本篇文章，我们将通过主流咨询公司的数据，看看中外企业的人机协同团队建设情况。看到最后会发现，现状可能和大多数人印象中的不一样。
01 渗透程度——AI技术广泛普及
首先看AI智能体的普及。我们前面的定义中，人机协同团队中，至少要有一个AI智能体被赋予固定职责，与上下游形成稳定工作关系，并拥有一定程度的决策权，以至于能够影响一个作业单元。从这个意义上说，大多使用AI智能体的企业，至少在一个组织局部范围内建起来人机协同团队。
我们首先引用两个国外机构的数据。
斯坦福HAI的数据说明，AI技术在企业内正在迅速普及。2024年78%的组织报告使用AI（2023年为55%），生成式AI的使用率从33%跃升至71%（Stanford HAI，2025）。
而麦肯锡2025年全球调研则显示，相对于AI技术的广泛应用，其应用深度严重不足。
从AI技术的普及上看，88%的组织已在至少一个职能中使用“AI工具”（上年78%）；再提高一下门槛，62%的组织至少试验过“AI智能体”；但是，仅有23%的企业在至少一个职能中跑通了规模化部署；而且，没有任何单一职能（如人力、财务类职能）内，各大企业在AI智能体规模化上的比例超过10%（McKinsey，2025）。
回到中国企业，穆胜咨询的数据显示，对于AI技术的应用似乎要保守很多，广度和深度都不让人乐观。
根据穆胜咨询的《2026中国企业平台型组织建设研究报告》，61.7%的企业使用AI的情况比较相似：一方面，大多企业都是部分员工在使用AI工具，另一方面，在这些企业里使用AI工具都是心照不宣的秘密。 换言之，AI工具在这类企业里并未完全普及，但也绝对不是稀缺品；大多员工不愿意公开承认，但也不会坚决否认。
至于搭建AI智能体，并接管部分职能，似乎并不在中国企业近期的计划里。当我们问道“贵司计划在1年内用AI工具替代多少现有岗位”时，66.7%的企业回答“暂无计划”，22.5%的企业回答“小于10%” ，两者合计比例为89.2%，这群相对保守的企业比例，相对2025年的81.7%反而上升了7.5%。
02 人力削减——假装“AI导致裁员”
再看AI技术的普及是否导致了裁员。裁员一方面降低了人工成本，直接提升了企业利润，另一方面也减少了人类员工的数量，客观上带来了人机混合（而非协同）的效果。在舆论中，这种“AI挤出人类员工”的现象已经大量发生，甚至还有媒体将“人类对抗AI”的故事写得有模有样，但现实真的如此吗？
Gartner给出了一个极其重要的纠偏数据：2025年上半年仅1%的裁员源于AI提升了员工生产力（Gartner，2026）。换言之，“AI导致裁员”的舆论热度，远远领先于“AI真的产生了足以裁员的生产力”这一事实。
事实上，对于AI是否足以替代人类，国外企业在实践后的认知已经开始趋于理性。德勤的数据显示，39%的商业领导者曾因AI裁人，但其中55%认为这是一个错误的决定（Deloitte/Orgvue，2025）。Salesforce和Klarna都曾大刀阔斧地在客服领域推动“AI换人”，但却发现这导致了服务质量下降，人类员工始终有AI无法替代的部分，于是，最终重新招聘了人类员工。
尽管硅谷企业大量以AI替代为名进行裁员，但这种现象在中国企业中并未大量发生。这种状态并非是基于实践后的理性，更有可能是实践尚未展开，或者对于“AI换人”的前景并不看好。
穆胜咨询的《2026中国企业平台型组织建设研究报告》用国内数据更深刻地揭示了“AI导致裁员”的真相。其中，这类裁员比例大于10%的企业仅有2.0%，可以忽略不计，比例小于10%的企业仅有17.6%。也就是说，仅有少量企业以AI替代为名进行了裁员，裁员比例也普遍小于10%。而且，这些企业中仅有极少部分（占比8.3%）企业进行了员工技能蒸馏。这说明，媒体上广泛宣传的”蒸馏员工”的现象更像是一个炒作流量的噱头。反过来说，也许大量员工并没有那么值得被蒸馏。
从员工的角度来看，他们的确有对抗AI的应激反应，主要体现在“找理由不使用AI工具”和“不承认AI给出的结果”，这部分人占比为45.0%。这种消极应对说明，AI智能体和人类员工之间的对抗，也许并不像媒体描述的那么激烈。
03 职能进展——哪些领域正在人机协作？
我们收集了全球的企业案例，发现人机协同团队主要在客服、研发、后台职能（HR等）、销售营销、供应链五个领域得到了积极应用。
我们发现，五类职能的应用中，人机协同的成熟度不取决于“AI有多强”，而取决于该场景的“例外处理机制”是否健全、“质量底线”是否被守住、“判断标准”是否可解释。这三者构成了人机协同团队能否长期稳定运行的三根支柱。
2025年，穆胜咨询以68家中国中大型客户企业为样本，针对老板或CEO（老板代理人）进行了一次问卷调研。在问卷中，我们针对一家企业的多个常见职能领域，计算了职能复杂程度（包括数据复杂性、决策独立行和决策封闭性三个维度）和AI适用效果（包括AI引用范围、AI实施效果和AI影响程度三个维度）。
