Artificial Intelligence promises to reshape industries, yet most companies are still struggling to see results. Despite record investments, nearly all AI projects stall before reaching real impact. Why do so many fail and what separates the few success stories from the rest? This article explores the “GenAI Divide” and shares strategies to help organisations cross it. Here let’s summarise it 15 points out from this super 麻省理工学院 NANDA 报告.
1.导言：人工智能的前景与问题
人工智能 (人工智能（AI）被誉为 21 世纪最具变革性的技术。随着机器学习、自然语言处理以及最近的生成式人工智能的兴起，企业纷纷抢着采用这些工具；然而，公司的方法可能是成功与无法实现价值的分水岭。然而，尽管在人工智能研究、基础设施和试点方面投入了数十亿美元，但大多数企业都未能看到可衡量的回报。.
最近的一项现实调查显示，尽管人工智能项目被广泛炒作和采用，但仍有 95% 的组织报告称，这些项目几乎没有产生任何价值。造成这种分歧的原因并不是技术本身缺乏创新，而是技术的应用、整合和管理方式。.
本文探讨了人工智能项目失败的原因、“GenAI 鸿沟 ”对企业意味着什么，以及哪些战略可以帮助企业释放人工智能的真正潜力。.
2.人工智能的应用规模
诸如 ChatGPT、Midjourney 或 Copilot 等生成式人工智能工具已经家喻户晓。全球数百万员工每天都在试用它们。银行、医疗保健和零售等行业的采用率很高。然而，采用并不等于转型。.
While pilots are easy to launch, turning them into production-ready, value-generating systems is far harder. Many organisations get stuck in pilot purgatory running multiple AI experiments without ever scaling them into business-critical processes.
3.解释 GenAI 的分歧
The “GenAI Divide” refers to the gap between AI adoption and AI transformation. On one side are organisations that treat AI as a shiny experiment, running disconnected pilots that fail to influence core workflows. On the other are the few roughly 5% who successfully integrate adaptive, learning-capable systems that transform operations.
这种鸿沟与获取技术无关。如今，每个组织都可以使用功能强大的模型。真正的区别在于方法和整合。.
4.人工智能项目失败的常见原因
为什么大多数人工智能项目都会失败？这其中有几个反复出现的主题：
- 缺乏明确的目标：许多项目在开始时都没有明确的业务目标。.
- 不切实际的期望：企业高估了人工智能的短期潜力。.
- 数据质量差：数据质量差会导致人工智能模型产生有偏差或不正确的结果，当模型在有偏差、不完整或不相关的数据集上进行训练时，就会导致项目失败。.
- 集成差距：试点项目孤立运行，无法扩展到实际系统中。.
- 文化阻力：员工往往缺乏培训或不信任人工智能产出。.
麻省理工学院和麦肯锡的研究表明，多达 80% 的人工智能试点项目从未投入生产，这凸显了执行力而非雄心壮志才是主要瓶颈。.
5.数据的作用：垃圾进，垃圾出
人工智能的好坏取决于它所消耗的数据。高质量、管理良好的数据是成功的关键，但许多组织却低估了这一要求。标注不清的数据集、缺失值以及训练样本缺乏多样性往往会削弱人工智能计划。在现实世界的部署中，不良的数据实践是人工智能失败的主要原因。.
Strong data management practices covering collection, governance, cleansing, and labelling are not optional extras. Without them, AI projects collapse under the weight of bad inputs.
6.不具规模的试点
人工智能试点之所以诱人，是因为它们启动迅速，易于展示。但是，没有扩展战略的试点是注定要失败的。许多高管都在庆祝概念验证演示，但这些演示却从未过渡到企业工作流程中。.
关键问题应该是 “这一试点将如何融入我们的日常运营、系统和关键绩效指标？” 如果答案不明确，项目就已经走向失败。有效的项目管理对于确保试点项目成功扩展到生产系统至关重要。.
7.错位的使用案例
人工智能计划往往是追逐炒作，而不是解决紧迫问题。例如，50% 的人工智能生成预算被分配给了销售和市场营销。 市场营销 项目，主要是因为它们能产生可见的产出。然而，研究表明，后台自动化往往能带来更好的投资回报率。.
Successful projects start with real pain points processes where automation, prediction, or insight can dramatically improve efficiency or customer experience. Identifying the actual use case guides the selection of the most effective solution, ensuring that the chosen approach truly addresses the underlying business problem.
8.人类与人工智能的合作：不是替代，而是合作
与人们普遍担心的相反，人工智能并不是要完全取代人类。相反，最成功的项目设计的是 "人在回路中 "的系统，人工智能可以增强而不是取代人类决策。.
例如，人工智能可以分流客户查询，将简单的问题标记为自动化问题，将复杂的问题升级为人工代理问题。这种混合模式可以建立信任、降低风险，并取得比人工智能或人类单打独斗更好的结果。建立一支技术熟练的团队来管理和监督人类与人工智能的合作，对于确保这些系统有效运行并取得最佳结果至关重要。.
9.人工智能影子经济
One striking trend is the rise of shadow AI employees using generative tools unofficially to boost productivity. Whether writing reports, summarising meetings, or automating spreadsheets, these personal AI hacks often deliver better ROI than formal initiatives. Often, it is the choice of the right tool for the task that drives these unofficial successes.
具有前瞻性思维的组织不会忽视或惩罚影子人工智能，而是对其进行研究和学习。非官方使用的模式可以为官方战略提供参考，帮助领导者了解人工智能真正的增值点。.
10.人工智能系统适应性的重要性
Generic, static models quickly reach their limits. Learning-capable systems that adapt to feedback and context are the future. Without adaptability, AI becomes brittle useful in a demo, but useless in complex, changing workflows.
