AI News AI资讯 11h ago Updated 1h ago 更新于 1小时前 52

Bain study finds companies miss AI savings targets because humans keep getting in the way 贝恩研究显示公司因人为因素未达成AI成本节约目标

The great AI productivity miracle is stalling, and the first body to be dumped in the harbor is the myth of the fully autonomous agent. According to a new Bain survey of nearly a thousand companies, almost 40 percent are achieving less than 10 percent in cost savings from their AI initiatives, a dismal performance when most had targeted 11 to 20 percent. The culprit, according to the report’s narrative, is that only 7 percent of companies actually run fully autonomous AI agents, despite building 贝恩最新调查像一盆冷水,直接泼醒了那些沉浸在AI降本增效美梦中的企业高管们。调查结果冷冰冰:近四成公司实际实现的AI成本节约不到10%,而他们最初的目标是11%到20%。这种集体“碰壁”并不意外,但官方给出的理由却滑稽得令人喷饭——“因为只有7%的公司真正部署了完全自主的AI代理,而人类总在中间碍事”。

75
Hot 热度
80
Quality 质量
70
Impact 影响力

Analysis 深度分析

The great AI productivity miracle is stalling, and the first body to be dumped in the harbor is the myth of the fully autonomous agent. According to a new Bain survey of nearly a thousand companies, almost 40 percent are achieving less than 10 percent in cost savings from their AI initiatives, a dismal performance when most had targeted 11 to 20 percent. The culprit, according to the report’s narrative, is that only 7 percent of companies actually run fully autonomous AI agents, despite building their business cases on that very premise. The headline spins this as humans "getting in the way," but that framing is dangerously backwards and reveals a fundamental misunderstanding of where the failure truly lies.

The real problem isn’t that humans are in the loop; it’s that the loop was designed by people selling a fantasy. For two years, the enterprise tech ecosystem has been peddling a singular vision: the magic button. An AI agent that will ingest your data, manage your workflows, and deliver 20% efficiency gains with minimal oversight. Sales decks and ROI projections are built on this "level 5 autonomy" promise, the corporate equivalent of a self-driving car that never needs a driver. The Bain data is the first major reality check, and it’s brutal. The 7 percent who claim to have achieved it are likely running narrow, highly constrained use cases—automating a specific code test or a rigid customer service script—not the transformative, cross-enterprise orchestration that was sold to the board.

What we’re witnessing is the trough of disillusionment, but with a twist. The disillusionment isn’t with the technology’s potential, but with the lazy, implementation-averse corporate strategy surrounding it. Blaming humans for "getting in the way" is a cheap shot, a vendor’s excuse to avoid accountability. What humans are doing is their jobs: they are applying judgment, context, and responsibility. They’re the ones who have to deal with the messy fallout when an autonomous agent makes a confident, high-stakes error. They’re the ones who understand that "autonomy" in a complex business environment often means "liability" and "unpredictable risk." The 93 percent of companies that haven’t deployed full agents aren’t failing to get out of the way; they’re wisely refusing to drive a car off a cliff just because the brochure said it could fly.

This exposes a deeper, more damning pattern in tech adoption cycles: the conflation of tool deployment with strategic integration. Companies threw money at AI platforms, assuming the software itself would generate the savings. But AI is an amplifier, not a magic wand. It amplifies your processes, your data quality, and, most critically, your strategy. If your process is broken, AI will automate the brokenness at incredible speed. If your data is siloed and poor, your agent will be a very fast, very confident idiot. The savings don’t come from the agent; they come from the painstaking, unglamorous work of re-engineering workflows, cleaning data, and redefining roles. That work requires humans at the center, not humans as an embarrassing obstacle to be minimized.

The Bain study subtly reveals where the real value is being found, and it’s in the 75 percent of companies achieving moderate savings. This suggests the winning model isn’t full autonomy but augmentation. The most effective AI deployments are those that make a human expert 2x faster or more accurate, not those that replace them entirely. It’s the coding assistant that handles the boilerplate so the developer can focus on architecture; the data analysis tool that surfaces anomalies so the strategist can make a decision. This is less sexy than the "autonomous future," but it’s where tangible ROI lives today. The hype machine, however, isn’t built to sell a powerful assistant; it’s built to sell a revolution, and revolutions require the obsolescence of old ways, including the human element.