在此基础上，我们计算了一个名为“导入AI积极度”的指标，即“导入AI积极度=AI适用效果/职能复杂程度”。这个指标很明确地说明了职能部门在不同的“先天条件”下导入AI的进展，直观来看，这个指标越大，越说明该职能积极地导入了AI，反之，则是对AI存在抵制太多。
结果显示，并不完全是那些易于使用AI的职能部门会导入AI，而是那些受到最大压力，且易于使用AI的部门会积极导入AI。说白了，从趋利避害的角度来看，所有职能部门的从业者都希望非标化，希望这块职能必须要依赖自己，但推动标准化，甚至引入AI的，还是那些承受压力最大的部门。一方面，法务、质量标准、采购、金融等职能部门的效率早已保守诟病，他们也有条件导入AI，其积极似乎可以理解；另一方面，生产/业务/经营/运营、财务、人力资源这类职能则是因为难度太大，在导入AI上尚需努力；除此之外，办公室、战略、研发、审计监察因为种种原因，被从业者们导向了凭手感的操作，AI自然是难以渗透。
结合麦肯锡的关于AI智能体尚未在大多企业内渗透到职能领域，以及在单项职能里并未普及的数据。我们推测，在国内人机协同团队的生根发芽，尚且有很长一段路要走。某些行业（如科技行业）的某些先锋企业有意愿在某些职能里发展人机协同，但其他企则业更像是压力之下的应对，而非寻求发展的主动为之。这些企业并不会完全因为某个职能容易导入AI智能体而积极行动，他们的积极性更大程度依赖于外部压力。
麦肯锡的国际样本里，AI高绩效企业（仅占6%）在多数职能中规模化部署智能体的比例，是其他组织的三倍以上，这些先锋企业必然带动趋势。但在我们的中国样本里，并没有太多的先锋企业给出让人信服的数据，即使像Salesforce和Klarna这类让人耳熟能详的“折腾案例”也没有。在人机协同团队的建设上，中国企业，要加油了。
04 业绩提升——是否带来了真金白银？
目前，对于AI智能体进入企业后带来的业绩提升，更多体现在成本端。如果AI智能体能够在确保工作质量的同时替代掉部分人力，“AI换人”带来的人力成本和周边成本减少就是实实在在的。
IBM用AI智能体AskHR承接了大量常规HR咨询与事务，在几年内节约了约8000人的编制，处理94%员工咨询，带来了每年6.4亿美元的结余。事实上，这类HR事务性职能是AI最易切入的场景。
但Salesforce和Klarna的案例充分说明，这种成本的削减并非没有极限，甚至可能带来负面效果。
Salesforce用AI智能体Agentforce承接了客服与开发工作，其支持团队从9000人减少至约5000人，且未新增工程师岗位，Agentforce的 ARR由8亿增加至15亿美元，增幅为240%。但在2026年4月，其招入了1000名销售岗位的毕业生。这充分说明，AI可以承接“有界、结构化”工作，人类则会回流至“无界、客户面、判断重”的岗位。
Klarna用AI智能体接管了客服工作，导致员工从2022年的7400锐减至2025年的2831左右，尽管人均收入从34.4万暴增至124万美元，但2025年5月后，他们开始重新招聘人工客服，导致Q3客服运营成本升至5000万美元。这充分说明，AI在一定程度上的确有效，但“过度依赖”却有“牺牲服务质量”的代价，质量与共情仍是人类领地。
当然，AI对于收入端的助力也有体现，但依然是在某些职能模块发挥作用。
在AI 销售 SDR（Sales Development Representative，销售开发代表）¹ 、AI 催收、AI 客服挽留/追加销售等领域，AI智能体已经开始直接作用于获客、回款、留存。
“11x AI SDR”用 AI 智能体替代或辅助人工 SDR 做自动外呼、邮件触达、约会议、线索初筛，在约90天预约了200多场会议，会议量约3倍，单会议成本降低70%。换言之，尽管这一环节未能实现收入的闭环，但却通过增加销售管道和会议量，增加了收入了可能行。
亚马逊的经典推荐引擎，是其运用最久、商业价值最直接的AI应用，贡献了35%的销售额。其核心是用算法替代了传统的“人工选品推荐”或“静态货架”模式。其AI搜索栏将访客转化率从平均2.17%提升至搜索用户的12.29%，提升了近6倍。
除此之外，亚马逊的生成式AI助手Rufus于2024年推出，2025年全面铺开，从“推荐商品”进化到“对话式导购”，直接替代了部分人工客服和导购的职能。亚马逊CEO预计，Rufus每年可带来约100亿美元的增量销售额。目前，其已积累2.5亿活跃用户，月活同比增长140%；与Rufus互动的用户，完成购买的可能性比未互动者高出60%。
至于对企业经营业绩的影响，则并没有太多的明确数据。
