Significant Changes in the ToB Market, Service Providers Adapt Their Approaches.

"Eighty percent of e-commerce clients report a clear decrease in software and infrastructure budgets."

"Information systems composed of menus and hyperlinks are now too easy to develop with AI."

"Previously, everyone wanted a tool; now, everyone wants a result."

This year, the flow of money in some enterprise markets is changing: in e-commerce, marketing, and office-related lightweight system areas, software and infrastructure budgets are being compressed.

The freed-up money is flowing towards AI. In the past, enterprise procurement was driven by mature business systems, where AI features might have been highlights or bonuses. Now, AI itself is driving enterprise procurement decisions.

At the same time, AI coding has inspired some large enterprises to consider self-development in various scenarios such as marketing, office operations, and customer service. Low-threshold scenarios with short chains, low barriers, and no reliance on deep private enterprise data are being self-built, while bosses are asking, "Where is the impact of our spending?"

Budget changes and the flow of money act as a baton, driving adjustments in parts of the software and service market. Observations from Shuzhi Qianxian indicate that many service providers are increasingly aware that their original service models are no longer viable.

Refine products? Enhance services? Deliver results?

A reconstruction is underway.

01

Changes and Constants After AI Influences Budget Mindset

In the first half of this year, an enterprise digitalization service provider observed a very clear signal among its e-commerce clients: approximately 80% of e-commerce clients reported a clear decrease in software budgets.

"There were signs in the previous two years, but this year it has been particularly evident," the vendor told Shuzhi Qianxian. This trend is not only happening in profit-constrained companies; some companies with better profits are also making similar moves.

However, there is also a view that budgets have not disappeared from these companies; rather, the allocation methods within many companies have changed, with some companies shifting their budgets towards AI.

Shuzhi Qianxian learned from the industry that the penetration of intelligent agents is accelerating this year. In many large enterprises, headquarters or top leaders are generally setting new tasks: certain work processes must be solved using AI. For example, a home appliance company requires certain processes to be AI-driven, while a large clothing company requires a certain proportion of new product designs to be completed by AI.

AI coding has significantly reduced software development costs and is reshaping the mindset of budget setters. "They may not know exactly how effective using AI will be, but they feel that large models have eliminated many software barriers, so this part of the software budget should be reduced and allocated towards AI," observed Liu Jingyi, a senior product operations expert at Lingyang AgentOne.

This top-down, hard- indicator -driven change has been confirmed by multiple sources. Sun Linjun, CEO of Shizhi Intelligent, an enterprise-level agent vendor, also sees that intelligent agents are greatly replacing traditional IT system construction in industries like e-commerce. In Q1 this year, Shizhi Intelligent Agent achieved a 150% growth in the e-commerce sector, with a large number of new clients signing up immediately. Sun Linjun attributes this to the fact that "the crayfish craze has educated the market; enterprise bosses have already done their research in the market and usually come with clear demands."

The reactions of many traditional industry bosses have also been swift. "This year, many traditional industries such as exhibitions and real estate have shown a clear interest in AI and are willing to invest large budgets in project transformations," a person from the enterprise market feedback.

Wang Ting, the head of AI success for clients at Linghe Shuzhi, a manufacturing AI service provider, has interacted with many traditional enterprises and observes that AI has become a "very clear" decision point in enterprise procurement this year.

Wang Ting has experienced the change nodes when technologies like ERP entered enterprise procurement decisions during enterprise digital transformations and deeply perceives that the current AI procurement decisions by enterprises also follow this cyclical pattern. "In 2008, during the rise of ERP, the industry said that implementing ERP was suicide, but not implementing ERP was waiting for death. Today, AI has reached the same decision node. Regarding whether to implement AI, many enterprise bosses have reached a consensus; they are just hesitate on how to do it, when to do it, and what conditions are needed," she said.

Some overseas research data even show the proportion of budget reallocation, with companies shifting recruitment and software spending towards AI. Goldman Sachs released a CIO survey in May 2026 that provided a more specific figure: only 33% of enterprise AI token budgets come from new allocations, while the remaining 66% come from reallocating existing budgets. The two main sources are labor budgets and application software budgets.

However, industry veterans believe that this reallocation mostly occurs in lighter software system departments such as marketing, customer service, and office operations. Complex production manufacturing, banking credit finance, and core ERP, finance, taxation, and legal areas of large enterprises remain the domain of mature enterprise-level system software, with AI penetration still in its early stages.

