What can be perceived from Ronglian Qimo is a practical and feasible industrial model landing path in the Chinese market. In this set of mechanisms, it can not only empower, but also self-grow and evolve. While strengthening the digital capabilities of enterprises, it is also building a flywheel effect of its own model.
This is not only the new enlightenment brought by Ronglian Qimo to intelligent customer service, but also the new thinking brought by this company to the next journey of China's large model.
Author | Pi Ye
Produced | Industrialist
How do you feel when your track suddenly becomes the focus of the market? Wang Chunsheng told us, "The impact is very high." He is the general manager of operations of Ronglian Cloud's digital intelligence cloud business group, and he is also the first group of people to face the changes in the wave of AI large-scale models that began in March.
According to a recent report titled "2023 China Smart Customer Service Market Report" jointly released by Sullivan and Toubao, Ronglian Qimo ranks first in the evaluation of domestic smart customer service companies from the two evaluation dimensions of growth index and innovation index. China's intelligent customer service track market leader echelon.
This kind of leadership in the smart customer service track also constitutes the first attention that people pay to this company when the big model comes, but this attention corresponds to pressure. That is, after OpenAI became popular, "smart customer service" first became the first industry that people imagined might be "replaced".
But for Wang Chunsheng, the "anxiety" he felt only lasted for a short time. "Now it's more of a surfer's mentality to challenge the limit." A surfer means overcoming difficulties and overcoming obstacles.
His perception is also a true portrayal of the current big model. That is to say, with the increasing number of large models, more and more real problems beyond the imagination gradually emerge. These problems not only appear in the many capabilities of general large models, but also in the real implementation of industrial and enterprise scenarios.
With many attempts, more and more people are facing the reality: how should the large model contribute real increment in the scenario of enterprise digitalization?
While feeling the changes, Wang Chunsheng and Rong Lianyun's team have already started to take action. At the 2023 World Artificial Intelligence Conference, Ronglian Cloud officially released the "Chitu Large Model" for enterprise applications, and at the same time, the "Generative Integrated Intelligent Customer Service Platform" was also officially unveiled.
The sailboat was already in the water. It is understood that before this press conference, the "generative integrated intelligent customer service platform" has become one of Ronglian Qimo's service options for enterprises.
From an objective point of view, Ronglian Qimo is a typical case in terms of the current node of "exploring how to implement the industrial model" . A large model model that can actually land and work well.
This kind of exploration corresponds not only to a new definition of the proposition of "smart customer service" in people's mouth, but also a new profile of the next stop of the current large-scale model industry.
1. 100 days from anxiety to excitement
"At first it was anxiety, but now it's more of a sense of excitement." Wang Chunsheng told us.
Turning the timeline back to March, with the discussion of chatGPT and OpenAI in the global market, China's industrial digital market has also fallen into a new paradigm of AI thinking. Among them, intelligent customer service bears the brunt.
"When the big model comes, or chatGPT comes, the previous model of intelligent customer service will be subverted, and it is enough to directly use the interface to connect, and the enterprise does not need to buy a customer service system." An investor said at the node in March Tell the industrialists.
But is this the case?
In fact, as a veteran of the smart customer service track, Wang Chunsheng and his team had already started thinking long before the heated discussions in the market. In addition, in addition to the route direction, a beta version was also accelerated.
"In the first March, we tested our 1.0 version internally for some customers, and collected a lot of customer needs and product opinions." Wang Chunsheng told us, "We also wanted to test whether our corpus was enough. And what is the customer's real demand based on the big model."
This is the unanimous choice of mature TO B companies. That is to say, for new technologies and new products, if we directly conduct independent and creative research and development based on the heat of the market, the end result will often be that the service and demand are not equal; the model of Ronglian Qimo is based on the needs of customers . Carry out product and market research and judgment.
After testing by Wang Chunsheng and his team, the route of the large model of Ronglian Qimo was quickly determined. That is, on the basis of the Chitu large-scale model that focuses on intelligent customer service capabilities, it integrates powerful general-purpose large-scale model capabilities on the market to connect and integrate upper-level intelligent customer service applications. In other words, Ronglian Qimo wants to "gather the strengths of a hundred schools of thought".