麦肯锡的调研显示，仅6%的受访企业可以被定义为“AI高绩效者”，即AI对于EBIT的贡献达5%以上，并看到“显著价值”。而这些企业对于AI的应用，对于人机协同团队的搭建，并不是尝试局部降本，也不是尝试局部增效，而是着眼于增长与创新。
穆胜咨询的中国样本进一步验证了这种观点。2025年以68家中国中大型客户企业老板或CEO为样本的调研显示，仅有7.4%的企业感知到AI智能体“整体提升了组织效率，并转化为经营价值”，他们在“AI适用效果”和“导入AI积极度”上分别有3个以上和4个以上的职能远高于样本平均水平（获得4分及以上的高分）。而其他样本在这两个指标上仅有1-2个高分职能，差距相当明显，换言之，他们在人机协同团队的搭建上，一开始就有整体思路，不仅是为了局部降本或增效。
这可能是对于AI、对于组织、对于人机协同团队的认知不同，造成的“动机差异”。我们在深度访谈中发现，中国老板的几种启动“AI化”或“人机协同团队”的观点很普遍，“AI肯定是趋势，公司也有盈利，可以去投资未来，先不管AI能做什么，先用起来。”“我们公司内部有几个地方是内耗的重灾区，与其让人去内耗，不如让AI把他们都替代了。”“当前我们公司人肯定是多了，先用AI这把刀来砍砍人，降降成本，增加在寒冬里生存的底气。”大多中国老板对于AI的使用，要么就是虚无缥缈，要么就是急功近利。
我们相信，如果不进行组织升级，不重新设计工作流，而是简单地将AI嫁接到原有的组织和流程里，企业无法获得AI带来的经营业绩增长。随着实践的深入，越来越多的企业或研究机构意识到，“AI换人后自动提升组织效率”的说法根本不成立，AI可以降本，在局部领域可能增效，但要把局部效率提升变成整体效率提升，且形成最后的经营结果，似乎还要考虑大量的约束条件。
微软2026年研究发现，组织条件（文化、管理者支持、人才实践）对AI成效的影响约为个人因素的2倍（67% vs 32%）；仅19%的企业进入“前沿区”，即个人能力与组织支持相互强化的状态（Microsoft，2026）。
但微软这种对于组织条件的描述，可能太泛化了，关键还是组织设计和周边制度。穆胜咨询的《2026中国企业平台型组织建设研究报告》显示，仅有26.7%的企业为了推动AI化转型，进行了组织结构调整。而其余企业认为，技术是技术，组织是组织。对于那些已经进行组织调整的企业，他们的认知显然更超前。其中，56.3%的企业推动的组织转型有明显的平台型组织特征。也就是说，一旦企业意识到导入AI技术需要进行组织变革或调整，他们大多都会走向平台型组织。
¹ 在 B2B 销售里，SDR 通常负责销售流程最前端：找潜在客户、清洗线索；冷启动外呼、发邮件、LinkedIn 触达；初步判断意向，预约会议/把合格线索转给销售（AE）；一般不负责最终成交签单。和它常对比的是 AE（Account Executive，客户经理/销售代表），AE 才主要负责跟进商机、谈判和成交。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
In this article, we will use the data of mainstream consulting firms to see how people in middle and foreign enterprises work together. At the end of the day, it would appear that the status quo may be different from what most people perceive.
01 Permeability — widespread use of AI technology
Let's start with AZAI agent. In our earlier definition, at least one of the AIAI agents in the HAB team is assigned a fixed responsibility to develop a stable working relationship upstream and downstream, with a degree of decision-making power that can affect an operational module. In this sense, most of the businesses using AZAI agent have set up teams of people working together in at least one organization.
We will first cite the data of two foreign agencies.
