Large-scale model knockout, dare to ask where the money is?

Text: Dong guides to talk about science and technology

The combined background of science, engineering and finance has given me a better understanding of technology, and I also like to discover changes from the perspective of investment, after all, capital never sleeps and capital is the most sensitive.

The two experiences of investment research in the domestic head private equity and the head financial self-media entrepreneurship have given me a structured thinking system and a better understanding that business operation is not about writing reports. Therefore, analyzing the business war has become the best perspective for me to understand business and discover opportunities.

Today, let's talk about the monetization of large models.

In the field of large models, it has quickly entered the "winner is king" from the "100 model war". For every player, although in his heart is the courage of "dare to ask where the road is, the road is under his feet". However, to win the finals, there is another question to be answered: dare to ask where the money is.

01 The B-side seems to be a little better

To Bor Not to B，that is a question。

There's something appealing about the B-side.

For example, customers have a high unit price and low price sensitivity, so they pay more attention to reliability and stability, and easily do not change partners. As long as the cooperation is pleasant, it can go from a one-shot transaction to a lifetime transaction. There is also no shortage of B-end companies such as SAP and shopify, which are bigger and stronger.

However, if we look back at the stumbling difficulties of B-end SaaS companies in China, we can clearly feel the disadvantages of B-side.

For example, the development cycle is long and the payment collection time is long; It is said that it is a product, but the demand is diversified, and if you don't pay attention to it, it becomes a customized service, and the gross profit margin has dropped significantly. Some small and medium-sized business owners may even feel that it is better to spend money on a set of IT products than to hire two more people.

Even in AI, the development experience of the last wave of industry stars "Four Little Dragons" also shows that the B-end market looks big, but it is not easy to eat meat.

However, for large model companies, the B-side is still an important monetization method that cannot be ignored. There are three reasons for this.

First, inevitability. Large models and AI are not just Internet proprietary technologies, but will become technological "infrastructure" like electricity. It is not just about "painting with poetry", but about combining with thousands of industries to improve efficiency.

This is one of the inevitable attributes of the large model, and it will naturally become a "commercial gold mine". As HUAWEI CLOUD said, AI for Industries is a new flashpoint for artificial intelligence.

Second, economy. The development of large models is already very costly and manpower-consuming, and the difficulty of developing the C-end at the same time will increase dramatically. What's more, the explosion of C-end products also has a lot of randomness. Being a "shovel maker" and selling shovels may have a greater probability of success.

Third, feasibility. Traditional enterprise-level IT products are indeed difficult to sell, because IT is sometimes not just needed by enterprises, or it is not directly effective. For example, information security products are thankless. Only when there is a breakdown, will it be taken seriously.

However, the big model is different. Because, at present, the application technology of large models can improve business efficiency. Such as customer service, marketing, video production, and so on. The benefits are more straightforward. As a result, companies will be more willing to purchase.

The industry leader has also taken the road of B-end. Although OpenAI has ChatGPT for the C-end, the company's revenue still comes from providing products and services to Microsoft and Apple. If you get the money owner, you don't have to worry about food and clothing, isn't this fragrant.

At present, the domestic B-end layout of enterprises is divided into three categories, and the mainstream is the "mantle inheritance" type of cloud business. For example, Alibaba believes that AI is a new form of cloud; Baidu Wenxin Yiyan and Huawei's Pangu model are also cut from the customer channel of cloud business. This kind of enterprise has existing customers, has a sales system, and has an advantage in competition.

Byte's bean bag model also has the group's own big customer. However, the sales system still needs to be further improved.

The second category is represented by iFLYTEK, which has good customer resources in education, operators and other fields, and AI can be combined with previous products to sell, although it does not explode, it can be stable and secure. The third category is pure "budding", hoping to enter the B-end market through technology and products. This type of survival is much more difficult.

Looking ahead, there are three elements for large-scale model companies to compete on the B-side.

One is the channel. The so-called, men are afraid of entering the wrong industry. Customers in different fields have different ability and willingness to pay. Just as semiconductor companies were very rich from 2018 to 2020 and photovoltaic companies were very rich from 2020 to 2022, the financial industry has always been a high-payment field, but it has also ushered in a period of pressure.

Different fields and different time periods have different "gold content". Strike at the right time to mine gold efficiently.

The second is the balance between productization and projectization. Productization means strong replicability and reduced marginal cost. Although every large-scale model enterprise wants to sell products, it is inevitable that there will still be a lot of project-based businesses that need customization and services. In particular, it needs to be integrated with the customer's business, rather than completely starting from scratch.

Therefore, it is necessary to balance costs and benefits in the implementation of large models. Avoid watching the excitement and actually losing money.