So, the real lesson here isn’t that we need to ruthlessly remove humans from the loop. It’s that we need to ruthlessly remove the fantasy of effortless transformation from the boardroom. The next phase of AI adoption will be defined by a sober return to first principles. It will require asking: What is the specific, measurable human task we are augmenting? What data foundation must we build for this to work? How do we redesign the surrounding process to capture the value? The companies that succeed will be those who treat AI as a tool requiring expert integration, not a plug-and-play solution. They’ll stop measuring success by the absence of humans and start measuring it by the enhanced capability of their human teams. The 40 percent coming up short aren’t a victim of human interference. They’re a victim of a flawed premise, and until that premise is corrected, those cost savings will remain just another line item in a presentation, waiting for an autonomous agent that never arrives.

贝恩最新调查像一盆冷水,直接泼醒了那些沉浸在AI降本增效美梦中的企业高管们。调查结果冷冰冰:近四成公司实际实现的AI成本节约不到10%,而他们最初的目标是11%到20%。这种集体“碰壁”并不意外,但官方给出的理由却滑稽得令人喷饭——“因为只有7%的公司真正部署了完全自主的AI代理,而人类总在中间碍事”。

这真是本年度最佳甩锅。把责任推给“人类参与”,就像厨师把菜做糊了怪食材太有个性一样荒诞。问题核心根本不在于“人类是否在环”,而在于这些公司从一开始就制定了脱离实际的幻想式蓝图。他们画出了一幅全自动、零干预的AI乌托邦,以为只要代码写好,利润就会像瀑布一样倾泻而下。现实是,绝大多数业务场景远比“输入-输出”的二元模型复杂,充满了需要人类判断的灰色地带、例外情况和情感交互。AI擅长的是模式识别和优化已知路径,而人类的优势在于应对模糊性、处理伦理困境、进行创造性破局。指望用AI全盘替代人类的决策环节,尤其是在降本这种牵涉流程重组、组织文化甚至岗位调整的敏感领域,本身就是一种战略上的懒惰和天真。

更讽刺的是,这份调查无意中揭开了当下AI落地中最虚伪的伤疤:商业案例的严重注水。这些公司在立项、融资或向上汇报时,几乎无一例外地将AI描绘成点石成金的魔法石,基于“全自主运行”的假定,堆砌出诱人的ROI数字。然而,当进入实施阶段,他们才发现所谓的“AI项目”往往只是一个需要持续人工标注、微调、监督和兜底的“高级自动化工具”。人类不仅没有缺席,反而成了维持系统运转不可或缺的“润滑油”和“消防员”。这哪里是“人类障碍”?这分明是AI当前能力边界的真实写照,是技术从PPT走向工厂车间时必然经历的“祛魅”过程。

所以,真正的问题不是人类“挡路”,而是企业决策者们集体患上了“技术冒进症”。他们被供应商的营销话术、同行的“军备竞赛”和资本市场对AI故事的狂热所裹挟,跳过了最关键的基础诊断:我们的业务流程是否真的适合用AI重构?我们是否有足够的数据治理和工程化能力去支撑一个“自主”代理?我们是否为可能出现的模型错误、偏见和责任界定预留了充分的人力监督与应急预案?直接跳到“用AI省人省钱”这个终点,而忽略了中间漫长而艰苦的适配、集成和能力建设过程,无异于没打地基就直接盖摩天楼,楼歪了只能怪风太硬。

因此,这份调查的最大价值,不在于揭露了一个数字,而在于撕下了一层遮羞布。它告诉市场:所谓“AI驱动”的降本增效,目前还远未到收获期,大量公司正处在投入期与幻想期的痛苦摩擦之中。那些将“人类干预”视为成本而非资产的公司,恐怕会越陷越深。真正的赢家,将是那些认清AI是“增强智能”而非“替代智能”,并致力于设计“人机协同”新流程的企业。他们不幻想一步登天,而是从具体痛点出发,将AI嵌入到人类工作流中,用它来放大人的能力,而非急于勾掉人的名字。这条路更慢、更笨,但也更真实。毕竟,在商业世界里,能真正解决复杂问题的技术才是好技术,而不是那个听起来最酷、最吓人的概念。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Agent Agent 评测 评测 部署 部署
Share: 分享到: