Information is trying to apply the "source" of the large model to the Inspur Information customer service platform to create an "intelligent customer service brain". An exemplary path.
Let AI become an IT expert
In the era of rapid development of cloud computing and artificial intelligence technology, Inspur Information achieved a rapid increase in server market share by virtue of the JDM (Joint Design Manufacture) model and forward-looking layout in the field of artificial intelligence, followed by "sweet troubles" ——The number of customers and service demand are increasing rapidly, and the non-standardized demand brought by the JDM model makes the service work more difficult.
In response to this situation, Inspur Information tries to achieve high-quality and sustainable development of service capabilities through digital and intelligent transformation. Among them, as the first window of service, the intelligent transformation of "customer service" has become the first hurdle.
Different from other industries, the hotline customer service of Inspur Information is not only an "operator" in the traditional sense, but also an "IT expert". They are the "center" of the entire service team, accepting customer service requests and understanding customers Problems, fault judgments, and solutions are provided; internally, customer needs are communicated, and the closed-loop service is completed in collaboration with scheduling, spare parts, and front-line engineers. Whether AI can have the professional technical ability and dialogue communication level of a customer service engineer has become a key breakthrough in "intelligence".
According to Zhang Yichuan, manager of Inspur Information Intelligent Service Department, for the development of intelligent customer service, Inspur Information initially adopted the industry's common FAQ question and answer, task-based dialogue and other models, and invited the expert customer service engineers of the call center to build and improve the standard question and answer library. "With the joint efforts of everyone, around 2021 our intelligent customer service can already answer about 60% of customers' questions, and the problem resolution rate will reach 50%." Zhang Yichuan said.
After achieving this achievement, Inspur Information continued to invest in the AI team and more than 40 expert engineers to continue to build multiple models and multiple knowledge bases, but the resolution rate of intelligent customer service was still hovering at 50%, and the marginal utility of the manpower and time invested was gradually increasing. Decrease, and the effect improvement encounters a bottleneck.
The capabilities of "customer service robots" that the public comes into contact with in life are often relatively simple, usually matching fixed answers to fixed questions, which is enough to meet the large number of procedural question-and-answer needs of most C-ends. However, the IT industry at the B-end is more knowledge-based services, showing the characteristics of high professionalism and high complexity. In Inspur Information, customers’ service needs often cover various issues such as product specifications, product usage, technical parameters, and fault repairs, and involve complex situations where multiple software, hardware, and different business scenarios and business processes intersect.
Picture: IT field has high barriers and low fault tolerance rate, and requires higher professionalism and communication skills of customer service
Therefore, the intelligent customer service of Inspur Information must be an "IT knowledge expert" who can truly understand customer questions and give professional and effective answers.
For example, there may be five situations behind the failure of the customer to install the system. The intelligent customer service needs to gradually guide the customer to clarify the specific problem scenario and give a targeted solution. It is not easy for AI to become a qualified "intelligent customer service" - at Inspur Information, a real employee with basic IT knowledge needs at least 6 months to 1 year of on-the-job training to become a qualified customer service Engineers, let alone AI.
In order to break through the bottleneck, the Inspur information service team decided to try a new path: large model "source".
The road to innovation: "Yuan" large-scale model to create an intelligent brain
The "source" large model is one of the world's leading AI large models, with 245.7 billion parameters and stronger general intelligence capabilities. With its super small-sample and zero-sample learning capabilities, "source" can be used as an algorithm infrastructure to generalize to a variety of application scenarios, effectively alleviating the dilemma of repeated modeling in fragmented development.
In order to better integrate the "Yuan" large model with the extremely professional and vertical data center service scene, the "Yuan" team joined hands with the Inspur information expert service team to successively compile more than 20,000 product documents, user manuals, and more than one million pieces of expert information. The engineer's service dialogue, hundreds of thousands of customer service logs, and work order data are used as the knowledge base to feed the "source" for learning, and it takes 6 months to deeply analyze the complex service business process, combined with knowledge distillation, compression and other technologies, based on "Yuan" has constructed the "intelligent customer service brain" of Inspur Information.
Wu Shaohua, director of AI software R&D at Inspur Information, compared "Yuan" to a "doctoral student" with strong learning ability, saying that it has a stronger level of intelligence and can complete learning tasks independently and quickly. This not only means that its training no longer requires a lot of manpower, but also means that the intelligent customer service brain it builds is no longer mechanically completing QA matching, but is capable of "deep thinking" and knowledge based on its own strong language understanding ability. Learn and refactor.
Take the scenario where a customer consults on the memory configuration of a server as an example. In the traditional training mode, engineers input question-and-answer databases of different product models one by one, and train robots to point to a best-matching answer in the question-and-answer database based on keywords of user questions. The "source" large model is based on the full text of product documents in the learning process. Combined with powerful contextual semantic understanding and analysis capabilities, it can more accurately understand customer problems, find related knowledge content, and then generate according to the powerful semantics. Extract key information and generate concise and professional answers.
At present, with the support of "Source", Inspur Information's intelligent customer service brain not only improves the natural language interactive service capability, but also intelligently manages and dispatches resources such as intelligent operation and maintenance and diagnosis of authorized IT equipment, expert engineers, and spare parts. In terms of capabilities, it has become the core "brain" supporting the intelligent transformation of the whole process of Inspur's information services. At present, this "expert data center intelligent customer service" has been named "Yuanxiaofu" and put into use.
Figure: Inspur information intelligent customer service gradually clarifies the problem and gives an accurate solution
"Yuanxiaofu" currently covers 92% of the problems related to Inspur Information's 8 major product lines, and solves 80% of the common technical problems in the data center, which can reduce the business processing time of complex technical consulting problems by 65%. Improve the overall service efficiency of Inspur Information by 160%.
In addition, the unsupervised self-learning feature of "Yuanxiaofu" also liberates the customer service engineers in the call center from the original extra work of filling FAQs and knowledge graphs, and improves the service team's human efficiency by 30%, helping Inspur Information The service team realized a leap in work efficiency.
Large-scale model landed, breaking through the deep water area of digital intelligence transformation
The success of Inspur Information's "intelligent customer service brain" has opened up a new path for the "intelligent change" of customer service. It not only proves that with the blessing of large models, intelligent customer service is no longer the "artificial mental retardation" ridiculed by the public. It also provides an exemplary path for large models to cross the gap between AI technology and complex industry scenarios.
Today, with the vigorous development of large models, it is believed that there will be more industry scenarios using large models as intelligent engines to break through the deep water area of digital and intelligent transformation.
END