您正在使用IE低版浏览器，为了您的雷峰网账号安全和更好的产品体验，强烈建议使用更快更安全的浏览器
此为临时链接，仅用于文章预览，将在时失效
0
导语：“太短的决策周期，太多的备选工具，从一开始就造成了太多的数据孤岛。”
10月20日，在以“助力中国企业全球化成长”为主题的Zoho北京媒体沟通会上， Zoho不仅向外界展示了其最新的企业级AI全景战略，更试图回答一个当前行业的普遍痛点：当AI大模型喧嚣了两年之后，为什么大多数企业在业务中依然觉得AI“用不起来”？
Zoho全球副总裁Mathivanan一针见血地点出了症结所在：“太短的决策周期，太多的备选工具，从一开始就造成了太多的数据孤岛。”当企业盲目堆叠各种带有AI标签的软件，不仅没有打破原有的信息孤岛，反而催生了“AI孤岛”。
Zoho中国COO夏海峰在随后的主题演讲中直言不讳：“如果AI只能回答问题、提供信息，却无法参与实际业务流程，它对商业的价值就微乎其微。”
这也揭示了企业AI应用面临的一道关键门槛：能够回答问题，不等于能够完成任务。AI要真正参与业务，不仅需要理解用户的需求，还需要获取相关数据、调用业务系统，并在权限和流程约束下完成具体操作。
为了跨过这道从“懂回答”到“能执行”的鸿沟，Zoho在此次会上系统展露了其沉淀多年的AI全景三层架构：底层给数据，中间层做连接，上层管执行。
在Zoho的AI全景图中，最上层是直接触达业务的核心。Zoho认为，真正的AI分水岭在于它能不能接下任务，并跑通多步骤流程。
首先是原生嵌入业务场景的对话式AI助手AskZia。它并非独立的外挂工具，而是深植于CRM、财务、客服等系统内。用户可以通过自然语言获取信息、分析业务数据，并在相应功能支持下完成内容起草与润色等工作。
但“会说”只是第一步，“能干”才是企业真正的刚需。Zoho推出了预置智能体Zia Agents，销售教练、SDR机器人等上百个开箱即用的角色，专门处理那些有规律、可复用、吃经验的工作。
而对于规则极其复杂、个性化极强的场景，Zoho提供了Agent Studio低代码平台。例如，一家汽车零部件客户就利用该平台，开发了专属于自身的成本预测智能体，有效应对国际贸易中由关税、运输等因素造成的材料成本波动。这种让非技术人员像搭积木一样定制专属AI的能力，正是Zoho区别于标准化SaaS产品的差异化护城河。
从Ask Zia到Zia Agents，再到Agent Studio，三者分别对应信息获取与处理、特定任务执行，以及面向个性化流程的智能体构建。它们的区别不只在于AI能力的强弱，更在于AI参与业务的方式发生了变化：从辅助员工完成工作，逐步延伸到承担特定任务。
如果说AI执行层是冰山露出水面的部分，那么数据和连接则是水面之下的庞大基座。在中间层，Zoho通过MCP（模型上下文协议）打通了内外生态。
夏海峰在会上描绘了这样一个场景：用户在WorkBuddy、豆包工作、千问办公等支持MCP的客户端中，配置自定义连接器，接入Zoho MCP，即可在外部直接对话操作Zoho CRM，实现数据的自动更新。Zoho MCP目前兼容Claude、ChatGPT等主流大模型与智能体，并提供300余款第三方服务的集成清单。通过标准化协议，Zoho不仅让自家的业务能力能够被各类外部智能体调用，更确保了敏感数据在离开系统前就被遮蔽与鉴权，实现了真正的开放与安全并行。
MCP的意义，在于为AI工具连接外部应用提供标准化的交互方式。对企业来说，这意味着AI的使用场景不必局限于单个软件内部，也可以向跨应用的工作流程延伸。
在算力与模型动辄追求百亿甚至上万亿参数的今天，Zoho在底层选择了一条极为务实的路径。Mathivanan表示，Zoho的投入不跟风竞争对手的烧钱规模，而是按客户真实需求决定。相较于消耗巨大资源的通用大模型，Zoho更倾向于自研部署在自有数据中心的中小模型。“例如自动回复邮件、图片内容总结或者情感分析等场景，只需要部署适合的模型即可，这不仅提高了业务精准度，也极大降低了推理成本。”
这种对业务边界和成本的极致把控，也体现在AI编程的应用上。面对AI写代码是否会取代传统软件的疑问，夏海峰用“冰山理论”给出了回答：AI编程只是冰山上露出水面的一角，源码托管、权限设计、安全运维、底层数据治理这些看不见的部分，才是真正下功夫的地方。Zoho将AI编程能力赋能给用户，是为了让他们在现有的、稳固的业务底座上做扩展，而非凭空去造一个悬在半空的应用。
在商业模型上，Zoho的选择同样“反直觉”。当部分厂商试图将AI能力作为独立套餐向客户收取高昂费用时，Zoho在会上主张，将AI融入产品体系，而不是简单地将其视为一项独立的增值功能，目前并不会向客户额外收AI费用。
这一决定的背后，是Zoho坚持了30年不融资、不上市的长期主义底气，对于企业级AI而言，底层投入的意义不仅在于提供算力，还在于支持产品运行、数据管理和不同业务应用之间的协同。
数据安全则是另一项不能绕开的基础要求。Mathivanan将其定义为“过渡期的底线”：不管是用企业自带的API Key接入外部模型，还是使用Zoho自研的模型，Zoho都承诺不拿客户数据训练通用模型，所有的技术探索最终都指向一个目标——保护客户的数据安全与隐私。
因此，企业部署AI不能只关注模型能力，还需要结合实际应用场景评估数据处理方式、权限配置和外部服务的接入边界。对软件厂商而言，AI能力的开放程度与数据保护机制并不是相互独立的两件事：只有明确访问范围和执行权限，企业才更有条件将AI用于真实业务。
正如夏海峰在演讲最后所总结：“随着智能体具备调用API和自主执行任务的能力，企业软件不再只是供人操作的功能工具，同时成为承载智能体开展业务的底座。企业上AI，不是选一个更聪明的模型，而是决定把哪些环节交给机器，以及为此准备好什么样的底座。Zoho要做的是把"AI + 自主执行"真正落地到企业全场景，而不是讲故事。“
从AI助手到智能体，从模型选择到系统连接，Zoho正在将AI能力融入企业现有的产品与业务流程，让AI从辅助信息处理，逐步走向具体任务的执行。
AI正在改变企业软件的使用方式，也在推动SaaS重新思考自身的价值。未来SaaS的竞争，不仅在于提供多少功能，还在于能否让AI真正参与业务，并将技术能力转化为实际的业务价值。
雷峰网(公众号：雷峰网)
为了您的账户安全，请验证邮箱
请验证您的邮箱
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
