Absrtact: in the age of intelligence, how can intelligent customer service and intelligent assistant really assume the name of intelligence? Huawei is rectifying its name in the name of AI.
In the driving, hands and eyes need to be highly concentrated, “please tell me how the road is today?” “Please tell me which way will get there faster?” Intelligent assistant will be the most reliable navigator around.
In the work scene, the guidance of old employees may be very important, but they can’t be the bodyguards of novices. “How can I repair this file?” “How to operate the financial process?” Intelligent customer service will be a more reliable and efficient means.
Obviously, whether consumers or ordinary users, intelligent customer service and personal assistant information system has been inseparable from life. Wisdom comes from data, continuous iteration of system and continuous updating of enterprise system. If the user gains a series of wrong answers and complicated operations in the process of using, not only the user experience is poor, the operation efficiency of internal and external linkage of the enterprise is limited, but also the brand image will be damaged. So in the age of intelligence, how can intelligent customer service and intelligent assistant really assume the name of intelligence? Huawei is rectifying its name in the name of AI.
Huawei’s R & D strength is obvious to all. In recent years, the global trend of intelligence has opened, and Huawei, standing at the forefront of the storm, has embarked on this historic process. In the process of the transformation of “intelligent Huawei”, it has launched a large number of intelligent means to improve the operation efficiency of enterprises. Intelligent customers and intelligent assistants are two of the most powerful tools. Huawei applies a large amount of experience and technology accumulated in the field of intelligence to the enterprise operation process in the form of intelligent customer service and intelligent assistant, so as to fully tap the internal technology and capability resources of the enterprise.
Which intelligent customer service can cross thousand mountains?
Before accessing intelligent customer service, one thing must be clear, that is, why intelligent customer service is needed? Are some of the customer service robots that we hear now intelligent customer service? Please take your time.
The purpose of setting up customer service is very simple, that is to solve the consulting problems of ordinary customers. At present, in the customer service industry, there are mainly two ways to deal with the general customer consultation, the traditional customer service robot and the telephone key navigation to manual customer service. However, the two have their own advantages and disadvantages. The traditional customer service robot has obvious problems, easy to answer, and the problem solving rate is less than 80%. At the same time, many online businesses can not be handled. However, the manual customer service process is lengthy and inefficient, usually waiting for the manual agent to submit the work order, upgrade the work order, close the work order, customer satisfaction evaluation and customer satisfaction evaluation Service quality inspection and many other links.
According to the data provided by Huawei’s customer service business, it currently has more than 10000 customer service agents, with 60 inquiry sheets per person per day, each of which takes about 10 minutes to process, with an annual loss rate of 130%, and a long training cycle. Robot customer service has low hit rate of dial test, low user participation rate, wrong answers and high cost of knowledge operation; manual agent queuing and work order upgrading are difficult, knowledge retrieval speed is slow, and the efficiency of manual summary is low So there’s no way to have the best of both worlds?
Obviously not. Huawei intelligent customer service is to solve this contradiction. In essence, the intelligent customer service solution needs to combine the two customer service businesses, and its overall process covers the following five steps.
· Intelligent navigation,In the navigation, AI intelligent diversion is realized according to the user’s portrait recognition to quickly understand the user’s intention, and some common problems are handled directly;
· Intelligent Q & AIn the aspect of robot customer service, based on NLP (natural language processing) technology, we have built all kinds of dialogue robots, so that users can do more self-service business;
· Seat assistance,In the manual agent service, automatic recommendation is realized in the aspects of business process, knowledge content, business script and promotion order classification, so as to improve the efficiency of customer service;
· Intelligent outbound call,In the user satisfaction evaluation part, according to the marketing clues to complete automatic mining, automatic completion of customer satisfaction return visit;
· Intelligent quality inspection,In the process of customer service quality inspection, it can identify the early warning of emotional abnormality in real time, detect the service quality and avoid sensitive information.
Through these five processes, Huawei intelligent customer service can realize that more than 85% of the user’s questions can be handled in a closed loop by the robot under the condition that no one dares to do so; and in the aspect of manual agents, it can achieve a thousand questions and answers in one minute to complete the recording, checking and upgrading of the order; and more importantly, the intelligent customer service system can realize 100% coverage of the whole question list to ensure that there are no errors and omissions The choice of beauty.
