Text | damask
SenseTime's ProIPO valuation level is more than 20 times P/S (price-to-sales ratio), and the highest direct dry after listing is more than 40 times. Following SenseTime, on January 27, Innovative Qizhi will also land on Hong Kong stocks, and its ProIPO valuation level is also 20 times P/S, and I don't know how to speculate when it is listed. But in any case, the AI track began to get restless.
Different AI companies often look at the flowers in the fog, in fact, according to the commercial type of 4 categories, you can quickly understand the difference between each:
Vision AI, the most important applications are face recognition and autonomous driving, the typical company is SenseTime-W (HK:00020), the current market value of 229 billion Hong Kong dollars.
Speech semantic AI, that is, text to speech, a typical company has iFLYTEK (SZ: 002230), the current market value of 112.6 billion yuan.
Decision-making AI, which you can simply understand as "AI playing chess", typical companies have a fourth paradigm, and the current ProIPO valuation is $3 billion.
Artificial intelligence robot, Boston Dynamics' robot dog knows it, the typical company in the domestic company has DJI, the current valuation of 166 billion yuan.
SenseTime and Innovative Chi chi chi we have all interpreted, today let's talk about the fourth paradigm.
Please enter the graph: Decision Class A I Market Share, Source: Prospectus Says
However, from a realistic point of view, the blessings and misfortunes are everywhere, and the distinctive characteristics determine that it must go through several hurdles to truly run through the business model, and unload the outside world's doubts about its "bloody IPO" and "IPO life". After all, the "cash + short-term investment + wealth management products" on the fourth paradigm account cannot withstand more than 1 year of losses; unlike SenseTime and Innovation Qizhi, the money on the account can still be built for several years.
01 Building a building block structure, transfer learning is the magic weapon
Before summarizing the business of the fourth paradigm, I would like to give a decision-making AI application scenario. Suppose a dessert owner sells 100,000 yuan for a certain product last month and 50,000 this month. So how much should I stock next month? If the product is single, experienced decision makers may be able to guess the difference.
But today's consumer preferences for products are changing rapidly, and today they are still pursuing "dirty bags", and tomorrow they will be the only ones who love meat and pine shells. When the product category is richer, the decision-maker cannot be superficial to the market law, and need to deal with the changes in market preferences.
The purpose of decision-making AI is to solve the above similar problems, and it makes scientific business decisions based on data. The logical support of decision-making AI is that statistical principles tell us that as long as we build a suitable model, we can infinitely approximate the most realistic results.
Traditional data analysis software is based on the database where different data are marked and "rigidly" calculated. This method can not flexibly generate new models and tables, but also requires a team that understands both software and industry to achieve flexible data transfer, which is too demanding.
The fourth paradigm provides a simple and flexible data platform to meet the needs of customers in different scenarios. The difference is that it is not trying to provide users with an analysis tool, but to provide users with a "building block" that can be arbitrarily matched with various functions.
Please enter the figure: "Building Blocks", Source: Prospectus
Sage AIOS: A visual operating system similar to Windows, which can improve the efficiency of computing power use and provide a venue for "building blocks".
HyperCycle: A code-free development tool built into the Prophet platform. Follow the prompts to stitch together AI components to complete the data closed loop of AI learning. The difficulty of use is about the same as drawing a scoop according to a gourd.
Sage Studio: Provides AI model editing tools for different programming difficulties. AI modules can be written or created according to different business needs and can be added to the original business data model in the form of composition.
◆ Prophet App: The developed application that can be used directly on the Prophet system is like a mobile APP.
It is not difficult to understand how to do these functions, and the use of transfer learning to make AI commands discover data laws and set data closed loops can achieve the above effects. Transfer learning is an AI algorithm, and it is also the magic weapon of Dai Wenyuan, the founder of the fourth paradigm, who is internationally renowned.
Transfer learning aims to transfer the labeling rules of one set of data to another set of data. For example, if the quality standard of one basket of peanut rice is known, how to classify the quality of another basket of sesame seeds? The migration algorithm uses the definition of large and plump peanut rice known to apply to the morphological classification of sesame seeds, so as to obtain the classification of good sesame seeds.
On the other hand, when the machine screens sesame seeds, it is found that sometimes large and full sesame seeds may be just an empty shell. By automatically adding rules for judging the weight of sesame seeds, AI can further improve the quality of the selected sesame seeds.
The customers of the Prophet platform are able to flexibly define the meaning of the data and improve the data results, and the recommendations given after the data are put into operation, the so-called decisions. The core role of the fourth paradigm product is to help enterprises after digital transformation make good use of data.
