From November 29th to December 1st,moment information 2023, "Guo Chao Qi · Everything" Liangzhu 2023 Science and Technology National Champion Industry Conference opened at the Intercontinental Hotel in Liangzhu.The conference invited hundreds of guests, scholars, entrepreneurs, investment institutions, and industrial service providers to create a national wave of science and technology.
The forum is guided by the China Brand Construction Promotion Association, hosted by the Yuhang District People's Government of Hangzhou City, Hangzhou Liangzhu New City Management Committee, China National Brand Network, and New Consumption of Tide.
On the special forum "Consumption+AI, Technology Empowering the All -Industry" special forum on "Guo Chao Qi · Everywhere" Liangzhu 2023 Technology National Champion Industry Conference,Zhang Dongxu, General Manager of the Intelligent Business Department of HKUST NewsThe keynote speeches were published "New Insight AI Marketing Insight: Activate every session interactive value". He said: "Prepare high -quality data assets, build a integrated knowledge base, and inherit the outstanding treasure of the enterprise."
The following is a record of lectures. The new consumption (ID: TIDESIGHT) editor is compiled, and there are deletions:
First of all, thank you for the invitation of the organizer. Hangzhou is the tide of the digital economy. I am very happy to share some of the practices in the application scenario of the big model in the big model application scenario with you here.
What I want to talk about today is the exploration of session value in the marketing link.First of all, let me briefly talk about the development of HKUST Xunfei in the general model. The birth of GPT has been from 2020, but GPT has attracted a lot of attention in China. To be precise, it was around November 30 last year. At that time, GPT3 was released..5.
We will find that the entire general artificial intelligence has reached the available level, and it probably subverts the past awareness of artificial intelligence.In the past, we often said the jokes of artificial mental retardation. General artificial intelligence really changed the world.
HKUST Xunfei has been focusing on the entire artificial intelligence industry since its establishment in 1999.The reason for our voice industry at that time was also the Chinese voice market was occupied by foreign companies such as Microsoft. The founding team of HKUST Xunfei wanted to develop domestic voice artificial intelligence companies. Now HKUST Xunfei has occupied 60 in the voice industry 60The %market and other markets are also occupied by other domestic companies.
Last year, the occurrence of GM's large artificial intelligence model made HKUST Xunfei break the opportunity of a new AI industry. Since December last year, the company's strategic tendency to generally use the GM's large model.At the end of last year, we formulated four plans for this year's press conference. On October 24 this year, we also released a plan for four upgrades.
At present, our effect is benchmarking GPT3.5, and some Chinese fields exceed the effect of GPT3.5.Our effect has also been recognized by the country. First of all, HKUST Xunfei led the formulation of the general cognitive system evaluation system, and it has also become the deputy unit leader unit led by the Ministry of Industry and Information Technology.In addition, in terms of the current evaluation agency, HKUST Xunfei's universal model has basically reached the leading and world -class level in China.
In the development of the marketing field, HKUST Xunfei attracted a large number of developers through an open platform.In order to better serve these developers, we provide traffic monetization services and build a complete marketing system.Starting from traffic and advertising, gradually expand to full marketing links.
In 2021, we began to get involved in the CDP and SCRM fields, focusing on post -link operations.In 2023, because we already have a large -scale model and can reduce the cost of analyzing in different fields, we can have more in -depth insight into the dialogue in the entire rear link.
The reason for this is because in the current marketing environment, brands generally face two challenges.First of all, the market incremental space has gradually disappeared, both Internet users and space for incremental consumption.Secondly, the brand pays more and more attention to the operation of the post -link, including the maintenance and operation of existing users, as well as expanding new markets and users.
In order to cope with these challenges, there are two options for brands.One choice is to go to sea and open up overseas markets; the other is to strengthen the post -link operation, including the operation of private domain users and stock users.Pinduoduo is a successful example. It achieved the market value exceeding Alibaba through the strategy of going out of the sea and private domain users.Although Pinduoduo has some controversy in marketing strategies, it really knows the characteristics of Chinese consumers and can effectively use these characteristics to achieve business growth.
For this link, the first is the data asset -based. The state's temporary management measures for the digital assets of the enterprise will be implemented from January 1, 2024. Enterprise data can become an enterprise's asset -liability statement; the second is to target these assetsHow to do insight; the third is how to enable enterprises to make deeper marketing.
The competition in the new energy vehicle industry is extremely fierce, and the phenomenon of rolls in the industry is also very serious.From the perspective of consumers and corporate links of new energy vehicle industry, it mainly includes online promotion, offline contact, private domain operations and offline interaction.Other brands are more or less involved in two to three links. These links will generate a large amount of corporate data, including telephone data, offline reception data, private domain text data, and transaction data.These data constitute the data assets of the enterprise, which is of great significance to the digital transformation and marketing strategy of the enterprise.
