Chen Zheqian of Yizhi Intelligence: Driven by large models, AI is the protagonist in the Marketing 3.0 era | An interview with 36Kr.
Author | Song Wanxin
Editor | Zheng Huaizhou
Cover Source | Provided by the enterprise
AI marketing has recently become a hot topic in the secondary market. Since November, AIGC-related concept stocks have triggered several rounds of daily limit up in A shares. The enabling effect of AI has begun to be truly reflected in the financial statements of many AI application companies.
Among them, advertising marketing has become the sub-field that first benefits from the efficiency improvement of the large model.
For example, the strongest technology stock in the AI marketing field has emerged this year. As of the close on December 17, the cumulative increase of the US stock "AI Application Bull Stock" AppLovin has reached 770% this year, far exceeding the increase of NVIDIA in the same period. The reason is that the company mentioned that the assistance of AI has continuously improved the AXON advertising engine, making it an important catalyst for business growth.
AppLovin's third-quarter report shows that the company's net profit increased by 300% year-on-year, far exceeding expectations, and the stock price soared by 46% after the financial report.
In China, there are also many technology marketing companies that have ridden the wave of the large model. Yizhi Intelligence, founded in Hangzhou in 2017, is the company with the highest market share in AI interactive marketing services in the domestic consumer retail field, covering more than 30% of top brand customers.
Now, based on the development of large model technology, Yizhi Intelligence has taken a new step in the AI marketing field. AI is increasingly demonstrating its value in improving the marketing efficiency of brand parties and managing the long-term value of consumers.
In an exclusive interview with 36Kr, Chen Zheqian, CEO of Yizhi Intelligence, said that the company has made multiple upgrades in the field of brand marketing based on AI multimodal interaction experience, which can solve users' complex personalized problems, and also allow brands to customize their AI personality, linking marketing reach with brand tone to make marketing services more complete and unified.
Why is marketing the first to emerge in AI application scenarios? Can AI become the protagonist of the new era of brand marketing in the future? The following is an edited excerpt of the exclusive interview between 36Kr and Chen Zheqian, CEO of Yizhi Intelligence:
About Marketing 3.0:
36Kr: After the large model was introduced, what important changes have occurred in the marketing industry due to AI? For example, before this, the AI-generated model images on Taobao were already very mature. Now, in the current marketing field, how is the penetration of AI?
Chen Zheqian: For brands, it is now the Marketing 3.0 era. In the 1.0 era, marketing was a one-way communication logic, such as TV advertisements, telling consumers that there is this product, but the immediate reaction of consumers is unknown to advertisers. In this era, the pain point of brand marketing is that there is not enough traffic. In the 2.0 era, marketing is a targeted recommendation logic, such as Toutiao and Douyin, which recommend the already generated content to users with corresponding preference labels through the recommendation algorithm. In this era, the pain point of brand marketing is that the traffic is not accurate enough. In the 3.0 era, behind the traffic are users who pursue personalized consumption experiences. Whoever can provide users with more unique shopping feelings and memories can truly have customers.
Currently, AI is still in the development stage in the marketing field. The marketing industry is undergoing a profound transformation led by digitization and intelligence, innovating new interactive AI marketing models, and generative AI is rapidly developing and reshaping the marketing industry. Yizhi Intelligence has proposed a new concept: AI interactive marketing, which is a methodology that integrates artificial intelligence technology and marketing practice, focusing on creating and providing personalized and highly interactive consumer experiences to stimulate and guide consumers' emotional responses and purchase impulses in an immersive environment.
36Kr: What breakthroughs have been made in specific application scenarios?
Chen Zheqian: We are building three major scenarios of AI marketing, AI sales, and AI customer service based on the capabilities of the large model, carrying out AI innovations from private domain operations, live-streaming scenarios to user calls, and promoting the innovation of the interaction between brands and users, thereby driving the continuous growth of user value.
Our large model is the "Taiyi" multimodal interaction large model, which can accurately understand the speaker's intention in the interaction, improve the ability to capture key information, and give answers that are in line with the marketing scenario. This year, after injecting a large amount of scene data accumulated in different subdivided industries into the large model for optimization and application, the accuracy rate of NLP natural language understanding ability has reached more than 96%, achieving a comprehensive algorithm-driven conversation scene, which can help brands achieve more than 90% satisfaction in scenarios such as brand promotion, event reach, marketing recommendation, preferential push, conversion and placing orders, after-sales care, and member services.
The improvement of voice interaction technology can enable in-depth communication with consumers from the initial 10 seconds and 30 seconds of AI conversations to more than 3 minutes now. It can be said that the integration of AI capabilities has greatly enhanced the marketing interaction experience.