Startups crossing the GenAI Divide tend to build narrow but highly adaptive systems. They prioritise domain fluency deep knowledge of a specific industry or process over broad general-purpose capability. These adaptive systems are treated as living products: dynamic, operational entities that are continuously monitored, versioned, and improved through real-time feedback and human oversight, ensuring ongoing business impact and seamless integration into enterprise workflows.
11.了解人工智能模型和解决方案
The critical factor that separates your successful AI initiatives from total failures? Deep, practical understanding of AI models and solutions. In your rush to adopt artificial intelligence, you’re overlooking the complexities that drive effective AI projects. This oversight is your leading cause of AI project failure you’re underestimating the importance of high quality data, robust training data, and the nuances of machine learning models.
In today’s business world, your AI pilots fail to deliver measurable return. This “GenAI Divide” isn’t just about your access to the latest AI tools or recent software updates it’s about whether you truly grasp how AI systems work, what their limitations are, and how to align them with your real business needs. Your inflated expectations, driven by hype, lead you to invest in AI features that look impressive in demos but fall short in production, especially when you ignore edge cases and integration challenges.
数据科学和数据科学家的专业知识是每个人工智能项目取得成功的核心。这些专业人员可确保您的人工智能模型在优质数据的基础上进行训练，经过严格测试，并能保留反馈信息和适应新情况。如果没有这个基础，即使是最先进的人工智能技术也会产生不可靠的结果，导致可衡量的零回报和投资浪费。.
The MIT study and resources like the AI incident database highlight your recurring theme: your AI projects fail most often due to poor understanding of underlying models, insufficient testing, and lack of focus on solving real problems. For your mid market firms and large enterprises alike, the lesson is clear your success depends on more than just deploying AI tools. You need commitment to understanding how these tools function, how they integrate with your existing systems, and how you can adapt them to deliver real value.
将这种理解放在首位的企业能够更好地驾驭人工智能计划的复杂性。你们认识到解决集成难题、规划边缘案例以及确保人工智能模型随着业务需求变化而发展的重要性。这种方法不仅能降低人工智能项目失败的风险，还能最大限度地提高投资回报，将人工智能从成本中心转变为业务增长的真正驱动力。.
In a landscape where you’re investing millions in AI initiatives, and where the line between your success and failure is razor-thin, your ability to understand and control AI models and solutions is paramount. Your teams and leaders who focus on this understanding rather than simply relying on hype or the latest technology are far more likely to deliver projects that succeed at scale, provide measurable return, and solve your real business problems.
最后，从过去的错误中吸取教训至关重要。人工智能事件数据库为您提供了宝贵的见解，让您了解人工智能项目失败的原因，从而加强您对严格研究、专注和持续教育的需求。通过将理解作为您发起的每项人工智能计划的基石，您可以弥合 GenAI 鸿沟，并确保您在人工智能方面的投资能够带来持久、变革性的价值。.
11.成功建设者的经验教训
如今蓬勃发展的人工智能公司都遵循一个共同的模式：
- 他们建立的适应性系统会随着时间的推移而不断改进。.
- 它们专注于特定的高价值用例，而不是庞大的功能集。.
- 他们优先考虑工作流程整合，将人工智能嵌入日常业务流程。.
这与那些只制作华而不实的演示而不将其嵌入员工实际使用的工具中的公司形成了鲜明对比。.
12.成功买家的经验教训
在买方方面，最有效的组织对待人工智能采购更像是业务流程外包（BPO），而不是传统的软件即服务（SaaS）。他们要求
- 根据他们的工作流程量身定制。.
- 基于成果，而不仅仅是功能。.
- 与供应商合作，共同开发解决方案。.
这种思维方式将人工智能从 “你安装的产品 ”转变为你发展的合作伙伴关系。.
13.下一个前沿：代理网络
Looking ahead, AI is moving towards an agentic web a network of autonomous systems that communicate and coordinate tasks without constant human intervention. These changes are already happening in some industries, where autonomous systems are being integrated into workflows and transforming how work is organized. Emerging protocols such as MCP (Model Context Protocol) and A2A (Agent-to-Agent) are paving the way.
在未来，系统将不仅仅是生成文本或图像，它们还将记忆、计划和行动，在最小的监督下适应各种工作流程。现在就为这一转变做好准备的公司将最有可能获得未来的价值。.
14.跨越 GenAI 鸿沟的战略
组织如何才能弥合试点采用与有意义的转型之间的差距？主要策略包括
- 确定明确的目标：将每项人工智能计划与可衡量的业务成果挂钩。.
- 投资数据：优先考虑治理、多样性和相关性。.
- Focus on ROI-rich use cases: Don’t just follow the hype automate where it matters.
- 支持人类与人工智能的合作：让人们参与监督和信任。.
- 从影子人工智能中学习：研究非官方采用模式，为正式战略提供指导。.
- 战略合作：将人工智能供应商视为合作者，而不仅仅是供应商。.
- 选择适应性强的系统：优先选择具有学习能力的工具，这些工具会随着使用而不断发展。.
如果没有这些战略，企业的人工智能投资将面临零回报的风险。.
15.结论：从失败到转变
The story of AI today is one of potential versus practice. While billions are invested, only a small fraction of projects deliver meaningful returns. The GenAI Divide illustrates that technology alone is not the problem it is approach, integration, and execution.
通过从失败中吸取教训、接受适应性和优先考虑整合，企业可以将人工智能从成本中心转变为增长动力。未来不在于试点，而在于能够学习、协作并改变工作方式的系统。.