Intelligent customer service product architecture
What scenarios can intelligent customer service play a role in? Tang Juxiang, intelligent customer service product manager of AI enabling Department of Huawei, said that the intelligent customer service solution based on intelligent routing, Q & A and task engine, intention recognition and other capabilities has been widely used in Huawei to achieve one click access to the required services; and the technology has also been widely used in user scenarios, internal connection scenarios, and various service scenarios of enterprises.
Scenario application, the key to intelligent customer service
Past experience tells us that if a product is not competitive and can not solve the problem in the scene, then it is difficult to call it a success. Huawei’s intelligent customer service solution can effectively break through the difficulty of scenario development when facing ordinary users, external engineers, small languages and many other scenarios.
Huawei vmall mall online customer service is a typical application, which can handle all kinds of consumer consultation in pre-sales and after-sales. In 96 countries, 700 + terminal product models and 14 emui / Magic UI system versions can be served. In the past solutions, customer service products often face many problems, such as complex system, difficult to remember knowledge, business surge risk, high seat pressure, emotional collapse in extreme cases, rapid staff turnover, difficult overseas recruitment and so on.
After accessing the intelligent customer service, the efficiency of the overall solution has been greatly improved, and the RPA robot process has been used before sales Product consultation and comparison tasks can be completed by RPA robot, knowledge mapping robot and entity recognition; after sales service can provide accurate maintenance guidance to the store through task robot, product mapping robot and user portrait; at the same time, gradual fault resolution can be completed through RPA robot, fault mapping robot and entity recognition ability.
Engineers’ problems may be more complex than ordinary users. In the past, when engineers were carrying out function comparison, product selection, product display, installation guide, fault guidance and version recommendation, it was very difficult for them to solve problems through robot customers, but they needed $80 / order to ask for help from manual agents.
In order to facilitate enterprise business engineers to solve problems quickly, Huawei enterprise business department invested one person month to build 14 main scenarios of task multi round self-help, realizing customer engineers’ self-help to quickly obtain solutions to product related problems, alleviating the pressure of human agent, and achieving labor cost saving. The biggest feature of this solution is that it can be based on the enterprise business intention recognition model. It is a solution that combines the enterprise product map with the scene task engine.
Compared with the small language scene, it will be simpler. Huawei has realized the ability of small language and English robot by using patented technology. At present, it has covered Spanish, Thai, Russian, Japanese, Korean, Arabic, French, German and many other languages. In the first mock exam, it usually has advantages and disadvantages, and can only be applied to specific scenarios. The system has special models to deal with some key languages, and it can form complementary after merging.
Huawei’s intelligent customer service solution is oriented to core small languages and mainly adopts multi model fusion (classical machine learning algorithm + deep learning algorithm) + voting decision-making method to train in different languages; for non core small languages, it mainly uses machine translation, English BOT ability and multilingual homologous Association of answers to realize fast coverage of small languages. Different scenarios have different solutions, in order to solve the problem more efficiently.
In summary, Huawei’s core technology of intelligent customer service has been significantly improved at the operational level. It can improve the knowledge reserve through automatic document extraction and automatic knowledge generation, and complete machine self-learning according to dialogue log statistical reports, hot and high-frequency problem statistics, and artificial service data feedback, and then with the help of manual annotation, knowledge health check, automatic knowledge generation, and language learning Material expansion and other capabilities to improve the efficiency of intelligent customer service. At the same time, through automatic training and manual release, the knowledge coverage of the system can be improved, the overall accuracy rate can be increased by 10% to 20%, and the answer is more accurate.
Compared with the previous intelligent customer service solutions, Huawei intelligent customer service has stronger customer understanding ability, which can enhance customer understanding, improve customer experience and reduce customer service pressure through NLP ability, multimodal emotion recognition, portrait based recommendation and multilingual intention recognition; in addition, it can provide out of the box ability, and realize 20% customer service through cloud and multi tenant isolation At the same time, customer service solutions cover many scenarios, such as internal consulting, business assistant and so on, which make the products more competitive.
Intelligent assistant, a new mode of information acquisition
We all say that now we have entered the era of data, especially after the development of enterprises, the scope of business continues to expand, systems and data continue to increase, the traditional way of information access is no longer so effective. According to McKinsey’s survey, about 40% of employees’ time is spent on obtaining different information, which will directly affect their work output.
So what are the ways to get information? Liu Yufeng, senior product manager of AI enabling Department of Huawei, said that the way of information acquisition has experienced three times of changes. In the era of 1.0, information is mainly obtained through the menu click mode, and the user’s access to content is limited; in the era of 2.0, information acquisition mainly relies on traditional search, which can obtain massive content, but the problem is that its response mechanism is vague, and it is mainly based on function driven, and the result is one-sided. In the era of 3.0, the emergence of intelligent assistant makes the channels for users to obtain massive information more diversified. It can achieve accurate response, and obtain information based on the scene drive. It can achieve active recommendation, and the content is more comprehensive and reliable.
In general, the way of information acquisition has experienced the upgrading from “people looking for services” to “services looking for people”. In fact, this is also the characteristic of the next generation of digital services, from active access to information to passive access after information integration, so as to revitalize enterprise digital assets and improve employee productivity. As a way of information acquisition for the future, how to build intelligent assistant?
The construction of intelligent assistant AI
In the past, when making products, we often choose technology driven, but intelligent products, especially intelligent assistant products, are often driven by operation, scene and data. When building an intelligent assistant, it is often necessary to obtain user behavior logs first, and then analyze them based on user behavior habits, so as to ultimately achieve core capabilities such as scene recognition, service design, intention recognition design, etc.; at the same time, it can also be used for status assessment and goal setting.
After the construction is completed and online, the subsequent update iteration is called “PDCA” transformation (PDCA: plan, do implementation, check verification, action improvement). After the product goes online, the product needs to be verified repeatedly, and then feedback iteration is used to realize the closed-loop of user problems, and finally the user experience is improved step by step. The first version of intelligent products is often not perfect, but it is in this series of closed-loop upgrades that the service ability and user experience of products are constantly improving.
Huawei’s internal intelligent assistant is mainly based on Huawei’s technology accumulation for many years. According to different ports, there are three main presentation modes: voice assistant, intelligent search and word search (flash search). In the mobile terminal, voice assistant can answer questions conveniently; text search can cross the terminal and process all kinds of content information, while word search can be intelligently embedded in all kinds of software to present all kinds of information in real time. Huawei’s assistant center can provide AI capabilities such as qabot, RPA, search bot, talk to data, task BOT, etc. to help improve employee productivity.
Multi scene landing, new year of intelligent assistant
When it comes to intelligent assistant, many people’s cognition of it may still stay on personal mobile phones such as Xiaoyi and Siri, but it is obvious that there will be more space for its landing than these scenes. Huawei’s intelligent assistant performs in HR service, conference assistant, data service, knowledge service and other fields.
HR service is not easy. Different from other network search, it often faces stranger search, most of which belong to fuzzy search. The intelligent assistant supports the internal search of employees based on the employee map, and classifies and associates search according to the different tags of employees. The tag gets electric shock from the beginning of employment, including the whole career cycle of job transfer, promotion, business trip, training, project operation and resignation. From the perspective of Department, it can also be distinguished from colleagues in department, secretary, hrbp, supervisor and project team To find the right person more accurately.
But there seems to be a lot of simple meeting services. When participating in the meeting, a series of operations such as booking the meeting room and booking the participants are needed before the meeting, and the meeting room may not be found at the same time. Huawei conference assistant can reserve the conference room in one sentence and automatically reserve the time of participants through RPA. The built-in indoor navigation can also avoid the embarrassment of not finding the conference room.
In meetings, attendance statistics, stranger information query, cross language communication, meeting minutes and task distribution, each one may be a pit. Huawei conference assistant can provide such functions as automatic attendance statistics (camera), stranger information recommendation, real-time conference script, simultaneous interpretation, photo recognition, one sentence creation task, etc. At the end of the meeting, it is often necessary to sort out the minutes of the meeting and follow up the tasks. The meeting assistant can record the multimedia meeting, generate video, voice and oral draft, and the summary oral draft text is smooth, which can automatically extract views; after the meeting, it can also use RPA to automatically follow up tasks.
