The newly streamlined chain embodies Kingdee's continuous reflections and evolving expressions over the past two years.

This includes a reinvigorated understanding of AI's "result delivery" standards, a concerted push towards the new capabilities of agent orchestration processes, and a fresh articulation of Kingdee's extensive data, permission comprehension, and enterprise management process insights accumulated over more than 30 years. These facets collectively form Kingdee's novel approach to enterprise management AI.

|Ye Pi

|Industry Insider

"After the launch of DeepSeek earlier this year, we visited numerous customers, and most have internally deployed large models," said Liu Zhongwen. "While scenarios like knowledge management yield results more easily, many enterprises haven't achieved ideal outcomes in management scenarios such as data analytics."

As Vice President of Kingdee China and General Manager of the R&D Center, he harbors deep insights into the unique AI environment within domestic enterprises.

Indeed, if 2023 marks the inaugural year of AI large models, aligning with people's unwavering belief in the AI era's arrival, then 2024 and 2025 can be seen as the years of large model implementation, reflecting the growing number of managers eager to leverage AI to upgrade their enterprise management frameworks.

According to a Gartner survey on enterprise AI, 93% of participating enterprises believe AI will propel revenue growth by 2025, while 66% fear missing out on generative AI opportunities will significantly challenge their medium- and long-term business development.

However, this is no easy feat. A separate enterprise AI survey report by Xiangjiang Digital Review reveals that over half of surveyed enterprises reported ineffective AI technology applications, citing mismatched business needs (87%), insufficient data quality (80%), and a lack of security audits (61.2%) as reasons.

Over the past two years, these inevitable challenges and needs have served as the backdrop for Kingdee's relentless promotion of its "All in AI" strategy. Simultaneously, it has sparked renewed contemplation for Kingdee, a leading software vendor in China: What kind of AI products can genuinely enhance enterprise productivity? What approach should enterprises adopt to integrate into the AI system? Alternatively, what should be the new pace of advancement in enterprise management in today's AI era?

"If our previous AI products were more like copilots, these products are now true end-to-end agent solutions that enterprises can readily use," said Li Fan, President of Kingdee China and General Manager of the Kingdee Cloud Platform.

In 2025, heralded as the first year of agent implementation, Kingdee is providing an answer. Or rather, this software industry leader is attempting to refresh the Chinese enterprise management model with AI.

I. The Surging "Enterprise Management AI Wave": Visible Needs, Unavoidable Pain Points

"We've tried many AI large models, but applying them to scenarios requiring efficiency improvements, like marketing, production, business travel, and finance, has been challenging," a CIO from a prominent domestic retail enterprise told Industry Insider. "Now we're reassessing how to utilize large models."

This is not an isolated case.

Over the past two years, it's no exaggeration to say that AI implementation within enterprises has been mixed. On one hand, AI technology has consistently surpassed existing capabilities across various rankings, with numerous general-purpose and scenario-specific agent products emerging and being hailed as "disruptive" by the capital market. On the other hand, for enterprises, besides experiencing the convenience of general-purpose AI in scenarios like meeting assistants and knowledge Q&A, AI has shown unexpectedly "low intelligence" and "unreliability" in core enterprise management scenarios such as finance, business travel, and human resource recruitment.

What are the reasons? Alternatively, behind the rankings, what should be the formula for converting the continuously upward-trending AI technology curve into a new enterprise management model?

"AI technology is just one factor in the evolution of enterprise management; there's also a series of engineering capabilities, including security, permission management, scenario understanding, and more," Liu Zhongwen explained.

Indeed, this is also the core difficulty enterprises currently face when implementing AI in management scenarios. Whether building in-house solutions or leveraging general-purpose agents like Manus, they inherently lack genuine industry and enterprise expertise.

These expertises, specific to enterprise management scenarios, are also the core underlying values Chinese SaaS enterprises have continuously refined over the past decade, encompassing process and organizational management for different links, control over content and data security and trustworthiness, personalized services for various enterprises and industrial scenarios, and more.

In reality, from an AI perspective, the challenges extend far beyond these. When implementing agents in specific professional scenarios, enterprises often follow the construction chain of "model - data - agent," with the data system's construction being particularly crucial. For most enterprises, model training and fine-tuning are no longer difficult, but constructing a data and process system genuinely fitting the corresponding scenario is not a capability they possess.

Moreover, for most enterprises, the challenge lies in embedding agents into existing software processes. That is, for most enterprises, agent intelligent bodies need to be appropriately embedded and orchestrated with inherent SaaS software processes. Only then can agent intelligent bodies truly become nodes flowing seamlessly within the chain of enterprise management scenarios.

