This article explores the three-tier architecture trend for ToB software over the next five years, arguing that AI will dominate the application layer to meet personalized needs, while SaaS will focus on the business layer and provide deterministic logic. The article also pointed out that the enterprise system will move towards a unified application layer, and the OA platform is expected to become the mainstream, eventually forming a winner-takes-all track standard product pattern.
During the May Day period, I made some deductions about AI-based SaaS, and also communicated with some founders and investors, and the conclusions I came to were even shocking
The original idea was to go for SaaS + AI, but the conclusion was AI + SaaS
In addition, there are more unexpected deduction conclusions, such as a large number of winner-takes-all tracks in the toB field……
Let’s talk about it in three paragraphs, from the following three perspectives to observe what will happen from today to the next few years:
- Perspective on individual SaaS products
- The perspective of multiple systems within the enterprise
- A global perspective of Chinese software
Finally, this sentence was verified – The higher you stand, the clearer the future you see.
Please come with me…
1. Single SaaS product: AI is responsible for flexibility, SaaS is responsible for certainty
Unlike the classic software layering model (Presentation Layer, Business Logic Layer, Data Access Layer),I layered a SaaS product based on “whether AI has the right to decide whether the program is running”:
The future application layer (including interaction and peripheral application capabilities unrelated to core business) will definitely be the world of AI. Either AI agents or other AI native forms.
This is determined by the uncontrollable needs of Chinese enterprise users. What they need is to be able to complete business work, try to match the usage habits of their own enterprise and individuals, and operate conveniently and efficiently.
I expect it will even develop into a scenario where a new employee starts with a super simple beginner page after onboarding; As they understand the position and improve their operational capabilities, the page will become more and more complex, but their work efficiency will also become higher and higher……
Yes, a bit like a game app – there is only one small city at the beginning, then the regional city and the national battle……
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
In the past, this was unthinkable for complex toB applications. But we can deduce today that AI can meet the individual needs of these businesses and employees at a low cost.
But we can’t rely entirely on AI either.
Microsoft Research’s Adam T. Kalai and his partners have mathematically proven that LLMs (large language models) that are pre-trained with corpus and generate content based on probability have a certain percentage of hallucinations.
The same is true for our intuition – LLM AI, like us Homo sapiens, uses neural networks; And human thinking will also make mistakes, deliberately deceive others, and even occasionally tell lies aimlessly.
The “instability” of AI-native applications, including agents, is the biggest obstacle to the toB system they use to pursue “certainty”.
Inference 1. The main solution to the current AI uncertainty problem is the three-layer structure I drew above: AI is responsible for flexibly meeting the interaction needs of personalized UI at the “application layer”, and SaaS is responsible for providing APIs containing deterministic business logic for AI application layer calls.
This is also the full use of the software investment of the whole society over the past two decades. The business of B-end customers has not changed much, so the business layer does not need to be rewritten.
Although business logic layer products also make extensive use of AI technology in product design, development, and testing, they will not hand over key decision-making power to AI during system operation. In this way, the security and control of processes, permissions, and data are strictly controlled.
Its product iterations are not automatically completed by AI, and each version update still follows the traditional “design-development-test” process, with human engineers controlling quality.
I talked to founders in the industry and found that a small number of SaaS companies have begun to act according to this model: splitting the application layer and the business layer, and packaging SaaS capabilities through APIs.
At this point, I still think that the future is SaaS + AI.
After all, the key to an enterprise’s digital system is still at the business level. But soon this idea was turned upside down……
Let’s listen to this article to continue to break down –
2. From the perspective of an enterprise: a unified application layer
If we think that the enterprise digital system of the future is still today’s multi-chimney structure – OA, CRM, ERP… It is a big mistake to have an independent mobile app and PC interactive page.
In the future, the application layer products of an enterprise system will gradually move towards unification. Because of the high flexibility and strong ability to quickly adapt to personalized needs of AI-native applications, it is no longer necessary to establish an application layer entrance for CRM, ERP, HR and other systems.
