跟随百分之五

跟随百分之五
Follow the Five Percent

Peter Cooper - AgileMesh - 2026年5月

摘要 / Abstract

95%的企业AI试点项目在投产前失败。成功的5%是上市公司,提交经审计的年报,记录其架构,年增长30%。本文研究了这5%究竟做了什么。答案在每个案例中都是相同的:他们保持了数据中的连接。

95% of enterprise AI pilots fail before reaching production. The 5% who succeed are public companies, filing audited results, documenting their architectures, growing 30% year on year. This paper examines what that 5% actually did. The answer is the same in every case: they kept the connections in their data alive.

2秒
Lemonade理赔处理
Lemonade claim processing
6.1x
AI领先者股东回报倍数
TSR multiple for AI leaders
320%
知识图谱3年投资回报率
Knowledge graph 3-year ROI
95%
AI试点失败率
AI pilot failure rate

1. 赢家做了什么 / What the Winners Did

两秒钟的旅程
图三 - 两秒钟的旅程 / Plate III - The Two-Second Walk

Lemonade: AI Jim在2秒内处理理赔。客户到保单到理赔到欺诈信号到付款 - 一条连续的链。典型的英国保险公司处理同样的理赔需要25-30天。

Root Insurance: 实时远程信息处理定价。驾驶员连接到驾驶行为。2025年收入15.2亿美元,增长29%。

Allianz: 暴风雨理赔在5分钟内分析完成。处理时间减少80%。

AXA Switzerland: 60%自动化率,大于99%准确率。

Lemonade: AI Jim processes claims in 2 seconds. Customer to policy to claim to fraud signal to payout - one continuous chain. A typical UK insurer takes 25-30 days for the same claim.

Root Insurance: Real-time telematics pricing. Driver connected to driving. $1.52B revenue in 2025, up 29%.

Allianz: Storm claims analysed in under 5 minutes. Processing time cut 80%.

AXA Switzerland: 60% automation rate, >99% accuracy.

2. 为什么95%看不到 / Why the 95% Cannot See

数据的两种条件
图二 - 数据的两种条件 / Plate II - On the Two Conditions of Data

SQL单一文化。大学课程中不到六分之一提供图数据库模块。毕业生构建他们知道的唯一形状。采购框架要求关系架构。DBA工会保护其领域专业知识。工具链只消费表格数据。

当AI在表格数据上遇到复杂的业务问题时,准确率仅为16%。当基于语义知识图谱时,准确率提高到54% - 提高三倍。

The SQL monoculture. Fewer than one in six university programmes offer a graph database module. Graduates build the only shape they know. Procurement demands relational architectures. DBA guilds protect their domain. Tooling only consumes tabular data.

AI accuracy on complex business questions over tabular data: 16%. Grounded in a semantic knowledge graph: 54% - a threefold improvement.

3. 谁实际上在付钱 / Who Actually Pays

可见的与沉没的
图一 - 可见的与沉没的 / Plate I - The Visible and the Submerged

保单持有人。患者。不是董事会,不是CIO,不是供应商。

NHS向Palantir授予了3.3亿英镑的合同。七年期。订阅模式。合同结束时没有软件转让给NHS。总成本预计超过10亿英镑。到2025年中,不到三分之一的信托机构在使用。

10亿英镑。无知识产权转让。三分之二拒绝采用。

The policyholder. The patient. Not the board, not the CIO, not the vendor.

The NHS awarded Palantir a GBP 330 million contract. Seven-year term. Subscription model. No software transfers when the contract ends. Total cost projected to exceed GBP 1 billion. By mid-2025, fewer than one in three trusts were using it.

GBP 1 billion. No IP transfer. Two-thirds declining to adopt.

4. 如何开始 / How to Start

掩体上的叠加
图四 - 掩体上的叠加 / Plate IV - The Overlay Upon the Bunker

语义叠加。不要替换现有系统。在它们之间画一张地图。试点成本:20-40万英镑。中型企业推广:50-200万英镑。培训:5-20万英镑。投资回报期:6-12个月。

本体论不是自上而下传播的。它们像病毒一样传播。Samuel Slater将Arkwright的纺织厂蓝图记在脑中带到大西洋彼岸。一个人理解了机器的形状,工业革命就在新大陆上病毒式传播。

Semantic overlay. Don't replace existing systems. Draw a map between them. Pilot cost: GBP 200-400k. Mid-enterprise rollout: GBP 500k-2M. Training: GBP 50-200k. Payback: 6-12 months.

Ontologies do not spread top-down. They spread like a virus. Samuel Slater carried Arkwright's textile mill blueprints across the Atlantic in his head. One person understood the shape and the Industrial Revolution went viral.

5. 百分之五宣言 / The Five Percent Manifesto

3.3亿英镑创可贴的时代结束了。

5%通过连接数据、赋能员工和拥有自己的智能来成功。

事实证明,人工智能最大的瓶颈可能正是我们自身组织上的缺乏。

跟随百分之五。

The era of the GBP 330 million band-aid is over.

The 5% succeed by connecting their data, empowering their staff, and owning their intelligence.

It turns out the biggest bottleneck to artificial intelligence might just be our own organisational lack of it.

Follow the Five Percent.