36Kr: Regarding live-streaming with goods, it still requires a lot of skills and atmosphere. Is the large model of Yizhi Intelligence self-developed or connected to an open-source model?
Chen Zheqian: Yizhi Intelligence is positioned as an AI application company based on a large model. Therefore, in the layout of the large model, we are more inclined to utilize the existing large model infrastructure and conduct a large amount of parameter adjustment and engineering work with industry vertical data, but this is not as simple as just putting on a layer of skin. For example, we have a joint research center with the Artificial Intelligence Institute of Zhejiang University, and we will conduct targeted parameter adjustment training on the basis of top open-source models at home and abroad. Another example is that we also actively embrace the large models of domestic large companies. Our algorithm team will build our own application ecosystem on the basis of the large models of large companies through methods such as RAG.
The "Taiyi" large model matrix launched by Yizhi last year, its biggest feature is not its size, but rather its smallness and speed. In the AI field, there is a consensus that a 10-billion-scale large model in a vertical scene has the same effect as a 100-billion-scale large model in a general scene. In the field of brand marketing, Yizhi Intelligence has accumulated 7 years of industry interaction data, and the business itself is still generating data every day - what questions consumers ask. These data are fed back to the background in real time, and then the large model is fine-tuned in various ways. Therefore, we attach more importance to the accuracy of vertical scene data rather than constantly pursuing the scale of model parameters. In the human-computer voice conversation scene, our first requirement for the large model is to respond quickly, without making users feel any lag. Our business concurrency is very high, with at least 200,000 conversation rounds triggered simultaneously per second. Therefore, if the model is too large for us, the cost will be very high. Our algorithm team has done a lot of research and development in the fine-tuning of the model itself and the model distillation and pruning, and has also invested heavily in the model inference engineering.
About AIGC and Brand Marketing:
36Kr: What was the initial idea that prompted you to engage in interactive marketing?
Chen Zheqian: I have always regarded Iron Man as my prototype. When I was in high school and watched the Iron Man movie, I really wanted to turn that kind of scene into reality, including the holographic projection and the scene where Iron Man's AI assistant Jarvis helps him. During my undergraduate and doctoral studies at Zhejiang University, I was researching many technologies related to human-computer interaction in the laboratory, such as holographic projection, human-computer conversation, and image generation.
Starting a business was an opportunity. The initial direction was to apply the human-computer conversation technology developed in the laboratory for many years to various industries. The development speed was very fast, represented by fast-moving consumer goods retail brands, and gradually extended to local life, games, education, finance and other industries. AI Voice Bot is receiving more and more attention from enterprises in user-centered interactive marketing scenarios. Marketing to acquire and retain customers is an eternal theme for all enterprises. AI is to magnify and accelerate the things that were originally done by manual marketing thousands of times. Therefore, from the initial application of Voice Bot in telemarketing, to the appearance of digital human anchors on various live-streaming platforms, to AI EBA shopping guides, and then to the call center AI BPO, we have been continuously exploring the scenarios and values of AI marketing.
36Kr: The large model can already produce advertising materials in a multimodal manner for AIGC marketing. Why is this necessary? From the demand side, what evidence and data support that users need it?
Chen Zheqian: In recent years, consumer goods have been competing on low prices, which is not a healthy competition. Returning to the essence of business, only when good products generate reasonable profits can high-quality innovation be promoted. Brands should pay more attention to the long-term value of consumers, meet more personalized needs, and establish a strong connection with consumers as a long-term growth strategy. Simple and rough marketing is also being abandoned. How to communicate with consumers efficiently and deeply, and how to have a wider and more personalized interaction with users to create a stronger stickiness will be a large-scale, deep-level personalized reach and marketing.
Now many new brands are investing in live-streaming with goods, and it is relatively difficult to achieve a positive ROI. The supply chain, operations, and product costs have dragged down the profit system without brand bonuses. However, live-streaming short videos are currently the largest traffic pool, and brands have to do it.
Applying AI large model content generation to brand marketing in aspects such as marketing interaction, sales conversion, and customer service significantly reduces the content threshold. The most obvious cost reduction and efficiency increase are in heavy scenarios such as live-streaming rooms and short videos. AI can generate copywriting, characters, and scenes, which are stable and highly productive. With just one idea, a marketing advantage can be quickly obtained.
We have statistically found that the AI marketing conversion rate in specific scenarios has increased by 300% compared to before. Compared to the live-streaming short-video work scenarios with real people, the marketing workflow generated by AIGC technology reduces costs by more than 10 times.
36Kr: What are the technical barriers of AIGC marketing? In the two years since its launch, what was the biggest challenge that Yizhi has faced?