HKUST Xunfei has created a smart hardware, which has a very unique advantage in it, which can help enterprises collect data from offline service scenarios, open up the last mile of digitalization, and make up the last key link of digital assets in enterprises.The cases we apply include automotive, real estate, medical beauty, retail and other industries.
So, how to dig data value and use the effect of large models?
After getting these data, we first need to use the effects of large models to tap the value of data.The big model has excellent dialogue text understanding and key information extraction ability, which allows us to extract the intention and needs of customers from the dialogue.By using the general model, we can analyze the WeChat dialogue of the teaching and training industry, and extract information such as customer intention, purchase objection point, and decide on the cycle of the next purchase.
These data are very valuable for the sales team.In the API into the company's WeChat SCRM tool or the company's own CRM tools, the sales team can get the reminder to communicate with customers the next time, so as to further follow up customers and improve sales performance.
HKUST Xunfei launched smart hardware, and in the scenario of automobiles, real estate, and medical beauty, it has achieved supply sales.These scenes usually require a longer consultation time, and the customer decision -making cycle is also more complicated, and detailed and professional product introduction is required.During the communication process, customers will directly express their views, views and needs on the product.
Based on dialogue -based understanding, we can extract full amount of customer portraits from the data.In addition, we also use the long -text long text understanding ability to summarize the reception.This summary can record what are the clients who are interested or not interested in the communication process, and what are the resistance factors that affect his decision -making.
On this basis, the large model can also provide personalized sales guidance suggestions. Based on the entire reception process, the salesperson told the salesperson how to follow the customer in the next step.Because most enterprises adopt the talent screening mechanism instead of the training mechanism and the cost of training is high, we can use machines and AI to help sales staff and provide personalized customers to follow -up guidance suggestions.
Let's discuss the insight into employees.Know thyself, ever-victorious.In addition to understanding the needs and behaviors of customers, we also need to understand the performance and characteristics of employees themselves.
By sorting out all service processes, we can refine the entire service process into different links and formulate quantitative scoring standards for each link.In this way, we can make the entire service process more quantified, traceable, and recurred.
We find that this method can increase the standardization of offline services of enterprises at least 25 percentage points, and at the same time reduces service disputes and complaints by more than 30%.For those companies with franchise stores or joint stores, this means can effectively prevent and handle disputes that may affect the brand image.
In order to better improve the service, Xunfei's large -fire marketing model has conducted in -depth directional training in terms of emotional treatment.
We can start with three important directions through the analysis of the service process:
First of all, in terms of demand mining, after the customer's results data and transaction data attribution, it is found that the best sales performance TOP10 and the most bad sales performance have at least twice the gap in the number of questions in question;Key movements include digging customer needs and building scenarios to allow customers to fully understand product characteristics; customers will ask some tricky questions, such as what is the price discount of comparing your prices with competing products, or even challenge the shortcomings. Can they be positive?response.
Through the way we know and know the other, we can conduct comprehensive insights on the entire session, thereby helping enterprises to precipitate an excellent interactive process.These processes are data assets accumulated by the enterprise in the operation process. They can use the concept of integrated knowledge base, large starry data model and the company's document resources to integrate and store, build the integration knowledge base of the entire enterprise, and form a treasure for the excellent inheritance of the enterprise.Library.
Xunfei has a very competitiveness in the number of people in the shopping guide.In the process of consulting, in the process of customer questioning, we are not passive Q & A, but active response to push the information that he wants to be more wanted to achieve the effect of automated shopping guide.
The first is the customer service scene.The traditional method mainly depends on the Q & A matching of keywords and fixed expression formats, but the effect is often not ideal, and the hit rate is usually only about 30%.Now we use large models+API and large model multi -mode capabilities, which can make the entire customer service hit rate of 75%or even 90%.
The second is the consultation scene.In the communication between customers and shopping guide digital people, we have learned excellent words and understand how to guide customers to express their needs.In this way, we can refine the needs of customers, supplement its causes and effects, and eventually make product recommendations according to customer needs.
At present, our ability can cover 8-10 rounds of dialogue scenes.For example, in the listing page forms and marketing, and in some private domain scenes, we can take the initiative to ask customers some questions, such as what kind of learning machines they need in the consultation sceneApplication Technology.
HKUST Xunfei hopes to help companies seize the opportunity of each marketing by insight into the journey of users in the entire marketing link.From the session asset -oriented, session insights to session marketing, connect the full link.