You are using a low-end IE browser and strongly recommend a faster and safer browser for your Thunder Peak account security and better product experience.
This is a temporary link, only for article preview, which will expire at the time
0
Introduction: “Too short decision-making cycle, too many alternative tools have created too much data isolation from the outset.”
On October 20, at the Beijing Media Communication Conference in Zoho, on the theme of “Facilitating the Global Growth of Chinese Enterprises”, Zoho not only showed the outside world his latest corporate-level AI Panorama strategy, but also tried to answer the general pains of a current industry: Two years after the AZLLM noise, why do most firms still feel AI “unuseful” in their business?
Zoho's global vice president, Mathivanan, has the crux of a blood spot: “Too short decision-making cycle, too many alternative tools, has created too many data islands from the start.” Instead of breaking the old information island, the "AI island" was created when companies blindly stacked all kinds of AI-labeled software.
Zoho, China's COO Summer Peak, put it bluntly in the ensuing keynote address: “If AI can only answer questions and provide information but is unable to participate in actual business processes, it will be of little value to business.”
It also reveals a critical threshold for enterprise AI applications: being able to answer questions does not mean being able to fulfil a mandate. For AI to be truly involved in business, it needs not only to understand the needs of users, but also to access relevant data, to access business systems and to perform specific operations under permission and process constraints.
In order to bridge the gap between “understanding” and “implementable”, Zoho systematically presented its three-tier AI panorama structure, which has been sank for many years: bottom to data, middle to connect, upper to upper tort.
In Zoho ' s AI Panorama, the top layer is at the core of direct engagement. Zoho believes that the real AI watershed lies in its ability to take over the mission and run through multiple steps.
The first is the dialogue-based AI assistant who was embedded in the business scene. It is not a stand-alone outreach tool, but is embedded in systems such as CRM, finance, customer service, etc. Users can access information in natural languages, analyse business data, and do content drafting and colouring with the corresponding functional support.
But “will say” is only the first step, and “capable” is the real business demand. Zoho has introduced pre-positioned AI agentia Agents, which sells hundreds of open-boxing roles such as coaches and SDR robots, which specialize in regular, reusable and experienced jobs.
