Boards Reject AI Spending Over Execution Bottlenecks, Fricke Says

There Is No ROI of AI: What Boards Really Ask Before They Fund It With Various Companies (Australia) — Ai4

Ai4

Corporate AI investments fail not from model deficiencies but because enterprises lack the operational capacity to absorb workflow redesigns, exposing directors to asset write-downs and regulatory liabilities.

AI-assisted summary of There Is No ROI of AI: What Boards Really Ask Before They Fund It With Various Companies (Australia) — original source. Check the original for full context.

Source published: 2026-08-19

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 19, 2026, this keynote features Leah Fricke, an Australian board director and audit committee chair. According to the video description, Fricke pulls back the curtain on boardroom deliberations to explain why traditional ROI metrics fail and shares the four key questions and strategic return categories directors actually use to evaluate AI proposals.

## TL;DR Leah Fricke argues that the traditional financial 'ROI of AI' is the wrong question for enterprises because value depends entirely on an organization's change capacity and strategic context rather than algorithmic capability. True enterprise moats reside in institutional judgment and proprietary context rather than rented third-party foundation models. She outlines four strategic, non-financial returns—decision quality, resilience, optionality, and alpha—and categorizes companies into archetypes like Preservers and Disruptors that require fundamentally different technology strategies. Ultimately, Fricke emphasizes that AI initiatives fail during execution rather than the pitch, underscoring that human leaders retain non-negotiable fiduciary accountability because AI cannot go to jail.

## Atmosphere & style The presentation is sharp, pragmatic, and refreshingly candid, delivering an unvarnished insider perspective from the highest levels of corporate governance. Fricke combines wry humor and battle-tested dot-com war stories with sober financial realism, puncturing Silicon Valley hype with discussions of regulatory risk and balance-sheet impairments. The tone is authoritative and conversational, leaving the audience with an engaging, no-nonsense masterclass in boardroom dynamics and strategic discernment.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=4s) The reported paradox between high enterprise AI spend and lack of measurable ROI stems from asking the wrong question rather than faulty measurement, as identical software tools yield vastly different value across different organizations. - [02:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=124s) The primary barrier to enterprise AI success is not model capability, but the widening choke point between vertical technology acceleration and the slower pace of institutional change absorption. - [02:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=124s) Boards frequently reject technically sound AI pitches if the company's operational 'change load' is too high to digest further organizational upheaval. - [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) Frontier AI models do not constitute a defensible moat; building solely on external foundational models leaves organizations vulnerable to vendor pricing power while real defensibility lies in proprietary data, judgment, and regulatory navigation. - [08:13](https://www.youtube.com/watch?v=cC8CR7B18A0&t=493s) Beyond financial ROI, boards evaluate AI initiatives across four non-financial strategic returns: decision quality, resilience, strategic optionality, and competitive alpha. - [11:17](https://www.youtube.com/watch?v=cC8CR7B18A0&t=677s) AI strategy must match an organization's specific archetype—Preserver, Scaler, Reinventor, or Disruptor—because identical technological bets that suit an aggressive disruptor can be catastrophic for a heavily regulated preserver. - [14:33](https://www.youtube.com/watch?v=cC8CR7B18A0&t=873s) Every board evaluates AI funding through four essential questions: the exact problem being solved, the category of return relative to company archetype, the implementation pathway, and non-productivity metrics paired with explicit kill criteria. - [14:33](https://www.youtube.com/watch?v=cC8CR7B18A0&t=873s) Conservative enterprises often deliberately wait for established, lower-risk enterprise vendors to ship 'good enough' solutions rather than adopting cutting-edge startups with governance and lock-in risks. - [17:06](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1026s) AI initiatives rarely fail during the pitch; they die during execution due to systemic failures to redesign workflows, leading to quiet project retreats and unexpected balance-sheet write-downs. - [17:06](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1026s) Fiduciary and legal accountability cannot be delegated to algorithms, leaving board directors as the indispensable 'biological sacrifice' responsible for governance and regulatory compliance. - [17:06](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1026s) The definitive competitive advantage for modern executive leadership is 'AI discernment'—the disciplined judgment to select appropriate returns, identify when human oversight is non-negotiable, and possess the discipline to wait.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=4s) The False Dilemma of AI ROI and the Real Enterprise Gap

Addressing an audience steeped in discussions about the return on investment for artificial intelligence, Leah Fricke offers a contrarian perspective from the governance side of the boardroom table. When enterprise teams pitch AI projects using language and financial metrics that directors neither trust nor prioritize, initiatives stall out before they even get off the ground. Over the past few years, tens of billions of dollars have poured into enterprise AI initiatives, yet the data presents a baffling paradox: major research reports indicate that 95% of enterprises have seen no measurable return, while other executive surveys claim three-quarters of organizations believe their deployments are performing well.