"In addition to the large Chitu model, we have also connected with Baidu Wenxin, Tsinghua ChatGLM, Huawei Pangu, etc., to jointly strengthen the intelligent customer service capabilities of Ronglian Qimo." Wang Chunsheng said, "Based on these general large models, the dedicated interface, we can also provide customers with more comprehensive AI intelligent customer service.”
This is the choice of most of the top TO B application companies on the market at present, that is, regardless of the huge investment in the general large-scale model route, the reason for choosing the compatible direction is that the underlying general large-scale model is constantly changing, including the corpus. , to gather the training capabilities of each company, so that the corpus covered by the enterprise's application products will be more complete.
In fact, for Ronglian Qimo, this is the best way to realize its true value. That is, on the basis of the computing power of the general-purpose large model, Ronglian Qimo can conduct proprietary training and fine-tuning of the model based on a large number of vertical industry corpora accumulated by itself, amplifying its own advantages in the direction of intelligent customer service.
"The entire fine-tuning process is based on our own past data on the one hand, and on the other hand, it is co-created with customers to conduct data training on actual scenarios under the premise of safety." Wang Chunsheng said, "The entire fine-tuning is done by our R&D team. Round by round, the whole is complicated and time-consuming.”
These directions and explorations finally constituted the "intelligent customer service generation integrated intelligent customer service platform" released on the site of the World Artificial Intelligence Conference on July 8.
2. Welcome to the era of intelligent customer service 2.0
So, how was the result? At the conference site, the figures given by Wang Chunsheng - 70%, 100%, 80%. They correspond to the improvement ratio of knowledge production efficiency and customer service efficiency and the reduction ratio of dialogue cost construction respectively.
In fact, before judging the results, the real status quo is that Ronglian Qimo's "intelligent customer service generation integrated intelligent customer service platform" is currently one of the few large-scale model products in the SaaS industry that has the ability to implement projects .
This is not an easy task. That is to say, for SaaS, although it has a certain amount of SaaS corpus, it is difficult for the capabilities corresponding to these corpora to truly "emerge" after several rounds of training, corresponding to the capabilities of the industrial track, which is comparable to the capabilities of the general large model. Compared with other industries, the increase is not large, and customers' willingness to pay is extremely high.
This is exactly the value of Ronglian Qimo AI products. That is to say, the confidence that it can be released to the outside world is that the bottom layer of the product is supported by the Ronglianyun Chitu large model that emphasizes intelligent customer service capabilities. At the same time, it is compatible with the general capabilities of other large models. Under the premise of ensuring that AI products complete "undergraduate education " , and a large amount of specialist education in the direction of intelligent customer service was given to him.
But that's not all. "We will equip enterprises that use large-scale model products with an AI trainer. On the one hand, they will assist enterprises in the use of large-scale model products, and on the other hand, they will establish a data training mechanism between large-scale models and enterprises." Wang Chunsheng told us, "Before these The AI trainer is more internally-oriented to help train our AI products X-Bot and AIcall."
For the delivery of large-scale model products, within Ronglian Qimo, it is divided into two modes: public cloud and private cloud (local deployment). The former is more of a subscription system, and small and micro enterprises can train themselves on Ronglian Qimo's products according to their own needs; the latter is a project system, that is, Ronglian Qimo will help build the company's own corpus and train the model . For these two types of customers, Ronglian Qimo will provide AI trainer related services.
"Basically, the public cloud can meet most of the intelligent customer service needs of enterprises, but for some vertical industries or product rules of more subdivided tracks, special training is required." Wang Chunsheng said, "In fact, there will be help enterprises to build question-and-answer rules before . library, but now that the large model has come, the original 2-3 months has been shortened to a few days to complete.”
One background is that in the process of local deployment of the intelligent customer service system, the common process is that the enterprise provides a series of question-and-answer rules based on products and services, and the service provider helps it build its own customer service vocabulary and corpus. It is often the core part of the implementation of the intelligent customer service system, and it will last for several months or even half a year depending on the situation of the enterprise.
In today's large model stage, it is greatly simplified. Wang Chunsheng said that enterprises only need to input their own data into the large model, and the large model can automatically generate the enterprise customer service corpus and use it directly online.