Moreover, in the future enterprise software, it is common for the AI application layer to call ERP and CRM business layer APIs at the same time to complete a task.
Inference 2. In the future, all employees of an enterprise will use the same unified application. It is not only a single sign-on for the portal, but also cross-system business operations such as outbound and inbound, order update, business opportunity follow-up, and employee transfer and transfer without jumping.
So who will provide this unified AI-native application layer? There are currently some suppliers:
A. Digital subsidiaries of large enterprises and a large number of custom development integrators in the market: They will pick up AI tools (such as Blue Lake Design, Cursor+ Claude, etc.) and continue to use customized development to meet the individual needs of customers. Although efficiency is greatly improved compared to previous methods, this path does not form a standard product.
B. Existing SaaS companies have also created AI native application teams, combined with their own understanding of the industry (FMCG, manufacturing… ) or field (HR, CRM…) business understanding, and make application layer products that are closer to the business needs of customer enterprises (but they are still chimney-like and cannot achieve the unification of enterprise applications).
C. OA platform companies (Qiwei, DingTalk, Feishu): Try to make standard AI products to meet the application layer needs of a large number of large, medium, and small enterprises. With the support of new AI technologies, these products will be able to solve the problem of personalized needs in the form of standard products; It not only satisfies customers, but also controls the cost per user.
Continuing the inertia of informatization in China in the past 30 years, today’s market share is A>B>C, but it is estimated that it will develop like this in the future:
- Stage 1: From today to the next 2 to 3 years, the AI application layer (B) of SaaS companies will gradually explode and surpass customized development companies (A);
- Stage 2: In the next few years, the more general application layer platform solution (C) will surpass B and win the most market share through a unified “AI application layer” product.
The reason is none other than the variable of new AI technology that provides the ability to meet personalized needs with standard products.
In the past, standard products could not solve the problem of non-convergence of demand from Chinese enterprises. But today with new AI technology, standard products have become a new solution. In such a situation, custom development companies with ultra-low profit margins will also seek transformation.
Inference 3. AI application layer products led by SaaS companies will gain new markets because they combine the flexibility of the AI application layer with the rigorous business logic of SaaS itself. However, in a few years, with the maturity of the “business logic layer” API products of various SaaS companies, platform products such as Qiwei/DingTalk/Feishu will occupy a larger market share because they can provide a “unified AI application layer” for an enterprise.
Let’s compare the “low-code/no-code” scheme that was very popular 5 years ago. At that time, I also pinned my hopes on her to solve the problem of non-convergence of corporate demand.
However, without the “three-tier structure” in the figure above, low-code applications are often started from scratch, which is only suitable for meeting simple needs within enterprise departments, and it is difficult to solve complex cross-departmental system problems in enterprises.
With AI, this underlying logic has actually changed; The impact results will be shown this year and next year.
3. Global perspective: On a large number of tracks, winners take all!
Continuing to the third step, I actually came to a conclusion that I had not expected beforehand –Winner takes all!
Due to the competitive situation between China and the United States, China’s digitalization will definitely form an independent ecosystem.
But there will be great integration within the ecosystem.
First of all, it is the integration of the “application layer”. As mentioned above, in order to meet the individual needs of small and medium-sized enterprises and the needs of large enterprises to integrate into hundreds of information systems, the “application layer” will have large products and large aggregations.
Secondly, the “business layer” will also have industry- and field-level winner-takes-all products.
The main business system of each large industry (such as MES in the auto parts industry) and key general systems (such as CRM in the equipment manufacturing industry) will be winner-take-all, and the general systems of decentralized small and medium-sized industries (such as general HR) will also be winner-take-all.
Why?
Because after the demand tends to be consistent, the black hole effect of data accumulation in the AI era will make the strong stronger. From this, we can deduce together how domino-like changes will occur in the next few years:
SaaS companies understand the industry/field better, and the personalized part is borne by the application layer through AI technology. Whether it is the above-mentioned A (customization company), B (SaaS company), or C (platform company), which party is responsible for the “application layer” product, the “business layer” under the application layer, which can only be responsible for the SaaS company, will be easier to make the “standard product” we dream of.