Shuzhi Qianxian contacted the big data department of a leading bank and learned that they do not even use AI systems to process data because traditional data analysis methods are very mature, and the financial system has high accuracy requirements, making the introduction of AI very cautious.

An ERP vendor also mentioned that they have been promoting AI capabilities in the ERP field to create benchmark clients and have been co-creating with benchmark clients for a year. However, the progress in covering large clients is very slow, and it is difficult to quickly scale up like in the office sector.

02

Client Self-Development Trends: Light Scenarios Being Penetrated First

Closely related to the changes in enterprise budget mindset and procurement willingness for "light systems" is the strong desire for self-development among demand-side parties after AI coding capabilities have increased.

Liu Jingyi observes, "Clients have become accustomed to conversational interactions and may feel that they can buy some cloud machines and computers to do it themselves for some ordinary products." Moreover, due to the rapid progress in intelligent agent capabilities, internal demonstrations within enterprises are extremely fast, causing service providers to no longer only compete with external rivals during the POC phase but also frequently compare effects with internally built tools by clients.

Xiao Yuyan, Deputy General Manager of NetEase Zhiji and Head of Cloud Business, also told Shuzhi Qianxian at the end of May this year that the most direct challenge they feel this year is not from competitors but from some large enterprises considering self-developing certain modules and product functions. "Clients may have a sizable internal technical team that needs to prove its value," she said.

The desire for self-development has also spread to the government side. Fang Yi, the founder of Daily Interactive, gave Shuzhi Qianxian an example: a local development zone leader no longer approves information construction demands submitted from below because someone in the office has already developed the system using AI. Some government ends, when planning to build similar government systems, will proactively learn from mature practices already in place in other regions, using AI tools to shorten the preliminary research and prototype-building cycles, and then make adaptive adjustments based on their own needs.

Fang Yi believes that AI has also impacted the bidding process itself. "The detailed functional point descriptions in bidding documents can accelerate preliminary demand sorting and development verification with the help of AI tools." Even bidding documents and tenders themselves have introduced AI assistance.

The scenarios that are first being penetrated share common characteristics: short chains, low barriers, and no reliance on deep private enterprise data. For example, in the after-sales evaluation link of e-commerce, Liu Jingyi introduces that for scenarios analyzing individual product reviews, clients tend to build their own intelligent agents, "consuming at most a few hundred dollars in computing power. The logic of selling a complete product for tens of thousands or hundreds of thousands of dollars in the past no longer works."

For complex systems heavily tied to process flows, industry knowledge, and private data, such as energy and chemical, industrial software, and ERP core modules, the possibility of AI replacement is still low. An enterprise-level agent vendor introduces that in some traditional manufacturing industries, for high-complexity scenarios and systems supporting the current core business operations of enterprises, companies often dare not directly replace them with AI but instead use interfaces or RPA to call relevant data. "The interaction methods may have changed, but they dare not randomly modify these assets and systems."

In exploration and trial and error, it is evident that the supply-demand relationship in the market is rapidly changing. Service providers can also observe that many companies are oscillating between self-development and external procurement.

Wang Ting from Linghe Shuzhi encountered a company with nearly 20 billion yuan in annual revenue that formed a 5-person AI team and used an open platform to build various tools for more than a year. The chairman evaluated that except for marketing, which could quickly produce images, there were no other effects. Afterward, this company turned to seeking external teams to take over. "We have encountered more than just one or two such companies," Wang Ting said.

Due to the sharp decline in software development barriers, IDC states that enterprise AI procurement is increasingly focusing on usage efficiency, cost control, and quantifiable implementation results. Many service providers have already felt that enterprises' procurement standards for external service provider products also require clear outputs and "results."

Cheng Weizhong, the founder of Cognitact, does not shy away from the challenges faced. Over the past six months, Cognitact's traditional tool-oriented digital human business has encountered significant Impact . "Previously, everyone wanted a tool; now, everyone wants a result. Having a digital human in an e-commerce live streaming scenario does not mean the goods will sell," Cheng Weizhong believes that not only digital humans but the entire SaaS industry needs to face clients' questions about whether the service provider's products can bring results and effects.

The emphasis on ROI and value makes it difficult for many traditional industry bosses to promote AI within their companies with a "token maxing" attitude like internet giants did at the beginning of the year.

Wang Ting from Linghe Shuzhi relayed a case from a manufacturing client boss to Shuzhi Qianxian: the enterprise spent 50,000 yuan on an office AI product and stopped cooperating after half a month. "They didn't even hear a sound for 50,000 yuan; it seems that even spending 500,000 yuan may not yield a sound." In her view, this is not a problem with the tool or platform. If enterprises cannot measure the value brought by AI after purchasing it, they will naturally find it difficult to truly pay for AI.

