The future is the world of agents, and multi-agent interaction will definitely become the future. But before that, ‘SaaS+AI’ will be an intermediate state, because as long as it involves TO B services, industry understanding, processes, permissions, security, etc. cannot be bypassed, and these are the core values.
“‘Is the SaaS model going to be replaced?’ This is a question that I have been thinking about frequently this year, and it is also a topic that many people in the industry are focusing on more and more. A founder of a domestic SaaS company told Industrialist.
If in the past two years, the argument that “SaaS model will not be the main service model in the AI era” has been more limited to investors’ forward-looking comments and a series of actions in Silicon Valley across the ocean, but this year, two events have brought this discussion closer.
In May this year, the first general-purpose Agent Manus was released, and although its impressive performance on the C-end brought about a discussion of ice and fire, on the B-side, Manus had a more far-reaching impact. “As soon as Manus came out, we started researching, and even many of the delivered processes, capabilities, and results interfaces that we started re-referencing in our software.” Another CRM business founder told us.
Not only him, but within a SaaS company oriented to data insights, Manus is also regarded as a landmark action of AI TO B. “This means that AI is starting to really have the ability to solve problems, and it can really be thought and planned to deliver results.” The company has made a series of structural adjustments in the past year or two, and the greater significance of Manus’ arrival this time is to “take a reassurance” for this structural adjustment.
Outside of Manus, another event that catalyzed this discussion was the “consensus report” of Sequoia’s internal sharing session in China. That is, at this sharing meeting with the participation of Microsoft, OpenAI and other companies, “outcome delivery” has become a new consensus among everyone on product delivery standards in the AI era.
Contrary to this, in the domestic SaaS track, in the past many years, whether it is CRM, ERP or industrial software, it is difficult for service providers to achieve complete meaningful result delivery in the trade-off between standardization and “customized delivery”.
How should SaaS companies change? In other words, is the SaaS delivery model still absolutely valid in the AI era? And what is the combination model of Agent and SaaS?
“SaaS is now returning to its very concept, it is not a form of software, but a form of delivery.” An investor told us.
Challenges and opportunities coexist. China’s SaaS is entering the AI era.
1. In the AI tide of SaaS, there are two paths that have been seen
In fact, if you look at the action, Chinese SaaS companies have already taken action.
From the perspective of paths, the overall path can be divided into two categories,On the one hand, the Agent PaaS is built based on the logic of PaaSThis path can ensure that SaaS service providers maintain the existing business model and better meet the current AI needs of customers.
After 10 years of interaction design, why did I transfer to product manager?
After the real job transfer, I found that many jobs were still beyond my imagination. The work of a product manager is indeed more complicated. Theoretically, the work of a product manager includes all aspects of the product, from market research, user research, data analysis…
View details >
Among them, some SaaS vendors will give priority to scenarios with high commercial value and clear needs, such as marketing, sales, data, customer service, etc., and the advantages of this path are quick results, accurate resource investment, and avoid the waste of resources for comprehensive transformation.
From the perspective of AI value,The characteristics of this path correspond to the problem of different development maturity and data quality in various sales scenariosThe use of AI to take the lead in landing in relatively data-rich and high-quality scenarios can not only truly meet the sticky needs of existing customer groups, but also provide technical endorsement for upward penetration of large customers. This style of play takes into account short-term commercial returns and the construction of long-term technical barriers.
The future outlook for this path is also clearer.For example, in communicating with many SaaS companies, many have told industrialists that agents also need personalized customizationBased on this, the “Agent PaaS” is already in preparation, and the underlying molecular capabilities are continuously accumulating from general scenario modules to industry-specific scenario modules.
From the perspective of the number of layouts, this is the mainstream method chosen by domestic SaaS companies, such as Saleseasy, Zoho, NetEase Digital Intelligence, etc.
The second path is to create an end-to-end agent product。 The so-called end-to-end also means that the new Agent product and the inherent software are not connected, in addition to the inherent service products, the new product faces new scenarios, and carries out new commercial pricing, such as Beisen, Kingdee’s related products.
Taking products such as “AI interviewer” launched by many manufacturers as an example, unlike the path of integrating AI with existing products, AI interviewer is an AI agent independent of the original product system, which will charge and operate separately in addition to the inherent product.
Corresponding to this path is the organizational structure that needs to be adjusted, that is, on this new AI commercial product, the new delivery triangle needs to have different personnel to deliver, and the delivery cycle, delivery process and delivery standards are no longer the same as before.
However, according to the understanding of industrialists, the same as the first path, the core of end-to-end products is also more in the scenario, that is, in addition to the inherent SaaS direction, enterprises need to confirm that the corresponding scenario has sufficient commercial value and market demand.
“In fact, this is equivalent to putting ourselves on the same battlefield as the current AI entrepreneurs, but we still have a great advantage.” One founder told us.
2. Three mountains: cost, organization and business model
In addition to the path, a more core question to think about is: Does the AI transformation of SaaS have corresponding value?
According to the “2024 AI Agent Market Trap Report” released by Gartner, more than 80% of agent promotional videos in the entire AI Agent industry demonstrate idealized scenarios, and the actual use effect is greatly reduced.
In fact, this transformation is far more difficult than imagined. A set of data shows that more than 60% of SaaS companies around the world are still in a state of loss.Most SaaS vendors have not achieved profitability or tight cash flow.