The demand for data services will be more complicated and the technical content will be higher. When looking for the right data content, the results are often too professional, leading to many front-line users unable to clarify the value and key of the data. Through the use of intelligent search, machine learning algorithm, intelligent assistant can accurately recognize the voice, analyze the user’s intention and dialogue with the user, so that the user can easily find the desired data.
For example, when the user wants the data of a certain area, the intelligent assistant can directly present the histogram of the relevant data, and can find a more appropriate visualization form according to the user’s instructions, so as to help the user understand the relevant data sets and problem points. Through the nl2sql (natural language to structured query language) technology of AI, the intelligent assistant realizes the transformation from natural language to machine language. Users only need the most natural language to combine data randomly, freely, multi dimensionally and logically. They can complete data exploration and analysis through one or more sessions to gain insight into the business nature behind the data.
Huawei intelligent assistant will provide more scene based help in knowledge service. The knowledge map based on product construction can complete the Q & a more accurately than the ordinary FAQ system, and realize the ability of product comparison, product FAQ support, attribute query and so on. The triple of product knowledge map itself is better defined, so the product attributes are relatively limited, and the data are mostly semi-structured data, which can be relatively simple to build; but the knowledge map of big concept will be more difficult, which needs to be built in the appropriate scene, and even needs to be solved by relevant logical judgment.
In the future, in-depth retrieval of product document content is expected to become a rigid demand. Only from Huawei’s internal point of view, the annual document volume can reach one million. In the past, it only supported employees to actively look for documents, but it took a lot of time to browse the contents of documents, and the efficiency of information acquisition was low. In the process of finding documents, due to the limited information covered by the document title, the search function often can not find or support excel, PPT, PDF, word and other file content search.
The intelligent assistant can parse the document, including page number, title, body and so on, and access the index library in the form of API. When users retrieve the relevant documents, the intelligent assistant calls the relevant content of the index library through intention understanding, and outputs the relevant content in real time through the intelligent search algorithm model. In summary, intelligent assistant can achieve deeper content search, accurate positioning, online preview and instant search.
Schematic diagram of core semantic identification service architecture
Core semantic recognition is an important difficulty of intelligent assistant. When doing Knowledge Q & A or search, we often encounter a problem. When users input a sentence, the importance of words is different. In the past, bad case caused by unrecognized core semantics has reached 35%. There are many solutions in the past, such as a large number of dictionaries or labels; but the problem is that the labor cost is too large, and the effect of case by case maintenance is not good.
Huawei intelligent assistant is based on the core word recognition model of unsupervised algorithm. When the user inputs a sentence, the algorithm model will automatically analyze the weight of multiple words in the language. Through the core semantic recognition, the user’s Q & A results can be more accurate, and the customer’s demands can be better understood.
In addition, in many cases, the user will search for some content with unclear intention, and the search topic will be more extensive; or the user’s memory is not clear when searching; moreover, when the user enters a tag, the content under the tag is too much, which leads to the user at a loss. In order to solve the problem of searching with unclear intention, Huawei intelligent assistant forms a topic map based on the relevant information of Huawei intranet, and draws a map according to the relevance of the relevant knowledge content under different topics, so as to guide users to make clear their intention.
When users explore knowledge, they can get it through the key knowledge extraction module under the topic; when the user’s input is fuzzy, the topic map can gradually guide the user’s related content; when there is more content, the topic recommendation ability created by the structured processing module and the query processing module can help users.
In general, the efficiency of traditional Q & a system is relatively low. Through intelligent upgrading and digital transformation, Huawei’s intelligent customer service and intelligent assistant can jointly help enterprises revitalize their assets, improve the overall operation efficiency, and give customers more high-quality service help. It can be said that intelligent assistant and intelligent customer service have become the “vanguard” of Huawei’s intelligent transformation and upgrading.
Click follow to learn about Huawei’s new cloud technology for the first time~