The third is endurance. The large model penetrates into the industry, which is a trend with high certainty but not easy to grasp the rhythm, and it is necessary to wait patiently until the moment of the outbreak, or there is no moment of concentrated outbreak, but a long-term flow pattern. In addition, generative AI is not the only route for AI, and there is still the possibility of technological change in the future, which also requires enterprises to have the endurance to survive technology iterations again and again.

Eventually, when large-scale model enterprises continue to take root in the industry, have services, products, and accumulate know-how experience, they will form strong "data and algorithm" barriers.

Therefore, in the domestic market, the B-end business of large models seems to be naturally unsuitable for small and medium-sized startups. Even, perhaps, it is just a redivision between the last wave of "cloud computing" players.

02 The C-end talks about profitability a little earlier

The C-end product is a little elf, which stirs the heart of every product manager, but always covers their eyes.

The benefits of the C-end are not mentioned, for example, the billion-level user base can generate good cash flow and high income, the user stickiness formed by the network effect can improve competitiveness, and the marginal cost can decline rapidly with the expansion of scale. However, there are many difficulties, and the two major difficulties are: users like and users pay.

Why do users like it? There doesn't seem to be a standard answer. Except for the phrase "digging into human nature". If you are superficially categorized, it may include: name, interface, function, etc.

The name is the first element of the C-end competition of large model enterprises. Among the existing products, Wenxin Yiyan and Tongyi Qianwen have the characteristics of the B-side, which reflects the reliability. The bean bun, KIMI, and ingots are more down-to-earth and close to users.

The difference in interface and function is actually in the choice of "single point and universal". According to the positioning of the super intelligent assistant, the functions must be multimodal and diversified; However, for the current commercial stage of AI, what is needed is more penetrating product functions.

That is, the advertisement hopes to "clearly convey information and accurately locate the target". For example, KIMI's "structured" ability, Tongyi's transcription ability, etc., and Yuanbao's official account data are easy to generate clear labels.

The difference between several companies is still the "customer acquisition strategy". Balance between short-term and long-term customers. Relatively speaking, companies that already have C-end products have more advantages when doing multi-functional, and can directly import and instill traffic to users. Otherwise, multi-function is actually a "thankless" positioning, and excellent functions may be hidden deeply, a bit like a C-class employee dragging down A-class.

There is also a big challenge for large model companies to do the "general function" of the C-side, which lies in the "entrance".

Apple, Honor, Oppo, Huawei, etc., all hope to use this wave of AI to highlight product differentiation and boost sales. The goal of mobile phone manufacturers is to combine hardware with software to achieve the "super assistant" effect. The best way is to complete a variety of task scheduling through the AI single portal on the mobile phone.

If the C-end products of large model companies are too generic and diversified, they will inevitably conflict with the AI of mobile phone manufacturers. Maybe there will also be a round of games like games and mobile phone dealers. If it only focuses on one-way functions, it may be more popular with mobile phone manufacturers.

When the product is already liked by users, the hardest challenge is to "get users to spend money".

The development experience of SaaS software in China for more than ten years has shown that the acceptance of domestic users for APP payment is extremely low. As a result, KIMI, who is quite young, took the lead in adopting the "reward model". But the effect doesn't seem to be necessarily impressive. What's more, domestic friends are ready to use free strategies to seize the market at any time.

In the field of monetization, although the large model brings "efficiency" improvement, it will not be a large group to pay for efficiency. And games and short videos have all verified that entertainment is just what is needed.

Therefore, domestic large-scale model companies make C-end products, and the unified caliber is "super early penetration", "not considering profitability", "market share first" and so on.

Perhaps, it is more feasible for the cost of large-scale C-end products to be included in the selling price of the hardware. From this point of view, in addition to mobile phone manufacturers, Baidu, Alibaba, and iFLYTEK all have hardware products such as smart speakers and learning machines, which can be used as carriers for AI.

In the C-end field, there are many companies that do application development based on large models. Unless these companies are already leaders in vertical categories, they can easily become cannon fodder. The large model company directly copies a product, isn't it fragrant?

Although the penetration rate of the C-end market is still low, whether it is burning money or landing paths, the attributes of giants are also relatively clear. But how to make money is not clear.

03 The wave of hardware change is even greater

In the previous Internet field, there were two ways to play to VC and to G (government). However, this wave of large models is not very easy to use. It is already a well-known fact that first-class investors are not having a good time. When there is no money, every investor is shrewd.

And because the large model is too tightly linked to the data, the business of to G is still the business of the giants.

On the whole, the era of large models is not an opportunity for small and medium-sized entrepreneurs. It is more beneficial to those giants with channels and customers, and use new technologies and new products to improve gross profit margins and create new revenues.

The opportunities for small and medium-sized entrepreneurs still come from the "hardware changes". It's like the wave of computers, smartphones.

If the hardware has not changed dramatically in the AI era, then the way to make money for large models still seems to be only a continuation of their respective products and customers. It's just that the B-side will be easier than in the IT era.