For most enterprises, whether it's the development details of agents themselves or their integration and cooperation within the enterprise, these are becoming real hurdles for AI implementation. Only by overcoming these hurdles can enterprises truly build their own AI productivity system.

As mentioned, this is no easy task. Over the past year or two, some vague answers have emerged, such as technical agents tailored for specific scenarios, helping enterprises embed agents within the enterprise based on the "low-code + AI" model, and assisting enterprises in building professional production knowledge bases. However, these answers are not optimal for enterprises, making it difficult for them to become out-of-the-box, end-to-end capability enhancements.

What should be the correct approach for AI to enter enterprise management scenarios? Alternatively, what should be the new model for enterprise management in the AI era? Where is the genuine AI entry point for enterprises?

II. Kingdee Provides an Answer

"Each of our agents is built by numerous technical agents. For example, behind the financial report agent, there are 16 agents, and other agents are also composed of many agents," Li Fan revealed.

This is Kingdee's new answer.

At the recently held 2025 Kingdee Cloud AI Summit, Kingdee officially unveiled five agents: Golden Key Financial Report (financial report analysis agent), ChatBI (enterprise data analytics agent), Recruitment Agent, Business Travel Agent, and Enterprise Knowledge Agent.

Specifically, Golden Key Financial Report, a novel financial report analysis agent, can help managers and financial personnel swiftly obtain financial report information in a very short time. Simultaneously, it can output comparisons of financial report operating data between different enterprises based on the agent, facilitating enterprise managers in understanding their own operating levels.

ChatBI (enterprise data analytics agent) corresponds to the most prevalent data BI scenario in China. Enterprises can independently mine data value through conversational chat, obtaining visual results in seconds from questioning to insight. Compared to previous complex code configurations, Kingdee's ChatBI agent is sufficiently accurate and efficient.

Recruitment Agent, Business Travel Agent, and Enterprise Knowledge Agent also target core management scenarios within enterprises: recruitment, business travel, and knowledge base construction. Based on these three agents, enterprises can directly complete the corresponding process links. For instance, with the Recruitment Agent, enterprises can complete the initial screening of talents, significantly enhancing recruitment efficiency.

The Business Travel Agent can formulate plans based on the enterprise's business travel rules and employees' corresponding needs, helping the enterprise complete the entire process from business travel plan formulation to final reimbursement in one stop, with the entire link closed-loop completed based on the agent.

Equally crucial is the Enterprise Knowledge Agent, aimed at the knowledge base scenario widely recognized as having the strongest certainty for AI implementation in today's large model wave. That is, based on the Enterprise Knowledge Agent, enterprises can reorganize internal materials based on special settings and formats embedded in the agent, thereby building an accurate, clear, and intelligent knowledge base for themselves.

"Where Kingdee differs from other market providers is that we have years of data accumulation and scenario understanding in corresponding links. These accumulations fundamentally involve understanding and classifying individual metadata formats at the bottom level. Based on these new metadata models, we provide agents that enterprises can genuinely use out of the box," Li Fan explained.

If the release of the five agents corresponds to Kingdee's answer to the most widely used AI scenarios in enterprises, then the 2.0 upgrade of the Kingdee Cloud Agent Platform can be seen as Kingdee's underlying capability and system enhancement to enable enterprises to truly leverage AI.

Compared to the Kingdee Cloud Agent Platform 1.0, this newly launched 2.0 version at the center stage boasts impressive upgrades across various AI engineering capabilities.

For example, richer templates and tools. Enterprises can swiftly develop enterprise agents through a vast array of task flow templates, prompt word templates, and preset tools. Another example is deeper SaaS integrations, addressing the pain points faced by most enterprises mentioned earlier, where agents are often in an "isolated island" state and cannot be integrated into the enterprise's existing processes. Based on the Kingdee Cloud Agent Platform 2.0, enterprises can seamlessly complete the access and integration between agents and their own SaaS software.

Additionally, there's a more secure enterprise-level platform and more open technical standards. The former also aligns with the necessity of security compliance for all agent and software value expressions within the enterprise, such as permissions, privacy protection, and content security, which are the most basic controls for agents to genuinely land in enterprise scenarios; the latter corresponds to enterprises' ability on the Kingdee Cloud Agent Platform 2.0 to connect to external resources from the model layer based on the MCP/A2A open protocol, akin to using "USB-C".

"We now emphasize shifting from a product-oriented approach to a results-oriented approach, judging our products from whether they can deliver results to enterprises," said Li Fan.

It's evident that whether it's the five updated agents or the upgrade of the Kingdee Cloud Agent Platform 2.0, behind these products on stage are genuine and feasible AI productivity tools. With the support of these agents and platforms in core scenarios, enterprises can build a newer enterprise management AI system, enhancing their inherent management efficiency through the combination of "humans + software + agents".