02 Frameworks sold to you, use well on your own
Using the prophet platform of the fourth paradigm, it is summed up in one sentence: the product is quite omnipotent, the master leads the door, and the practice looks at the individual.
The fourth paradigm sells three types of products to customers: Prophet platforms and kits, AI hardware (servers) customized with third parties, and custom development services. Customers are mainly system integrators, in the overall solution to do the software part, does not provide a relatively complete digital upgrade plan, terminal enterprises use the platform modeling to rely on their own hands.
Relying on the Prophet platform for AI digital construction, well-used terminal companies will open up a new cycle of business life. As far as the current situation is concerned, the financial industry and the manufacturing industry will have a relatively high acceptance of the platform because of the high degree of similarity of business. With the huge financial and manufacturing fundamentals, the fourth paradigm has also reached a relatively good income level.
Fourth paradigm income, source: W IND figure says
The fourth paradigm of not having direct access to customer data and the fact that the platform is built on demand means that the actual effectiveness of AI depends on how the customer uses it.
In order to differentiate itself from the direct customized solutions of other SaaS vendors, the fourth paradigm changes the charging standard to the form of software licensing + hashrate so that customers can purchase on demand. However, unlike companies that directly own hardware platforms such as Alibaba Cloud, the fourth paradigm does not have a clustered data center, and at this stage, it can only lease servers from cloud service providers and then sell hashrate quotas to customers.
In addition, the Prophet platform is highly modular and standardized, making it much faster than traditional enterprise-level software deployment time. But this also means that the fourth paradigm does not earn engineer on-site service fees in the process of deploying software.
Then, the upstream computing power must be purchased outward, and the value-added services brought by downstream labor cannot be obtained, which makes the gross profit of the fourth paradigm eventually locked at about 40%, which is significantly lower than that of the head AI company (SenseTime 70%+).
Gross margin of the fourth paradigm, source: prospectus
Looking at the characteristics of the fourth paradigm of end customers/users, they all have different degrees of digitization, and the amount of data generated by the business is huge and difficult to integrate effectively. Under the pressure of massive amounts of data, any tool that can replace human analysis or flexible integration of data will be their prescription.
The fourth paradigm carried out in-depth cooperation with Yonghui in August 2019, and the system after the improvement of the adjustable digital model can make personalized recommendations for customers, and finally achieve an increase in the transaction volume of customer orders and increase the corresponding revenue by hundreds of millions. The success of the result comes not only from Yonghui's active digital reform, but also from the Prophet platform.
In 2017, Yonghui decided to lay out the new retail business from the launch of sub-brand super species, but the two-year effort did not have much effect.
In the second half of 2019, Yonghui reached a cooperative relationship with the fourth paradigm, and Yonghui Supermarket launched the Yonghui Grocery Shopping APP, which is very similar to the Yonghui Life APP. Trying to transform online and do a good job of delivery business, but the conflicting APP is considered to have serious differences within the company.
In March 2020, the Yonghui Grocery App was removed from the shelves, and the two apps will work together to develop the home business.
In July 2020, yonghui yunchuang, its digital platform, was re-withdrawn to facilitate the company to better integrate resources and improve the efficiency and service quality of online business. Yonghui Yunchuang, which has returned again, has integrated with Yonghui Supermarket and completed consistency in management.
In the first half of 2021, after the completion of the integration of the retail system, Yonghui online sales reached 6.81 billion yuan, an increase of 49.3% year-on-year, accounting for 14.1% of the main revenue. Yonghui's digital transformation officially began to enjoy a welfare period.
It can be seen from the event line that while the fourth paradigm cooperated with Yonghui in 2019, Yonghui is also striving to "self-help" in the digital retail business. Digital and intelligent transformation is easy to say, but it involves various processes and organizational management reconstruction. The fourth paradigm of the AI platform supports a variety of functions, can be tossed around and also proves the reliability of the prophet platform in architecture.
But tossing the digital platform means that end enterprise users should be the "first people to eat crabs" in the industry. This "crab" can be both delicious and costly to businesses. For them, even the best tools and suggestions are not as good as a tailor-made set of solutions. How good is the AI platform as a tool? Just like a work of art, no one would praise a painting for being good because of the high quality of the brush.
03 The four hurdles of the fourth paradigm
In companies with SaaS products, there is usually a department called the Customer Success Department. The mission of this department is to help customers get positive benefits from good products, thereby improving the reputation of products and customer renewal rates. But for AI companies like the fourth paradigm, if you want users to get good results, you need to cross a few hurdles.
First of all, whether the cooperation model of benchmark customers can be promoted to the whole industry is the first big problem.