In this case, if an enterprise wants to carry out relevant AI transformation, even if the current token price is low enough, under the corresponding agent inference training, AI transformation is also a huge expense. According to data, an HR SaaS company spent 2.7 million yuan on data cleaning alone to build a talent portrait system.
The problem is not only in cost, but also in itEstablishment of a new service model。 As mentioned earlier, there are two paths for SaaS vendor transformation, one is to integrate AI into the original product, and the other is to create a new end-to-end Agent product.
Take the former as an example,In essence, AI is used as a value-added service for existing products, which corresponds to two practical problems. That is, one of them is the blessing of AI and original products, which is difficult to quantify the effect of AI within the enterprise, and the bigger problem is the price. One fact that many manufacturers tell industrialists is that “as a value-added service, it is difficult for customers to pay separately, and even if they pay, the cost is very low.” ”
This has also led to many SaaS vendors starting to convert “value-added services” into independent agent products this year, although they still have to be embedded in the original product and embedded in the existing software process, but the pricing mechanism of the product is clearer and “uncompromising”.
In addition, the difficulty is also in delivery. “Now we basically deliver it ourselves, and there are almost no external channel providers for delivery based on AI Agent, and we are also a little bit of a waddle.” A manufacturer leader told us.
The problem with the latter is just as obvious. That is, as mentioned above, the end-to-end “start anew” also corresponds to the verification of the needs of new business scenarios, and enterprises need to go through the verification process of PMF again, but from the current real water temperature of AI product entrepreneurship, with the gradual iteration of AI technology, it is more difficult to start a business from the AI application layer at this stage than before, not only to stay away from the range of large factories, but also to stay away from the development range of AI technology.
Behind these problems, it corresponds to the certainty of SaaS vendors seeking change and the confusion of direction.
That is, with the emergence of general agent products such as Manus, the result delivery and agent form are becoming more and more deterministic TO B service forms.SaaS companies must break the inherent product and delivery model by themselves, but the capability boundaries of large models have not yet been broken.
In addition, the TO B service form based on Agent delivery is not the same as the inherent SaaS service model, and its corresponding service processes, delivery logic, etc. need to be re-changedFor SaaS service providers, they need to be ruthlessly completed to complete the “pain” adjustment to build a new delivery model.For example, the service method of new CRM enterprises is to align the project indicators first, and finally calculate the product and service form backwards from the indicators.
3. SaaS + AI, what path should be taken?
“In fact, Saleforce doesn’t know what is the best way, it is now walking on two legs.” A person in charge of the enterprise told us.
Saleforce is currently the most valuated SaaS company in the world. An introduction about it is that with the continuous advancement of products such as Agentforce, its stock price has risen for many consecutive weeks, and the market continues to be optimistic.
That’s true. From the perspective of Saleforce’s path, its update action for AI large models is, on the one hand, it continues to strengthen its existing CRM product model, embeds the corresponding underlying AI capabilities into the original product module, and carries out product-side evolution; On the other hand, the end-to-end product of Agentforce has been launched, and based on Agentforce’s strong Agent PaaS capabilities, it helps enterprises build a variety of Agent products.
In addition, in order to ensure delivery on the architecture side, Saleforce has also re-recruited new AI salespeople and established a new AI delivery system.
One fact is that the AI transformation of SaaS, or the native transformation of AI, is difficult to complete in one step, and once there is the idea of the “Great Leap Forward”, it is difficult to go for a long time. Just like in the early days of the development of autonomous driving, most of the autonomous driving manufacturers were educated by the market, and slowly changed from the L4 level “one-step” path to the L2 level “gradual” path.
SaaS vendors also need “gradual” AI transformation to go through the “semi-automatic” transition stage。
For example, Salesforce’s Agentforce emphasizes autonomous task execution, but still relies on the original CRM data model and API interface, essentially using AI agents as “intelligent plug-ins” for existing systems. Domestic manufacturers such as DingTalk’s AgentStack allows enterprises to combine AI capabilities and existing functional modules through a low-code platform to form a “toolbox” solution.
The rationality of this intermediate state lies in the fact that it can not only reduce the risk of technical reconstruction, but also improve user stickiness through AI function increments.
In addition, the actions of more international manufacturers are also becoming new references. For example, Microsoft proposed the vision of “Agent Stack”, trying to build an intelligent hub for cross-enterprise applications, scheduling multiple agents to complete complex tasks through a unified framework, and even planning to replace traditional databases as the core operating system of enterprises. Large model manufacturers such as OpenAI aggregate developers through the API ecosystem to promote AI Agent to become a standardized service interface, and with the introduction of the MCP/A2A protocol, this interaction model is gradually becoming more feasible.
But either way, the market consensus that can be seen is that the inherent form of SaaS is changing。
“The future is still the world of agents, and multi-agent interaction will definitely become the future, just like the enterprise procurement process may be completed by multiple agents – demand analysis agents connect with the financial system, supplier price comparison agents integrate e-commerce platforms, contract generation agents link legal modules, and finally realize the whole process of no intervention.” The founder of the company at the beginning of the article tells us.
“But before that, ‘SaaS+AI’ will be an intermediate state, because as long as it involves TO B services, industry understanding, processes, permissions, security, etc. cannot be bypassed, and these are the core values.”