Most industry commentators treat this massive discrepancy as an information or reporting problem. Fricke argues that the underlying premise is entirely mistaken: "the question is wrong." The gap does not stem from faulty measurement, but from the reality that different organizations cannot extract the same value from identical software tools. Nuances governing whether an implementation succeeds run much deeper than any headline metric from a CEO survey. Having spent the last decade sitting on boards of directors, Fricke works to help executives and board members distinguish between technology investments that generate true, durable enterprise value and those that merely offer minor, fleeting productivity uplifts.

- Tens of billions of dollars have been invested globally into enterprise AI projects over recent years. - Contradictory studies highlight the disconnect: 95% of enterprises report zero measurable financial return, while 75% claim their initiatives are thriving. - Technology decisions in corporate boardrooms are never made through the theoretical purity of a single ROI equation.

---

[02:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=124s) The Institutional Choke Point: Change Load and the Cost of Waiting

In boardroom deliberations, technology itself is almost never the real obstacle. Organizations repeatedly run into trouble because introducing advanced software does nothing to resolve the systemic friction points that plague large transformation projects. The actual breakdown occurs in the widening gulf between what a model can technically do and what an organization is capable of putting to work, digesting, and governing. While artificial intelligence capabilities accelerate upward at an almost vertical trajectory, human institutions operate at the comparatively sluggish speed of human coordination. Whenever that gap expands, choke points emerge, value leaks out, and projects falter.

A vital, underappreciated metric for directors is an enterprise's current "change load." Even when a team presents an immaculate business plan paired with a cutting-edge tool and a mathematically stellar projected ROI, a board will routinely turn it down if the organization lacks the cognitive and operational bandwidth to absorb further upheaval. For directors, the central inquiry is not "what is the return of AI," but rather "what return actually matters for us at this point in our strategy and what is the cost of waiting?" Boards frequently stumble in both directions: they either commit capital to the wrong category of return, or they paralyze themselves waiting for absolute certainty that will never arrive. As Fricke asserts: "technology creates capability and leadership and institutions determine if that capability is ever converted into any kind of value."

```mermaid flowchart TD accTitle: Technology Capability vs Institutional Absorption Gap subgraph TechCapability["Technology Trajectory"] A["Frontier AI Capability"] -->|"Accelerates Vertically"| B["Raw Algorithmic Potential"] end subgraph HumanOrg["Enterprise Reality"] C["Institutional Absorption"] -->|"Moves at Human Speed"| D["Change Load & Governance"] end B --> E{"The Value Choke Point"} D --> E E -->|"Unabsorbed Value Leaks"| F["Project Stalls & Failures"] E -->|"Aligned Governance & Leadership"| G["Enterprise Value Realized"] ```

---

[05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) Lessons from the Dot-Com Era: Why the Model Is Not Your Moat

Reflecting on her tenure as an executive during the dot-com boom—noting wryly that if younger audience members cannot recall the era, "ChatGPT can tell you all about it"—Fricke recounts her time as employee number 13 at Australia's first successful fintech to complete an initial public offering. Despite possessing absolute technological superiority in the marketplace, the company was abruptly blindsided by an existential regulatory threat surrounding anti-money laundering (AML) mandates. Enforcing the drafted regulations as originally written would have shuttered the business immediately, rendering it "dead on arrival."

The fintech survived not because of its codebase or algorithmic prowess, but because leadership engaged in intensive government relationship management, lobbying, and educational dialogues. They demonstrated the product's strategic value and successfully worked with Australia's export regulatory authorities to help them comprehend—for the first time in the nation's history—that an enterprise could export a service. Fricke uses this historical parallel to deliver a stark warning to modern organizations: "The model is not your moat." The frontier crown shifts hands continuously, open and cheaper models follow immediately behind, and developments in China threaten to decimate business models currently celebrated in Silicon Valley. Building an entire value proposition atop an external foundational model amounts to renting a moat from a vendor—one who will inevitably leverage their pricing power to capture the economic surplus while leaving the enterprise with whatever "scraps you are allowed to keep." True moats reside exclusively in institutional judgment, proprietary data, and contextual understanding that cannot be captured in a mathematical ROI spreadsheet.

---

[08:13](https://www.youtube.com/watch?v=cC8CR7B18A0&t=493s) The Four Forms of Non-Financial Return

When boards evaluate technology proposals, financial profit alone is rarely the determining variable. Fricke outlines four strategic forms of return that enterprise leaders should target instead:

```mermaid flowchart LR accTitle: Four Strategic Forms of Return for Board-Level AI subgraph StrategicReturns["Non-Financial Enterprise Returns"] R1["Decision Quality"] R2["Resilience"] R3["Optionality"] R4["Competitive Advantage: Alpha"] end R1 -->|"Beats rivals & reassures regulators"| O["Enterprise Survival & Leadership"] R2 -->|"Withstands market shocks"| O R3 -->|"Move cheaply at own pace"| O R4 -->|"Uncopyable core capability"| O ```

* **Decision Quality:** Enhancing the speed, accuracy, and depth of high-stakes corporate decisions serves as a direct lever to outmaneuver competitors, satisfy demanding regulators, impress clients, and defend the organization's fundamental license to operate in competitive markets. * **Resilience:** The capacity of an enterprise to absorb structural stress and market shocks without fracturing. In volatile environments, possessing built-in operational shock absorbers distinguishes lasting survivors from firms that collapse under strain. * **Optionality:** Preserving the strategic freedom to maneuver on the organization's own terms and timetable. This allows leadership to act as a first mover when conditions warrant or advance incrementally and inexpensively while staying safely within the firm's defined risk envelope. * **Competitive Advantage ("The Alpha"):** Fricke draws a sharp line between standard technology adoption—baseline table stakes required merely to keep operational costs on par with the market—and true competitive advantage. Parity simply keeps a firm in the game; the real return comes from establishing an proprietary "alpha" that competitors cannot easily duplicate.

---

[11:17](https://www.youtube.com/watch?v=cC8CR7B18A0&t=677s) Enterprise Archetypes: Why the Identical Bet Can Be Fatal

There is no universal AI strategy because there is no universal business strategy. The diverse boards Fricke sits upon have AI initiatives that, if drawn as a Venn diagram, would represent completely detached circles with zero overlap. To illustrate how corporate context dictates technology decisions, Fricke categorizes enterprises into four broad archetypes:

| Enterprise Archetype | Primary Moat & Context | Target Strategic Return | Board Governance Priority | |:--- |:--- |:--- |:--- | | **Preservers** | Highly regulated sectors (banking, healthcare, insurance); heavy compliance and stakeholder guardrails. | Decision quality and resilience. | Extreme caution around patient/client risk; back-office labor tweaks are immaterial. | | **Scalers** | High-growth digital platforms embedding AI directly into core products. | Product-level alpha to outpace market rivals. | Aggressive product differentiation beyond mere back-office efficiencies. | | **Reinventors** | Mid-transformation legacy enterprises adapting to changing operating environments. | Strategic optionality and organizational learning. | Iterative exploration to modernize capabilities without destabilizing core business. | | **Disruptors** | AI-native challengers free from legacy infrastructure, tech debt, and compliance overhead. | High-velocity market disruption and category creation. | Bold, aggressive bets without traditional institutional constraints. |

Fricke emphasizes that if a Preserver and a Disruptor place the exact same technological bet, both are fundamentally wrong. The Preserver would assume existential risks intolerable to its regulators and patients, while the Disruptor would display uncharacteristic timidity when its sole structural advantage is speed and audacity.

---

[14:33](https://www.youtube.com/watch?v=cC8CR7B18A0&t=873s) The Four Questions Every Board Asks Before Writing the Check

To bridge the communication divide between technical advocates and directors, Fricke provides the four-question framework used in the boardroom to filter pitches:

1. **What is the problem we are solving?** Board members are constantly pitched new, shiny toys under the banner of innovation. Directors need to know the specific business failure or friction point being eliminated, rather than hearing generic excitement about deploying AI for its own sake. 2. **What kind of return are we buying, and is it matched to our archetype?** Pitching minor labor-efficiency gains to a Preserver board fall on deaf ears. Fricke notes having well-intentioned discussions with insurtech vendors pitching minor operational cost savings for desk-bound office staff; in an insurance company's audited accounts, that expense is practically immaterial. If an existing team performs adequately, marginally lowering their headcount costs offers negligible strategic value. 3. **What is our pathway: experimenting, piloting, scaling, or waiting?** In conservative, large-scale enterprises, leadership will deliberately wait for established, vetted enterprise vendors to ship an AI solution that is "close enough." A dependable, lower-risk vendor with an understandable pricing structure will almost always triumph over a startup offering a technically superior solution accompanied by governance risks and vendor lock-in. Crucially, the pathway must explicitly articulate where human judgment remains in the loop. 4. **How will we measure success beyond productivity, and what would make us halt?** Proposals must define rigorous, non-productivity success criteria alongside pre-agreed triggers that force a project shutdown before sunk costs run out of control.