Stanford HAI data indicate that AI technology is spreading rapidly within enterprises. Seventy-eight per cent of organizations reported using AI in 2024 (55 per cent in 2023), and the rate of use of generated AI jumped from 33 per cent to 71 per cent (Stanford HAI, 2025).
For McKinsey ' s 2025 global study, the depth of application is severely inadequate relative to the wide range of AI applications.
In terms of universal access to AI technology, 88 per cent of the organizations have used “AI tools” for at least one function (78 per cent in the previous year); a further increase in the threshold, with 62 per cent having tested “AIAI agent”; however, only 23 per cent of the enterprises have managed to deploy in at least one function; and there is no single function (e.g. human, financial) and large enterprises have over 10 per cent of the size of AIAI agent (McKinsey, 2025).
Turning back to Chinese firms, the data from the Mu Sheng consultancy show that the application of AI seems to be far more conservative and less promising in terms of both breadth and depth.
According to a study on the construction of platform organizations in Chinese enterprises in 2026, consulted by Mu Sheng, 61.7 per cent of enterprises use AI in a similar way: on the one hand, most enterprises are part of the workforce using AI tools, while on the other hand, the use of AI tools in these enterprises is an implicit secret. In other words, AI tools are not fully available in such enterprises, but they are by no means scarce; most employees are reluctant to acknowledge them publicly, but do not deny them categorically.
As for the construction of AIAI agent and taking over some of its functions, it does not appear to be part of the recent plans of Chinese enterprises. When we asked, “How many jobs do you plan to replace with an AI tool in one year”, 66.7 per cent replied, “not planned for the time being”, and 22.5 per cent responded, “less than 10 per cent”, with a combined ratio of 89.2 per cent, an increase of 7.5 per cent from 81.7 per cent in 2025.