Zoho provided Agent Studio ' s low-code platform for a highly complex and personalized set of rules. For example, a car spare parts customer used the platform to develop its own cost projection AI agent to respond effectively to material cost fluctuations in international trade caused by tariffs, transport, etc. This ability to customise the exclusive AI of non-technicals, like building blocks, is the difference between Zoho and the standardized SaaS product, the differentiated moat.
From Ask Zia to Zia Argentinas, to Agent Studio, each corresponds to the acquisition and processing of information, the performance of specific tasks, and the construction of AI agent for personalization processes. Their difference lies not only in the strength and weakness of AI, but also in the change in the way it is engaged in business: from supporting staff to carrying out work, to taking on specific tasks.
If the AI enforcement layer is part of the surface of the iceberg, then data and connections are a large subsurface. At the intermediate level, Zoho has managed to access internal and external ecology through MCP (model context agreement).
At the meeting, the Summer Peak painted a scene where users, among their clients supporting MCP, such as the WorldBuddy, Doubao jobs, and the Sun Query Office, configured a custom connector to access Zoho MCP to operate Zoho CRM in direct external dialogue to achieve automatic data updating. Zoho MCP is currently compatible with mainstream LLM and AI agent, such as Claude, ChatGPT, and provides an integrated list of over 300 third-party services. Through the standardized agreement, Zoho not only allows its own operational capacity to be accessed by various external AI agents, but also ensures that sensitive data are concealed and protected before leaving the system, achieving real openness and security in parallel.
The significance of MCP is to provide a standardized and interactive approach to the external application of AI tools. For an enterprise, this means that the use of AI need not be limited to the individual software, but can also be extended to the workflow process across applications.
Zoho has chosen an extremely pragmatic path at the bottom of the ground, today, when computations and models are moving in pursuit of billions or even trillions of parameters. Mathivanan states that Zoho ' s input is not the size of the price set by the competition, but is determined by the real needs of the client. Zoho prefers self-study of small and medium-sized models deployed in its own data centres to the generic LLM, which consumes enormous resources. “As for example, an automatic response to e-mails, photo summaries or emotional analysis, it is sufficient to deploy a suitable model, which not only increases operational accuracy but also significantly reduces the cost of reasoning.”
This extreme control of operational boundaries and costs is also reflected in AI programming applications. Facing the question of whether the AID writing code will replace traditional software, the summer summit answered with an “icing theory”: The AI programming is just the tip of the surface of the iceberg, and the invisible parts of the source-code hosting, permission design, security, bottom-level data management are the real places to go. Zoho has given the AI programming capacity to users to expand on existing, secure business floors rather than to create a half-spaced application.
In the commercial model, Zoho's choice is equally “anti-intuitive”. When some manufacturers attempted to charge their clients high fees for AI capabilities as a stand-alone package, Zoho argued at the meeting that AI would be integrated into the product system rather than simply treated as a stand-alone value-added function and that no additional AI fees would be charged to clients at this time.
Behind this decision was Zoho ' s persistence in a long-term, non-financing and unmarketable background for 30 years, and the bottom-up input is not only about providing calculus but also about supporting synergies between product operations, data management and different business applications for the enterprise-level AI.
Data security is another fundamental requirement that cannot be bypassed. Mathivan defines it as “the bottom line of the transition period”: whether using the company's own API Key access to external models or using Zoho's self-study models, Zoho is committed not to using a generic model for customer data training, and all technological exploration ultimately points to one objective - to protect customer data security and privacy.
Therefore, an enterprise ' s deployment of AI cannot only focus on modelling capacity, but also needs to assess the data-processing approach, permission configuration and access boundaries of external services in conjunction with the actual application of the scene. For software manufacturers, the level of openness of AI ' s capabilities and data protection mechanisms are not separate from each other: only with clear access and permission implementation will an enterprise be better equipped to use AI for real business.
As Summer Peak concluded at the end of his speech: “As AI agent has the ability to call API and to carry out its tasks on its own, enterprise software is no longer just a human-operated functional tool, but at the same time a base for AI agent operations. It's not about choosing a smarter model at the corporate level, but about deciding which links to the machine and what floors to prepare for it. Zoho is going to put "AI + Self-Enforce" on the ground, not tell stories. “...
From AI ' s assistant to AI agent, from model selection to system connectivity, Zoho is integrating AI ' s capabilities into the existing products and business processes of the enterprise, allowing AI to move from ancillary information processing to the implementation of specific tasks.
AI is changing the way in which enterprise software is used and is also driving SaaS to rethink its value. Future SaaS competition lies not only in the number of functions to be provided, but also in the ability to engage AI in real business and to translate technical capacity into real business value.
Thunder Peak Network (WeChat Official Account:Ray Peak Network)
For your account security, please verify the mailbox.
Please verify your mailbox.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.