03

Service Providers' Responses: Refine Products, Enhance Services, Deliver Results

Based on the real actions of different types of vendors over the past six months, Shuzhi Qianxian found that agent service providers are adjusting to the market's self-development trends and demands for delivery results by making changes at multiple levels, such as customer segmentation, finer product granularity, and delivery results, to match market demands.

Taking Lingyang AgentOne as an example, they are breaking down the granularity of service scenarios. Previously, they served clients with entire products or platforms; now, they need to adapt to the AI era by breaking down product skills and MCP interfaces. Lingyang told Shuzhi Qianxian, "Only with finer granularity can we find more customer entry points in the market."

In this process, customer segmentation is inevitable. Shuzhi Qianxian learned from Lingyang that a few enterprises at the top of the pyramid have already built many agents internally and seek service providers to supplement the "last-mile skills they lack." Lingyang's agents can integrate into the enterprise and achieve skill complementarity through agent-to-agent calls.

The second type is the current mainstream, such as some e-commerce enterprises that have received the mandate from top-down to "use AI to improve e-commerce operation efficiency." They have opened up the operational scenarios of several e-commerce stores, with more than ten agency operators needing to upload images, adjust prices, register for events, and place ads every day. Enterprises need to improve efficiency and reduce costs in these deterministic scenarios. For such clients, Lingyang can provide a managed digital employee solution, similar to an "AI operation employee," allowing the digital employee to "onboard" like a real person, be assigned a cloud-based employee computer, come with pre-configured job skills, and have the enterprise configure the digital employee's guidelines and permissions.

There are also some large clients with relatively vague demands who only have a need for AI but no clear ideas on how or where to use it. For these scenarios, Lingyang will use an FDE (Frontier Deployment Engineer) team to enter as experts, "see them off on the last journey," and then precipitate the scenario demands and functions back into the enterprise's standard products.

Service providers redefining customer boundaries are not uncommon.

Xiao Yuyan from NetEase Zhiji also saw two typical client profiles this year: "one says I want to self-develop, and the other says I failed at self-development and came back." For top enterprises with internal self-development capabilities, Xiao Yuyan believes that there are massive long-tail demands within these enterprises that can be autonomously developed through the enterprise's general-purpose middleware capabilities. In fact, a fully customized development model closely following top clients does not align with the business models of most software service providers. Therefore, NetEase Zhiji will focus more on serving mid-to-lower-tier clients who lack the self-development resources and capabilities of large enterprises. To serve these enterprises well, in addition to refining implementation scenarios and product capabilities, they also provide a series of tool combinations to help improve AI operation capabilities after product delivery.

Delivering more closed-loop results is also the strategy Cheng Weizhong from Cognitact is currently adopting. Cheng Weizhong introduced that around the result-oriented product delivery, over the past six months, Cognitact has been shifting from pure SaaS sales of digital humans to effect-oriented services, focusing on building a closed-loop service chain from traffic acquisition to conversion.

Cheng Weizhong believes that enterprise-level agents will only have two models in the future: one oriented towards results and the other towards light customization and private deployment, heavily occlusion with client data and SOPs. "Short-chain workflows will basically be eliminated as large model capabilities upgrade."

Therefore, in the original arena of digital human live streaming, the energy distribution of the Cognitact team is already different from before. Only 20-30% of the energy is spent on live streaming digital human products, while 70% is spent on filling the capability gaps in the enterprise's sales business chain outside the live streaming room. For example, how to drive traffic, how to generate high-conversion-rate viral short videos, and then to the AI digital human live streaming room for undertake , striving to achieve more comprehensive chain coverage and help clients complete the closed loop of product sales using digital humans.

To complete this closed loop, they have also opened up new product directions. Currently, AI assistants are becoming new traffic entry points for various services and consumption. Cognitact has built a SmartAds platform that connects to ChatGPT's paid advertising and became an official partner of OpenA in July, using its own algorithms to help overseas enterprises improve the effectiveness of ChatGPT ad placements. In addition to direct effect conversions, the acquired traffic can also flow to website digital human customer service or viral short videos on social platforms for content matching. Ultimately, Cognitact's digital human business may form a new closed loop from traffic acquisition to sales conversion with them.

04

Exploration and Inquiry into the Last Mile of Implementation

The adjustments made by service providers have, to some extent, answered the questions of what to sell and how to sell under the new trends. However, enterprise-level AI still faces the issue of how to truly be used in business.