---

[17:06](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1026s) Execution Failures, Tech Debt, and the Biological Mandate of Discernment

Looking ahead over the next few years, Fricke predicts a widespread, quiet retreat from numerous corporate AI initiatives that were initially trumpeted with immense fanfare. These retrenchments will rarely be labeled as formal write-downs; instead, public relations teams will frame them as strategic consolidations or dismiss the technology as overhyped. In reality, these retreats will not stem from technical software bugs, but from systemic governance failures where executives bought into raw capability while skipping the arduous organizational work of redesigning workflows. "AI doesn't die in the pitch," Fricke reminds the room, "it dies in the execution."

From her vantage point as an audit committee chair, the financial ramifications are already surfacing: capitalized technology investments are facing severe balance-sheet impairment write-downs as legacy software is hastily replaced with ill-governed AI tools. This accumulating tech debt hits operating earnings directly.

Reflecting on governance accountability, Fricke quips about the non-negotiable boundary of human leadership: > "As a director, let me tell you the one thing, the only thing that AI can't do that I can as a director on a regulated board and that is go to jail. The human desire to have someone to whip in the town square to criticize and to send to jail... I am the biological sacrifice that will continue to exist for that."

While artificial intelligence rapidly closes the capability gap on tasks requiring wisdom, empathy, and analytical judgment, human leaders retain the ultimate responsibility for "AI discernment." True discernment means grasping which returns matter to your specific enterprise, knowing which critical judgments must remain human, resisting shiny distractions, and having the courage to wait.

- Failed AI initiatives will increasingly be masked under the guise of corporate reorganizations and consolidations rather than admitted failures. - Unplanned impairments on capitalized legacy technology being replaced by AI represent a tangible, bottom-line financial drain that audit committees scrutinize closely. - Legal, fiduciary accountability cannot be outsourced to software; regulatory frameworks require a biological human to shoulder governance liabilities and face penal consequences. - The definitive competitive advantage for leaders going forward is "AI discernment"—the disciplined judgment to select appropriate returns, time investments correctly, and anchor strategy in enterprise context.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=4s) the question is wrong. — Leah Fricke

> [02:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=124s) technology creates capability and leadership and institutions determine if that capability is ever converted into any kind of value. — Leah Fricke

> [02:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=124s) what return actually matters for us at this point in our strategy and what is the cost of waiting? — Leah Fricke

> [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) ChatGPT can tell you all about it — Leah Fricke

> [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) The model is not your moat. — Leah Fricke

> [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) scraps you are allowed to keep. — Leah Fricke

> [17:06](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1026s) AI doesn't die in the pitch, it dies in the execution. — Leah Fricke

> [17:06](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1026s) As a director, let me tell you the one thing, the only thing that AI can't do that I can as a director on a regulated board and that is go to jail. The human desire to have someone to whip in the town square to criticize and to send to jail... I am the biological sacrifice that will continue to exist for that. — Leah Fricke

## Mentioned Leah Fricke, ChatGPT, Silicon Valley, China, Australia, Ai4

## Video description Leah Fricke - Board Director/Chair Audit Committee, Various Companies (Australia)

In my job, the only thing AI can't do is go to jail.

Accountability, judgment, the human on the hook for the call: that's what boards are really weighing when they decide whether to fund AI, and it's one of the biggest reasons strong cases never get approved.

Drawing on years spent on the boards where technology investments get approved, or quietly killed, I'll take you to the other side of the table and show you how directors actually decide, starting with a better question than "what's the ROI of AI?"

I'll share a practical framework built around the four questions every board asks before it funds AI, and the forms of return, beyond productivity, that separate an investable case from a science project.

Whether you build AI, sell it, or have to get it funded inside your own organisation, you'll leave knowing what the people who write the big cheques are really asking.

[Source](https://www.youtube.com/watch?v=cC8CR7B18A0)