02 Manpower cuts - pretending to “AI leads to layoffs”
Let's see if the spread of AI technology has led to layoffs. Retrenchment, on the one hand, reduces labour costs and directly increases business profits, and, on the other hand, reduces the number of human employees, with the objective effect of mixed (rather than synergistic) machines. In public opinion, this phenomenon of “AI crowding out human employees” has occurred extensively, and even the media has written the story of “Human versus AI”, but is that true?
Gartner gave an extremely important correction: only 1 per cent of the layoffs in the first half of 2025 resulted from AI ' s increased employee productivity (Gartner, 2026). In other words, “AI leads to lay-offs” is far ahead of the fact that “AI actually produces sufficient productivity”.
Indeed, the perception of foreign firms after practice has begun to be rational about whether AI is sufficient to replace human beings. Data from Deloitte and Touche show that 39 per cent of business leaders had been assigned to AI, but 55 per cent considered this to be a mistake (Deloitte/Orgvue, 2025). Both Salesforce and Klarna promoted “AI-for-persons” in the area of customer service in an aggressive manner, but found that this had led to a decline in the quality of service, and that there was always an irreplaceable part of AI for human employees, which eventually led to the re-recruitment of human employees.
Although the Silicon Valley enterprise has made a large number of staff reductions in the name of substitution by AI, this has not occurred in large numbers in Chinese enterprises. This state of affairs is not based on post-practical reason; it is more likely that practice has not been developed, or that the prospect of “AI-change” is not good.
The study on the construction of Chinese business platform organizations in 2026, which was consulted by Mu Sheng, sheds more light on the reality of “AI leading to layoffs” with domestic data. Of these, only 2.0 per cent, or 17.6 per cent, of the enterprises with more than 10 per cent, are negligible. In other words, only a small number of enterprises have been retrenched under the name of AI, and the proportion is generally less than 10 per cent. Moreover, only a very small number of these enterprises (8.3 per cent) have been distilled in terms of employee skills. This suggests that the phenomenon of “distilled workers”, which is widely publicized in the media, is more like a fiery flow. Conversely, perhaps a large number of staff do not deserve to be distilled.
From the perspective of the staff, they do have a counter-AI stress, mainly in terms of “justification not using the AI tool” and “non-recognition of the results given by AI”, which account for 45.0 per cent. This negative response suggests that the confrontation between AZAI agent and human employees may not be as intense as the media described.
Functional progress — in which areas is human collaboration taking place?
We have collected global business cases and found positive applications in five areas: passenger service, research and development, back-office functions (HR, etc.), marketing, and supply chain.
We find that in the application of the five categories of functions, the maturity of human synergy does not depend on “how strong AI” but rather on the soundness of the “exceptional mechanism” of the scene, whether the “quality bottom line” is maintained and whether the “judgment criteria” are explained. They constitute three pillars of the long-term sustainability of the human resources team.
In 2025, a survey was carried out on the employers or CEOs (agents of the owners) of 68 large Chinese and medium-sized customer enterprises. In the questionnaire, we calculated the complexity of functions (including three dimensions of data complexity, decision-making independence and closedness) and AI application (including three dimensions of AI reference, AI implementation effectiveness and AI impact) for a number of common functional areas of an enterprise.
On this basis, we calculated an indicator called “Import AI Activeness”, which is “Import AI Activeity = AI Application Effect/Function Complexity”. This indicator is a clear illustration of the progress made by functional departments in importing AI under different “congenital conditions”, and the larger the indicator, the larger it is, the more the function is actively importing AI and, conversely, there is too much resistance to AI.