Chen Zheqian: Especially last year, AIGC was flourishing, including text-to-text, text-to-image, text-to-video, etc., and the quality may seem uneven.
Yizhi is at the forefront in the simulation of the AI large model. However, from technology to application, it is still very challenging to make merchants pay for it. Facing commercialization, the value of AI marketing needs to be explained. Whether the customer needs are clear, whether the customer needs are solved, whether it is stable and efficient, and the most important thing is to help customers make money, these are all extremely challenging.
First, in the modeling process of the digital anchor, the facial expressions, such as the direction of the facial muscles when speaking, need to be simulated.
Second, in terms of sound, if the speech synthesis is not good, it can be heard with one ear that it is synthetic. We actually have some breakthroughs in the ability of expression control and body control, which our peers do not have. This is the main point we will focus on in the first half of next year.
Then the third point is that the digital anchor also needs a smart brain, which is the human-computer interaction ability behind it. Having a smart brain is a technical barrier, but I think even if the technical barriers are overcome, the operation is extremely important. Today, the digital human only replaces the anchor, but in fact, for a live-streaming to be successful, the operation team is very important.
36Kr: How big is the market for AI marketing, and how do we determine the opportunities in it?
Chen Zheqian: With the popularization of the Marketing 3.0 concept, more and more industries and companies will apply AI marketing. AI not only improves marketing efficiency but also changes the interaction between enterprises and consumers. It is predicted that by 2025, the market size of AI marketing will reach 40 billion US dollars, which is the general trend.
Currently, there are roughly three types of AI companies in the market: AI computing power hardware and energy, general large models, and vertical industry small model application companies. Computing power and energy are the underlying infrastructure, and large models require a considerable amount of investment to be put into application. Currently, the most entrepreneurial potential areas are mainly concentrated in the application layer: using AI capabilities to solve practical problems, enabling service providers with complete business solutions to improve efficiency quickly. These hidden barriers and customer solutions are the moats of AI vertical model companies.
This type of application company understands the customer scenarios, can quickly deploy AI to replace some repetitive and light decision-making processes, commercialize the large model, provide high-quality products, and build applications that are highly reliable, extensible, and able to handle general and personalized cases. Low latency and seamless integration with existing systems are the key points for the next stage of AI development.
Of course, the commercial success of AI is not easy. AI marketing will first be applied in enterprises that have deployed cloud computing, and be quickly implemented through API and light deployment methods. These companies that have benefited from the wave of cloud computing upgrades are now enjoying the efficiency dividend of AI Agent. They can reach customers with lower costs and higher conversion rates than their competitors, pay for the content quantity, and pay for the results. These enterprises and service providers have entered a growth cycle. AI is improving the brand efficiency and in turn feeding back to the service providers to make AI better. Yizhi Intelligence is also fortunate to be in this growth cycle.
36Kr: This year, the focus of marketing has shifted from competing on price to competing on value, and consumption is returning to rationality. I understand that Yizhi's AI intelligent marketing products and digital human products have been applied to many domestic and foreign brands such as L'Oreal Paris, Estée Lauder, LAMER, Decathlon, UR, and Luckin Coffee during the year-end shopping season. What is the core AI application in this?
Chen Zheqian: In the recent year-end consumption season, Yizhi will provide more than 300,000 AI digital employees online simultaneously to help brands reach more than 600 million consumers. Whether it is handling common customer inquiries or solving complex personalized problems of brands, digital employees can complete the tasks well.
What Yizhi really wants to do is Agent, not just Copilot. Agent needs to evolve the ability to break down goals and the ability to operate and execute. Behind it, it requires the reasoning ability of Yizhi's "Taiyi" large model, the ability of scene data, and the ability of multimodal recognition input and output support. AI, as an important driving force for the new round of scientific and technological revolution and industrial transformation, will have a profound impact on global economic and social development and the progress of human civilization.
36Kr: Yizhi is an entrepreneurial enterprise with the background of Zhejiang University and has also jointly established the Zhejiang University - Yizhi Artificial Intelligence Research Center. In the future, how do you view the development of Chinese AI technology and its application in the business field?
Chen Zheqian: In addition to the competition in technology, Yizhi's original intention is to apply AI to the real world, to make valuable products and a socially meaningful enterprise. When everyone is keen on chasing traffic, we advocate for slowing down the traffic and allowing brands to provide better experiences and services to users through AI applications, forming better reach and interaction between the two sides. When AI becomes the main battlefield of competition, we are against involution. The technological development and large-scale application of AI require the joint promotion of the entire industry. Chinese AI applications must be more pervasive and widely applied in the global field. In this regard, China already has a leading innovative perspective and advantages.
Follow for more information