The change is not only in the construction link, but also in the effect link. " It used to require 10 people, but now basically 3 people are enough. For example, an AI agent can solve the problems of 50 people a day."
In addition, unlike the fixed customer service language in the past, the large model can give more personalized and emotional answers based on the corpus uploaded by the enterprise and its own general customer service corpus.
Some real data on the enterprise side were also presented at this conference. The first A-share listed company in the home furnishing industry used Ronglian Qimo’s AI products to solve the inherent problems of no one on duty at night and untimely reception during peak hours. Constantly optimize.
The same is true for another leading dairy group in China. Based on Ronglian Qimo’s AI products, its voice robot responds quickly in the process of communicating with consumers. The response time does not exceed 2 seconds, which is 500% higher than the average 12s response when the robot was not used in the past. , the overall call satisfaction increased by 27%; in addition, the artificial replacement rate of intelligent return visits exceeded 85%, and the average annual labor cost can be reduced by one million yuan.
It is understood that in today's Ronglian Qimo products, AI product options will be provided separately for enterprises, namely artificial seats and AI seats, and enterprises can freely customize according to their own needs.
This grounded engineering capability is also feeding back Ronglian Qimo's large-scale model service capabilities. " The general capabilities of the model capabilities trained and optimized on the private cloud side will be fed back to public cloud products, helping enterprises using public cloud products to obtain a better AI intelligent customer service experience."
3. Where is the next journey of China's large model?
Where is the next stop for large models? In the answer to this question, industry is an inevitable proposition. But as far as the moment is concerned, it is difficult to see that there are enough proprietary models suitable for the industry in China, even on the basis of general-purpose large models.
However, what can be seen is that companies like Ronglian Qimo have already started to take action, and have initially established a benign large-scale model closed loop. "This year we will build an industry corpus with enterprises in 10 industries." Wang Chunsheng told us.
Actions based on this plan have already begun. According to him, the AI trainer system will be continuously optimized and upgraded in the future, and its closed loop on the data side will be strengthened on the basis of the original big model "customer success" role.
What can be seen is that Ronglian Qimo is becoming the first batch of people who run domestic large-scale models on the industry track. This kind of industry-based action is not only based on its own inherent intelligent customer service capabilities, but also on the mechanism model of the large-scale positive cycle in the industry.
In this model of both project implementation and feedback, Ronglian Qimo has truly taken root in the enterprise, based on AI training, industry corpus and other methods to help enterprises build a digital environment that adapts to large models from data to applications, and these capabilities are not only It can be used in the direction of intelligent customer service, and it can also be reused in other links within the enterprise.
In Wang Chunsheng's plan, Ronglian Qimo will also strengthen the AI capabilities in more subdivided directions in the industry track, "Such as clothing, shoe accessories and other more detailed industrial directions, it is very important to empower these directions based on AI. valuable."
In addition, he will lead the team to continue to cooperate with various offline call centers to test the effect of AI products on actual enterprise activities. In his words, "continuous exploration and continuous optimization."
It can be said that what can be perceived from Ronglian Qimo is a practical and feasible industrial model landing path in the Chinese market. In this set of mechanisms, it can not only empower, but also self-grow and evolve. While strengthening the digital capabilities of enterprises, it is also building a flywheel effect of its own model.
This is not only the new enlightenment brought by Ronglian Qimo to intelligent customer service, but also the new thinking brought by this company to the next journey of China's large model.
Write at the end:
What will the future intelligent customer service look like?
Wang Chunsheng described to us the future of the industry he imagined, "In the future, intelligent customer service will have emotions, which can recognize people's emotions, feelings, personality, etc., and can truly liberate manpower. This is the direction of our future efforts."
In May of this year, Sequoia Capital sent a letter to all the invested companies. In the letter, there was such a sentence that in the changing times, "The last (company) that survives is not the most powerful, not the most popular, nor the most popular. Not the smartest, but the quickest to adapt to the environment."
Now that large models are coming in, in the much-discussed Chinese smart customer service track, the evolved Ronglian Qimo is becoming the best proof of this sentence.