The results show that it is not entirely those functions that are easy to use AI to import it, but those that are under the greatest pressure and are easy to use AI to actively import it. To put it simply, from the point of view of profit avoidance, practitioners in all functional sectors want to be de-standardized, hoping that this function will have to rely on themselves, but that those that promote standardization and even introduce AI are the ones that are most under stress. On the one hand, the efficiency of the legal, quality standards, procurement, finance and other functions has long been conservative, and they also have conditions for the importation of AI, which seems to be understandable; on the other hand, the production/business/operation, finance and human resources functions are too difficult to work on the importation of AI; in addition, offices, strategies, research and development, audit inspections are, for a variety of reasons, guided by the hands-on operations of practitioners, and AI is naturally difficult to penetrate.
In conjunction with McKenzie, data on AZAI agent has not yet permeated in most enterprises and are not widely available in individual functions. We speculate that there is still a long way to go before human beings in the country form their team. Some pioneer enterprises in certain industries (e.g. technology) are willing to develop human synergies in some functions, but others are more like coping under stress than pursuing development initiatives. These enterprises are not motivated solely by the ease with which a function can be imported into AZAI agent, and their motivation is more dependent on external pressure.
In McKinsey ' s international Z111Z, the proportion of high-performing AI firms (only 6 per cent) deploying AI agent on a scale in most functions is more than three times that of other organizations, and these pioneers are bound to drive the trend. In our China sample, however, there are not too many convincing data from pioneer companies, even in well-known “show cases” like Salesforce and Klarna. The Chinese enterprise is about to refuel in the building of a team of human beings.
04 Improved performance — does it bring real money?
At present, the increase in the performance of AZAI agent after entering the enterprise is more evident at the cost end. If AZAI agent is able to replace part of the workforce while ensuring the quality of the work, it is true that the human and peripheral costs of “AI-change” are reduced.
Using AIAI agentAskHR, IBM undertook a large number of routine HR advice and services, saving approximately 8,000 people over the years and processing 94 per cent of staff counselling, resulting in a balance of $640 million per year. Indeed, this type of HR transactional function is the most accessible scenario for AI.
However, the cases of Salesforce and Klarna amply demonstrate that such cost reductions are not unlimited and may even have negative effects.
Salesforce took over the services and development work with AIAI agentAgentforce, whose support team was reduced from 9,000 to about 5,000, with no additional engineer posts, and Agentforce's ARR was increased from $800 million to $1.5 billion, an increase of 240 per cent. In April 2026, however, it recruited 1,000 graduates from sales positions. This is a strong indication that AI can take over the “boundary, structured” jobs, and that humans will return to the “situated, client-oriented, critical” jobs.
Klarna took over the passenger service with AIAI agent, resulting in a sharp drop in staff from 7,400 in 2022 to around 2831 in 2025, and the re-recruitment of manual passenger service after May 2025, which resulted in the operational cost of Q3 passenger service rising to $50 million, despite the increase in per capita income from 344,000 to $1.24 million. This amply demonstrates that AI is effective to some extent, but that “overdependence” is at the expense of “the quality of services”, and that quality and convergence remain human domain.
Of course, AI has also demonstrated its contribution to the income end, but still works in certain functional modules.
AIAI agent has begun to play a direct role in the areas of sale of ISDR (Sales Development Representative, sales development representative), AI collection, and AI client retention/additional sales.
“11x AI SDR” uses AI AI agent as a substitute or auxiliary manual SDR for automatic outside calls, e-mails, conferences, initial screens of clues. Over 200 meetings were booked in approximately 90 days, with approximately three times the number of meetings and a 70 per cent reduction in single meeting costs. In other words, while the chain fails to close the revenue chain, it is possible to increase revenue by increasing distribution channels and volume of meetings.
The Amazon classic recommended engine, the oldest and most direct commercial-value AI application, contributed 35 per cent of sales. At its core, algorithms have been used to replace the traditional “manual selection recommendations” or “static shelves” model. Its AI search column increased the visitor conversion rate from an average of 2.17 per cent to 12.29 per cent for search users, almost sixfold.