The newly streamlined chain embodies Kingdee's continuous reflections and evolving expressions over the past two years—this includes a reinvigorated understanding of AI's "result delivery" standards, a concerted push towards the new capabilities of agent orchestration processes, and a fresh articulation of Kingdee's extensive data, permission comprehension, and enterprise management process insights accumulated over more than 30 years. These facets collectively form Kingdee's novel approach to enterprise management AI.

III. 2025, Chinese Enterprises Officially Enter the AI Era

"For Kingdee, there are also many challenges. For instance, shifting from product delivery to result delivery requires changes in many internal product logic, delivery processes, and development logic," said Li Fan. "And defining the results that everyone recognizes, and forming these criteria, is still a work in progress."

Yet, behind this challenging yet resolute transformation lies Kingdee's new thinking on enterprise management models. "We believe that in the AI era, the enterprise's digital and intelligent platform will evolve into an enterprise management AI, from ERP to EBC, and then into today's EMAI (Enterprise Management AI)," said Zhao Yanxi, Executive Vice President of Kingdee China and President of the Business Center, at the conference.

One interpretation of the EMAI concept is that it corresponds to utilizing AI to digitally and intelligently reconstruct a series of management scenarios within the enterprise, gradually transitioning from the inherent "human + software" model to the "human + agent" model, truly making AI the driver and executor in enterprise management scenarios and helping enterprises build new production chains based on intelligent attributes.

This is also the new underlying philosophy behind Kingdee's product launch this time.

That is, whether it's the introduction of five new agents or the upgrade to Kingdee Cloud 2.0, beyond the products themselves, Kingdee's ultimate goal is to assist enterprises in genuinely identifying entry points for AI integration. These include financial report analysis, business travel management, agent invocation leveraging MCP, and the seamless integration of new agents with existing processes. All these efforts align with helping enterprises take a leap forward into AI, building upon their existing enterprise management frameworks – transitioning from the era of assemblable software to the age of enterprise management AI. This transition positions AI not merely as a technology provider but also as an executor and catalyst for specific processes.

"In the first half of this year, we engaged in exchanges with numerous enterprises, and every chairman and executive expressed keen interest in AI and sought out potential breakthrough scenarios," said Zhao Yanxi. "However, from the current perspective, the implementation of AI varies significantly across different fields and systems."

This aligns with Gartner's AI business speed stratification model, which predicts that record systems, differentiated systems, and innovative systems will be progressively reconstructed by AI. For enterprises, this presents an optimal starting point: beginning with AI experiments on foundational, universal systems, then implementing them within their own enterprise systems and specific industrial links, ultimately reaching AI-native scenarios and constructing a comprehensive EMAI system.

This journey is lengthy and challenging. It involves enhancing AI agents for specific industrial links, developing a low-threshold AI component building platform tailored to enterprise needs, managing different orchestration processes between agents and traditional software, and invoking and centrally managing external agents based on the platform. Enterprises must navigate these complexities amidst the new wave of AI evolution.

"We now possess an AI implementation methodology, abbreviated as 'AIGO'. These four letters correspond to Analysis and Architecture (A); Implementation and Execution (I); Governance and Management (G); and Operation and Optimization (O). This includes assessing the current state, needs, vision, and strategy of enterprise digitization, designing the enterprise management AI architecture, and managing the digitization implementation process. It provides enterprises with a clear path for AI transformation," explained Zhao Yanxi.

This is a pivotal challenge for Kingdee. Transitioning from a software service provider to an AI service provider necessitates not only a high level of abstraction and new expressions of past capabilities but also the exploration of novel technical engineering systems, measurement standards, and implementation models. In fact, this has been Kingdee's primary focus over the past few years, as evidenced by this latest launch.

Notably, this focus has garnered external recognition. According to Gartner's 2024 report, Kingdee ranks among the top ten in China's market share for generative AI models, and among the top ten, Kingdee is the sole enterprise management software vendor.

It's evident that not only has Kingdee's external service model evolved but so has the company itself. Within Kingdee, the AI Innovation Award is presented monthly, and all employees are encouraged to participate in the AI Speech Contest. Furthermore, "AI First" has been established as the core enterprise development strategy.

"Starting this year, we will embark on a new journey. Instead of 'smashing' through challenges, we will plant a tree every year – the tree of enterprise management AI. We firmly believe that this industry will undergo a transformative shift due to AI," said Xu Shaochun, Chairman and CEO of Kingdee Group.

In 2025, Chinese enterprise management will accelerate its stride into the AI era.