中文简报

Fricke称：董事会因执行瓶颈否决AI支出企业AI投资的失败并非源于模型缺陷，而是因为企业缺乏吸收工作流程重构的运营能力，从而使董事面临资产减记和监管责任风险。 Leah Fricke认为，对企业而言，传统的“AI投资回报率（ROI）”是个提错的问题，因为其价值完全取决于组织的变革能力与战略语境，而非算法能力。真正的企业护城河存在于制度性判断力与专有语境中，而非租用的第三方基础模型。她概述了四种战略性非财务回报——决策质量、韧性、选择权和阿尔法收益，并将企业划分为“守护者”与“颠覆者”等需要截然不同技术策略的类型。最后，Fricke强调，AI项目往往在执行阶段而非立项提案阶段失败，并着重指出人类领导者承担着不容推卸的信托责任，因为AI不可能去坐牢。针对企业在 AI 领域巨额投入与衡量回报之间的悖论，Leah Fricke 指出“AI 的 ROI 是多少”本身就是一个错误问题；同款软件在不同组织中的成效差异极大，核心在于该投资能否转化为持久的企业价值而非短暂的微弱生产力提升。 (0:04) 企业推进 AI 的真正卡点不是模型性能，而是技术垂直爆发与组织制度吸收速度之间的严重脱节；当企业的“变革负荷（change load）”已达上限时，哪怕账面回报再完美的项目也会被董事会果断否决。 (2:04) 前沿模型本身无法构成企业的防御护城河，单纯基于外部大模型搭建业务无异于向供应商租借壁垒，而真正的竞争壁垒只能来源于专有数据、深度的业务上下文理解以及合规与监管驾驭能力。 (5:01) 董事会审核 AI 预算时聚焦于四大非财务战略回报：决策质量、抗风险韧性、战略选择权与独特竞争优势（Alpha），且必须与企业所处的“守成者”、“颠覆者”等组织原型严格契合。 (8:13) 获得董事会资助需厘清四大问题：明确解决的具体业务痛点、契合企业原型的回报类型、落地的实施路径（包含人类决策参与点及是否选择等待成熟供应商），以及设定非生产力衡量标准与明确的止损退出机制。 (14:33)

Reading summary

Company director Leah Fricke argues that calculating a single financial ROI for artificial intelligence is fundamentally flawed, as board-level funding decisions depend on organizational change capacity, risk context, and non-financial strategic returns. The Disconnect Between AI Capability and Enterprise Absorption (0:04) Tens of billions of dollars have flowed into enterprise AI initiatives, but industry data reflects a stark divide: 95% of organizations report no measurable financial return, while 75% claim their deployments are performing well. Fricke asserts this discrepancy is not a reporting failure but evidence that identical software yields different value depending on the organization. The true choke point is the gap between AI's rapidly accelerating technical capabilities and the human speed of organizational coordination and governance. Even when pitches present high projected returns, boards routinely reject initiatives if an organization lacks the operational bandwidth to handle additional disruption, a constraint Fricke terms 'change load.' Directors focus less on raw technology potential and more on whether their specific enterprise can absorb the upheaval, what category of return actually matters for their strategy, and the cost of waiting versus moving immediately. Foundational Models and the Illusion of Competitive Moats (5:01) Drawing on her experience at Australia's first successful publicly listed fintech during the dot-com era—where survival hinged on navigating anti-money laundering regulations and export frameworks rather than superior code—Fricke warns that external software models do not create defensible competitive moats. With frontier models shifting continuously, open-source alternatives following closely behind, and international developments threatening software pricing, relying on commercial foundation models amounts to renting capability from vendors who will capture most of the economic surplus. Sustainable enterprise moats depend instead on proprietary data, organizational judgment, and institutional context. For boards, true return comes from establishing proprietary 'alpha'—capabilities competitors cannot easily copy—rather than settling for standard technology adoption that merely preserves operational parity. Evaluating Strategic Returns Through Enterprise Archetypes (8:13) Because financial profit alone rarely justifies technology investments at the board level, Fricke identifies four alternative forms of strategic return: improved decision quality, operational resilience against market shocks, strategic optionality to maneuver over time, and proprietary competitive advantage. How these returns are prioritized depends entirely on an organization's specific profile: * **Preservers:** Regulated entities such as banks, insurers, and healthcare providers that prioritize decision quality and resilience over marginal back-office labor savings. * **Scalers:** High-growth digital platforms embedding AI into core offerings to secure product-level differentiation. * **Reinventors:** Legacy companies executing mid-transformation adaptations to acquire optionality without destabilizing existing operations. * **Disruptors:** AI-native startups unburdened by legacy tech debt or compliance overhead, competing primarily on speed and aggressive category creation. Fricke stresses that applying the same AI strategy across different archetypes can be disastrous; an identical deployment could impose unacceptable regulatory risks on a Preserver while causing a Disruptor to lose its core advantage of speed. Governance Filters and the Financial Toll of Execution Failure (14:33) To evaluate capital allocations, boards apply four core screening questions: what specific business failure is being solved, what category of return matches the enterprise archetype, what is the implementation pathway (experimenting, piloting, scaling, or waiting), and what criteria define success or trigger an early project shutdown. Conservative institutions often deliberately choose to wait for established enterprise vendors offering 'close enough' solutions rather than adopting superior startup tools that bring governance uncertainties or vendor lock-in. Fricke forecasts a quiet retrenchment of corporate AI initiatives over the coming years, driven not by software flaws but by failures to redesign human workflows. These setbacks will appear as corporate reorganizations and balance-sheet tech debt impairments when legacy systems are hastily replaced. Ultimately, legal and regulatory accountability cannot be delegated to algorithms, leaving human leadership and disciplined discernment as the non-negotiable backstops of governance. Fricke concludes that the decisive capability for leadership is 'AI discernment'—the capacity to match technology investments to enterprise constraints, resist market hype, and determine precisely where human judgment and fiduciary accountability must remain in control.