For example, the demo may be excellent during POC, but the effectiveness is always lacking during actual use. The industry believes the crux lies in the fact that POC typically verifies a small-scale, single-point demand, while true implementation in enterprise scenarios involves not only single-point demands but also people, rules, and a large amount of tacit knowledge within the enterprise.

These various gaps are commonly referred to as the "last mile of AI implementation" in the industry. Shuzhi Qianxian observes that service providers have already begun to diverge in their solution paths.",

An enterprise service professional told Shuzhi Frontline that the greatest value of FDE lies in its extensive experience with numerous products and enterprises. Typically, FDEs possess a deep understanding of both AI products and the pain points of enterprise operations. They can leverage their experience to help enterprises screen relevant scenarios, manage expectations, and ensure subsequent delivery and implementation. For large enterprises that are unsure of their specific goals but still want to utilize AI tools, the role of FDEs is to accompany clients through a certain phase while enabling them to eventually operate independently.

However, some believe that in the domestic market, due to differences in payment capacity and willingness between enterprise markets in China and overseas, it remains to be seen whether the bridge role connecting business needs and AI capabilities will be primarily led by service providers or generated internally within enterprises.

'Palantir's average customer price starts at $5 million, and there are few clients in China with such payment capacity and willingness,' said Fang Yi, founder of Meiri Hudong. He believes that the FDE model will not be well-suited to the Chinese market. Instead, he advocates for bringing clients' personnel to service providers for training, enabling them to work independently, and then returning them to the enterprise to transform the organization. Fang Yi refers to this model as CCE (AI Collaborative Creation Engineers).

Fang Yi predicts that in the AI era, private data is becoming a key barrier to AI intelligence and usability. For many enterprise users, AI must be deployed where the data resides, rather than moving the data to the cloud. At the same time, since enterprises need to cultivate AI talent with business acumen internally, they require a set of AI application development tools. Therefore, they have designed a multi-layered capability combination of 'AI programming capabilities + workstation hardware + trusted delivery' to meet market demands for AI implementation.

Fang Yi introduced that currently, Meiri Hudong's Tongren Programming lowers the barrier to application development, allowing non-development personnel within enterprises to develop applications. The process of co-creation with CCE is based on Tongren Programming, and the developed applications can be deployed on the Gezhi Intelligent Workstation, which aligns with the procurement habits of government and enterprise clients and meets the requirement of keeping data within its original domain.

There are also views in the industry that cultivating AI implementation capabilities internally within enterprises, in addition to avoiding the high costs and scalability issues of FDE, may also enhance internal motivation and prevent issues where changes in Agent rules render them ineffective.

The service model of Linghe Shuzhi's CASM (AI Customer Success Team) also follows the approach of 'teaching others to fish.' This approach consists of three steps: AI consensus, AI co-creation, and AI symbiosis—first managing expectations and clarifying boundaries, then conducting business interviews and goal design, and finally accompanying the enterprise through organizational-level adjustments. 'Every scenario created is like a pearl. When the pearls are strung together, the organization will undergo transformation,' Wang Ting believes that by adopting a slower approach, they can avoid the current industry situation where many AI sales teams withdraw after delivering a single scenario, leaving clients unable to truly utilize it, with scenarios lacking interconnection and sales teams constantly chasing new deals.

However, regardless of the path chosen during implementation, the fate of AI products after delivery ultimately points to changes within the organization itself.

Wang Ting believes that one cannot focus solely on the short-term costs of AI. Indicators such as process structuring and the transformation of individual capabilities into organizational capabilities require attention. 'Organizations also need to build their muscles.' Among the enterprises she has engaged with, leading companies have already begun to adjust their incentive mechanisms. For example, some bosses have proposed 'paying Agents wages' and distributing the cost savings generated by AI as incentives to those who trained it.

Sun Linjun, founder of Zhipu AI, mentioned that over the past few months, their primary focus has shifted from product functionality to 'organizational transformation.' He observes that client demands are now transcending the tool level. 'The logic used to be simple: if you bought our product, we would reduce headcount and replace work hours. Now, many people are considering how to transform into an AI-driven organization.'

As for how to forge an organization more adapted to the AI era, that is another, even larger, topic. Essentially, AI is bringing about a paradigm shift in development. The clear and stable SLA, delivery agreements, and delivery models of the previous generation of software are being disrupted, and both service providers and enterprises require time to dynamically adapt to this transformation.