As we pointed out when we analyzed the innovation Qizhi before, there is a huge basic disk for the downstream customers of AI companies, and the financial industry is part of the huge basic disk for the fourth paradigm. But this set may not work in the retail industry.
For example, the successful cooperation with Yonghui has shifted to the operation of convenience stores, the advantage of the former is to provide consumers with long-term purchases, so as to obtain consumer portraits, and then optimize the recommendation and purchase algorithms, and ultimately reduce turnover.
For convenience stores, the location of the store is more important than the category of goods provided, which leads to the so-called successful case may be suspected of being rigid in the eyes of others, and finally have to develop a targeted function of intelligent site selection.
Second, there may be no end to investment in research and development.
The Prophet platform actually provides users with a complete set of toolboxes, it seems that users can build freely according to their ideas, but in fact, because of the problem of optional functions, they are limited to a certain extent. This leads to a continuous investment in research and development in order to meet the possible needs of customers in order to meet this hypothetical solution.
The first reason is that THE AI algorithm is still in the process of a high-speed update iteration, and the AI algorithm cannot be used stably for 5-10 years in one version like the traditional ERP software. If customers cannot upgrade their models in a timely manner, they will use more computing power and reduce efficiency as their business grows. This leads to the need for customers to consider extra hash rate allocation and process rationality when writing models.
Development of the fourth paradigm, source: brocade
The second reason is that once there is a unique new demand, the process for AI companies to respond to delivery is no different from that of traditional ERP companies, and all need to be customized development. If the requirements are clear, then why not find a software company to develop it in the first place? Taking a step back, the order generated between this demand and response may be cut off by a third-party service provider. This leads to the fact that the functions and apps of the Prophet platform must be constantly anticipated and developed for customer needs.
Third, the issue of talent management.
The Prophet platform allows people with no code learning experience to quickly eat the benefits of AI application landing by building blocks. Good products need to invest huge amounts of research and development, but once the research and development is completed, how to retain talents?
Algorithms are academic resources, which means that any developer with a design core is likely to step out of the fourth paradigm and start a company in the future. For competitors, the fines for non-compete agreements and the fines for possible infringement in the future are like spicy noodles bought for fifty cents compared to personally investing in research and development costs. Once the product matures and lands, the research and development costs have not been recovered, and the price is likely to decline rapidly because of the inner volume.
The fourth paradigm, as the leader of the modular platform, has to continue to come up with new features to open up the gap with the latecomers.
Fourth, the intellectual property rights of customers to develop models independently.
Suppose that in the cooperation between the fourth paradigm and the Ningde era, it is known that the Ningde era itself has certain data programming capabilities. So who does the AI model tailored for itself by using the Prophet platform in the Ningde era belong to?
If it belongs to the Ningde era, the propaganda strategy of the so-called industry benchmark customers of the fourth paradigm may not be established. After all, customers use it well, and it does not have much to do with product manufacturers. If it belongs to the fourth paradigm, is the global leading level production control system model developed in the Ningde era going to flow to the second and third lines? As a user who develops the model independently, he will naturally not be willing.
Now these problems are not exposed because they are properly solved, but AI is still in the early stages of entering the industry, and the possibility of future occurrence needs to be predicted in advance.
Artificial intelligence in the domestic market is close to the outbreak period today, but step by step is not necessarily the ultimate victory. ON THE ONE HAND, AI TECHNOLOGY COMPANIES ARE VERY OPTIMISTIC ABOUT THEIR LOSSES, AND ON THE OTHER HAND, THEY ARE VERY PROUD OF THEIR INVESTMENT IN RESEARCH AND DEVELOPMENT AND OCCUPYING THE FOREFRONT OF SCIENCE AND TECHNOLOGY. But isn't this investment really forced? Although the ideal of the product is good, it is still far from the maturity of the landing.
It has been four years since the establishment of the fourth paradigm, but the purpose of fundraising is only to step by step technology research and development, salary increases, marketing promotion, and investment and mergers and acquisitions to break into the target industry. It seems to be a very normal choice, but it reveals the dilemma of the company being tied to the donkey pulling model: constantly falling into the cycle of research and development marketing, and can not be like the head AI company, planning to build its own AIDC, landing its own AI acceleration chip, to improve profit margins and product added value.
The fourth paradigm of the AI prophet platform is a good product, but maintaining the advantage is too much to pay.
Looking back at the large number of small shareholders above the shareholding structure, is the fourth paradigm really only a pursuit of development rather than a salvation attempt?
Shareholders of the fourth paradigm, source: prospectus