精炼阅读摘要

澳大利亚资深董事兼审计委员会主席 Leah Fricke 指出，传统的人工智能投资回报率（ROI）计算在董事会层面往往失效，企业 AI 举措受阻的根源在于技术能力跃迁与组织吸收能力之间的结构性脱节。董事会真正考量的核心并非孤立的财务指标，而是企业当前的变革负荷、战略回报类型、组织原型适配度以及无法转嫁的人类治理责任。虚假的回报悖论与组织吸收力瓶颈 (0:04) 近年来全球企业在 AI 领域投入了数百亿美元，然而行业调研却呈现出极端矛盾的数据：主流研究报告显示 95% 的企业未能获得可衡量的财务回报，而另有高管调查却声称 75% 的组织认为其项目表现良好。Fricke 认为，这种落差并非统计或信息失真，而是因为“ROI 这一提问本身就是错误的”——不同企业无法从完全相同的软件工具中提取同等价值。技术决策在董事会中从来不是单凭理论公式完成的纯粹数学推演。在董事会审议中，技术本身极少构成真正障碍，核心堵点在于前沿大模型垂直上升的能力轨迹与人类组织缓慢的协调速度之间的裂痕。Fricke 特别强调了“变革负荷”（Change Load）这一指标：即便技术方案无懈可击且纸面 ROI 极佳，若企业当前的运营带宽已无力吸收进一步的动荡，董事会依然会予以否决。Fricke 明确指出，技术负责创造能力，而唯有领导力与组织机制才能决定这种能力能否转化为实际价值。基础模型并非护城河与供应商租金陷阱 (5:01) 结合自己在互联网泡沫时期担任澳大利亚首家成功上市金融科技公司第 13 号员工的经历，Fricke 回顾了企业当年虽具绝对技术优势，却险些被反洗钱（AML）监管草案彻底扼杀的危机。该企业最终得以存续，并非依赖算法优势，而是管理层通过密集的政府关系沟通与教育对话，协助监管机构首次理解了“服务出口”的战略内涵。历史经验表明，纯粹的技术领先难以构成企业抵御外部冲击的真正防线。基于此，Fricke 警告现代企业切勿将底层模型视为护城河（The model is not your moat）。前沿模型王座频繁易主，低成本开源模型快速跟进，来自全球各地的竞争随时可能颠覆现有的商业模式。如果一家企业的核心价值完全建立在外部通用大模型之上，实质上只是向供应商“租赁护城河”，最终供应商将凭借定价权攫取绝大部分经济剩余。企业真正的护城河仅存在于无法被电子表格量化的机构判断力、专有数据沉淀以及具体的业务上下文理解之中。董事会评估的四种非财务战略回报 (8:13) 在董事会评估技术提案的现实语境下，单纯的短期账面利润极少成为决定性考量。Fricke 提出了董事会更看重的四种非财务战略回报形态：首先是“决策质量”（Decision Quality），即提升高风险核心决策的速度、精度与深度，以此跑赢竞争对手并满足严苛的监管要求；其次是“韧性”（Resilience），即组织吸收市场震荡与结构性压力而不至于破裂的缓冲能力。第三种是“期权性”（Optionality），指企业保持在自身风险承受范围内按自身节奏行动的战略灵活性，既可在时机成熟时果断抢占先机，亦可以低成本小步试错；最后是真正的“超额竞争优势”（Alpha）。Fricke 强调，必须将维持行业平均运营成本的“入场底牌”与真正的 Alpha 区分开来。追平同行标准仅能保证企业留在牌桌上，只有竞争对手无法轻易复制的核心能力才能算作战略回报。企业战略原型与董事会立项四问 (11:17) 由于企业不存在统一的商业战略，AI 战略也绝无通用模板。Fricke 将企业划分为四大原型：受强监管约束的“守成者”（Preservers，如银行、医疗和保险），其治理重点在于保障患者与客户安全，后台人效微调在其财报中微不足道；嵌入核心产品的“规模化者”（Scalers），追求产品级 Alpha；处于转型期的“重塑者”（Reinventors），注重期权性与组织学习；以及轻装上阵的“颠覆者”（Disruptors），追求极速市场颠覆。两类极端原型若采取相同的技术下注方式，必然导致灾难性后果。为打破技术团队与董事会之间的沟通壁垒，Fricke 归纳了董事会在签署支票前必问的四个问题：第一，这项投资正在解决什么具体的业务失效或摩擦点？第二，企业正在购买何种回报，它是否与自身原型匹配？第三，落地路径究竟是实验、试点、规模化扩张还是主动等待？Fricke 特别指出，保守型成熟企业往往更倾向于等待经过合规检验的成熟厂商交付“足够好”的方案，而非承担初创团队的治理风险与锁定效应；第四，如何制定超越生产力的衡量标准，以及触发终止项目的预设红线是什么？执行败局、技术债务与人类法律责任 (17:06) 展望未来数年，Fricke 预测企业界将出现一轮对早期高调宣传的 AI 项目的隐性撤退，公关团队通常会将其包装为业务整合或归咎于技术炒作。但实质上，AI 项目鲜少死于立项演讲，而是死于工作流重新设计的漫长执行期。从审计委员会的视角来看，仓促引入治理不足的 AI 工具替换遗留软件，正在导致资本化技术资产面临严重的资产减值损失，这笔不断累积的技术债务已开始直接冲击企业的营业利润。在合规与治理责任层面，Fricke 指出人类领导力存在不可替代的制度底线：算法无法承担法律后果，监管框架始终需要有血有肉的人类董事作为承担问责与刑事处罚的“生物学牺牲品”。因此，随着 AI 在智识与分析能力上逼近人类，企业高管最核心的不可替代优势将是“AI 辨别力”（AI Discernment）——即清晰识别何种回报契合企业自身、洞察哪些关键裁量权必须牢牢掌握在人类手中、抵御新潮工具的盲目诱惑，并具备在不确定环境中保持审慎等待的战略定力。 Leah Fricke 总结认为，企业引入 AI 的真正挑战在于治理重塑而非技术选型。在模型商品化与供应商寻租的背景下，董事会必须摆脱狭隘的生产力指标崇拜，将审议重心聚焦于组织吸收带宽、非财务战略护城河以及法律问责底线；能否建立与自身原型严密契合的“AI 辨别力”，将决定企业是在执行败局中承受资产减值，还是在审慎治理中转化出持久的组织价值。