In addition to this, the Amazon ' s Generating AI Assistant Rufus was launched in 2024 and rolled out in 2025, from the evolution of “recommended goods” to “dialogue-led purchases”, directly replacing some of the manual passenger service and procurement functions. The Amazon CEO predicts that Rufus could bring about a $10 billion annual increase in sales. At present, 250 million active users have accumulated, representing a 140 per cent increase in monthly activity; users who interact with Rufus are 60 per cent more likely to complete their purchases than those who do not.
There is not much clear data on the impact on the performance of enterprises.
Research in McKinsey shows that only 6 per cent of the enterprises interviewed can be defined as “AI high performers”, i.e. AI's contribution to EBIT is more than 5 per cent and sees “significant value”. And their use of AI, in partnership with teams, is not an attempt at partial reduction, or local efficiency, but rather a focus on growth and innovation.
This view was further confirmed by China ' s sample, which was consulted by Mu Sheng. In 2025, a study of 68 large Chinese and medium-sized customer business owners or CEOs for sample showed that only 7.4 per cent of businesses were aware that AZZ AI agent “comprehensively improved organizational efficiency and translated into business value” and that they had more than three and four functions in “AI application” and “Importing AI motivation” respectively, well above the average of sample (high score of 4 points and above). Other sample have only one or two high-scoring functions on these two indicators, and the gap is quite clear, in other words, they have a holistic approach at the outset, not only for the partial reduction or efficiency gains.
This may be a “motive difference” between AI, the organization, and the human team. In our in-depth interviews, we found that the Chinese boss's ideas of initiating “AIization” or “human teamwork” were widespread, and that “AI must be a trend, and companies are profitable and can invest in the future, no matter what it can do, use it.” There are several areas within our company that are severely affected by internal consumption, rather than internal consumption, let AI replace them." There must be a lot of people in our company right now, cutting people off with this sword, cutting costs and adding to the atmosphere of survival in the cold winter.” Most Chinese bosses use AI either in vain or in short supply.
We believe that without organizational upgrading, rather than redesigning workflow, and simply marrying AI into the original organization and process, the business will not be able to achieve the growth in its business performance. As practice deepens, a growing number of enterprises or research institutions realize that the term “automated organizational efficiency after AI has changed” is not valid at all, that AI can be downgraded, that there may be efficiency gains in local areas, but that a large number of constraints appear to have to be considered in order to transform local efficiency gains into overall efficiency gains and to produce final results.
Microsoft ' s 2026 study found that organizational conditions (culture, manager support, talent practice) affect AI performance approximately twice as much as individual factors (67 per cent vs 32 per cent); and that only 19 per cent of enterprises enter the "front zone ", a state where individual competencies and organizational support reinforce each other (MS, 2026).
However, the description of the conditions of the organization, Microsoft, may be too broad, and the key is its design and peripheral system. According to a study on the construction of a platform organization for Chinese enterprises, which was consulted by Mu Sheng, only 26.7 per cent of enterprises have undergone organizational restructuring to promote the transformation of AI. For the rest of the enterprises, technology is technology and organization is organization. For those enterprises that have undergone organizational adjustments, they are clearly ahead of schedule. Of these, 56.3 per cent of the business-driven organizational transformations are clearly platform-based. That is, most of them move to platform-based organizations once they realize that the import of AI technology requires organizational change or adjustment.
1 In B2B sales, SDR is usually responsible for the front-end of the sales process: looking for potential customers, cleaning clues; cold-starting exterior calls, e-mailing, LinkedIn touch; pre-judgemental intent, booking meetings/transfer of qualified leads to sales (AE); and generally not responsible for the eventual signing of a signature. In contrast, AE (Account Export, Customer Manager/Sales Representative) is primarily responsible for following up on business opportunities, negotiations and deals.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.