The main competitors of SaaS companies are not SaaS companies, but large and small custom development companies. In the future, no matter how large or small, enterprises will switch to the above three-tier architecture. Perhaps in 5 years, 90% of custom development companies will have to transform, and SaaS companies that provide business-layer APIs will have more and more advantages in “standardized products” and data accumulation.
The threshold for doing business-level products is a deep understanding of the business, and this threshold is very high. This is different from the genes of companies that do flexible application layer products, and in the future, the “application layer” and “business layer” are more likely to be divided into two types of companies.
The threshold for “business layer” products will become higher and higher with the accumulation of data and the application of business layer AI technology (data analysis and forecasting, etc.). Companies that make “business-layer” products will get what Professor Zeng Ming called the “black hole effect”.
In the application layer, SaaS companies have the advantage of understanding business better in the short and medium term, but the three platform companies of Qiwei, DingTalk and Feishu are the “friends of time” – AI technology makes general product capabilities stronger and stronger, and with the support of the “business layer” API, it will gradually cover and crush the “application layer” products of SaaS companies. Perhaps in 5~10 years, most enterprises will choose the unified “application layer” products of platform companies. At present, our SaaS companies are more likely to retreat to the capabilities of the “business layer”.
So can SaaS companies choose not to provide “business layer” APIs to other companies’ “application layer” products? Let’s deduce: there are 2nd and 3rd places on this track, if you don’t provide it, others will provide it; This is the customer’s choice, not the control of the SaaS company.
Therefore, in the end, professional people do professional things: “application layer” products will be aggregated according to toB, toC and other customer types, and “business layer” products will also be aggregated according to industries and fields. The winner takes all, and this is where it comes from.
The final result of this deduction is that from the 1990s to the present, the unsolvable problem of “demand without convergence” in the informatization of Chinese enterprises for 30 years will finally be solved by AI’s flexible “application layer” products. Moreover, very successful standard products will appear at the application layer and business layer.
The government, state-owned and central enterprises will also use public cloud “application layer” standard products, the only difference is that their “public cloud” is “government cloud”/”financial cloud”/”state-owned cloud” and other clouds controlled by state-owned enterprises. We have also seen the determination of governments and state-owned enterprises to embrace AI in this wave of Deepseek. Digitalization and intelligence are related to the national fortune and must be done.
4. There are not many options for SaaS companies, but they are more worth looking forward to
Therefore, there are several key paths to choose from among the thousands of Chinese SaaS companies:
A. Separate the application layer and the business layer, enhance the business layer API, and provide services for its own and external application layer products;
B. At the same time, we will make efforts at the business layer and the application layer, invest heavily in AI applications, and strive to maintain our application layer.
C. (For companies with too thin business layers) Abandon the business layer and reinvent itself as an AI-native product company.
For software companies that have invested hundreds of millions or billions to build PaaS, although it is painful, they also need to put aside the capabilities of PaaS at the application layer (custom pages, reports, BI, etc.), and focus on the PaaS capabilities at the business level (process configuration, custom objects and fields, etc.).
Focusing on the business layer, helping customers solve real business problems, and leaving the problems of the application layer (personalized interactive pages, personalized needs, etc.) to the native AI team and internal and external implementation teams to solve may be the solution to the standardization problem of Chinese SaaS products.
Finally, I give the SaaS team development advice in the past two years:
focus on the business layer, get close to customers to help customers solve business problems; Both industry SaaS and general SaaS must try to participate in the digital transformation of industries or fields (such as HR, marketing) and make differentiated advantages.
Package business-layer capabilities in your product into APIs for internal and external teams.
Consider adjusting the charging method from annual fee to charging based on business volume (e.g. number of service layer API calls), billing per performance (e.g. commission on sales).
Strive to become the boss of the track (a secondary industry or a general field)
Find the logic of constructing the “black hole effect” from the data level