Detailed notes — zh

## Background 该视频由频道 Ai4 于 2026 年 8 月 19 日发布在 YouTube 平台上。视频主讲人 Leah Fricke 担任多家澳大利亚公司的董事会董事兼审计委员会主席，内容主要分享董事会在审批 AI 投资时真正衡量的核心问题、责任归属以及超越传统 ROI 的非财务回报框架。 ## TL;DR 演讲指出企业单纯向董事会汇报 AI 的投资回报率（ROI）存在认知脱节，因为技术的突破往往受限于人类组织的消化与治理能力。真正的竞争护城河在于独特的业务判断力与上下文，而非频繁易主的基础大模型；董事会更看重决策质量、韧性、选择权和竞争优势这四种非纯财务回报。此外，不同战略原型（守成者、规模扩张者、重塑者、颠覆者）需要匹配截然不同的 AI 推进路径，企业必须警惕治理失败与流程脱节导致的盲目上线。演讲强调人类不可替代的核心壁垒在于“AI 洞察力”以及最终必须由自然人承担的法律与伦理兜底责任。 ## Atmosphere & style 演讲展现出资深董事会成员特有的清醒、老练与高度务实的专业气场。主讲人以犀利而幽默的洞见（如“AI 唯一做不到的就是坐牢”）直击当下 AI 狂热背后的盲区，用亲身经历与严密框架层层剥离技术光环。整场分享节奏从容沉稳，兼具宏观治理的冷静克制与深层战略启发，带给听众一种穿透商业表象、直面决策本质的震撼感与获得感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=4s) 董事会对 AI 项目的考量并非基于单一的 ROI 财务公式，而是关注技术能否转化为企业级价值，而非仅停留于局部的生产力提升。 - [03:22](https://www.youtube.com/watch?v=cC8CR7B18A0&t=202s) 技术本身只创造可能性与能力，组织的领导力和承载制度才决定了能力能否转化成真实价值，需重点考量组织的变更负荷（Change Load）与等待的代价。 - [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) 底层 AI 大模型无法构成企业的核心护城河，单纯依附大模型等于向供应商租赁壁垒；真正的护城河是企业专有数据、业务上下文和组织判断力。 - [08:13](https://www.youtube.com/watch?v=cC8CR7B18A0&t=493s) 董事会审视 AI 回报时涵盖四种非纯财务维度：决策质量（Decision Quality）、组织韧性（Resilience）、选择权（Optionality）以及竞争优势（Competitive Advantage）。 - [11:17](https://www.youtube.com/watch?v=cC8CR7B18A0&t=677s) 不存在普适通用的 AI 战略，企业需根据自身原型（守成者、规模扩张者、重塑者、颠覆者）匹配完全不同的 AI 投资策略与风险偏好。 - [14:33](https://www.youtube.com/watch?v=cC8CR7B18A0&t=873s) 向董事会提案 AI 项目时，必须首先剥离技术光环、明确解决的核心业务痛点，并确保所追求的回报形式与企业战略定位高度契合。 - [15:31](https://www.youtube.com/watch?v=cC8CR7B18A0&t=931s) 在大型受监管企业中，“战略性等待”是合理路径；选择高信任度、成本清晰且方案“足够接近最佳”的成熟供应商，往往胜过盲目追求技术性能但风险不可控的初创方案。 - [17:00](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1020s) AI 不会死于提案 PPT，而会死于实际执行；若缺乏对工作流程的根本性重塑，技术落地极易演变成预算失控与资产减值的治理失败，董事会须预设叫停机制（Kill Switch）。 - [18:30](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1110s) AI 唯一做不到的是替人类承担法律与监管责任；人类真正的核心壁垒在于综合宏观决策、知晓何时克制与等待的“AI 洞察力（AI Discernment）”。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=cC8CR7B18A0&t=4s) 重新审视 AI 的投资回报率（ROI）：董事会视角的认知脱节在当前大量讨论聚焦于“AI 投资回报率（ROI）”的背景下，董事会的视角往往截然不同。如果企业在向董事会汇报 AI 战略时，依然采用董事会不信任或根本不在意的语言和指标，项目就会面临停滞的巨大风险。过去几年中，全球投入到 AI 项目中的资金高达数百亿美元，但市场研究却给出了相互矛盾的极端结论：一方面，有主流研究指出高达 95% 的企业没有看到任何可衡量的财务回报；另一方面，又有调研显示 75% 的企业认为自己的 AI 应用进展顺利。 ```mermaid flowchart TD accTitle: AI投资回报率认知差异与董事会决策逻辑 subgraph Gap["认知冲突与问题重塑"] A["95% 企业未见可衡量回报"] --- B["75% 企业自认表现良好"] A & B --> C{"核心问题不是信息偏差 而是提问方式错误"} end subgraph Decision["董事会考量的核心维度"] D["组织变更负荷 (Change Load)"] E["能力吸收与治理边界"] F["非财务回报形式"] C --> D & E & F end subgraph Impact["战略决策重构"] D & E & F --> G["技术仅提供可能性 (Capability)"] G --> H["领导力与机制转化为企业价值 (Value)"] end ``` 这种“相同技术却得出完全相反结论”的现象，常常被误认为是信息不对称的问题。但事实上，这是因为提问的切入点本身就错了。该差距反映出：并非所有公司都能通过同一款产品获得相同的成果，商业落地能否成功的细微差别，远非高管或 CEO 调查报告中的简单标题数字所能概括。董事会成员需要帮助管理层厘清：究竟哪些 AI 能够创造切实的“企业级价值（Enterprise Value）”，而哪些仅仅是带来了一些局部的“生产力提升（Productivity Uplift）”。 - 董事会的决策从来不是基于单一纯粹的 ROI 公式，而是受多重复杂因素制约。 - 企业数字化转型屡屡受挫的根源极少出在技术本身，而在于“技术能做到的”与“组织能承载、消化并进行有效治理的”之间存在巨大鸿沟。 - AI 的技术能力正以近乎垂直的速度迅猛演进，但组织与制度的演进只能维持“人类速度”，其中的堵点几乎全与人相关，这也正是价值大量流失的盲区。 --- ### [03:22](https://www.youtube.com/watch?v=cC8CR7B18A0&t=202s) 组织变更负荷与“等待的代价” 董事会评估 AI 项目时的一个核心指标是“组织的变更负荷（Change Load）”。即便团队带着看似能创造巨大价值的 AI 工具、完备的项目计划和漂亮的 ROI 测算模型来到董事会，如果当前企业已经不堪重负、无法消化该项变革，那么算出来的 ROI 就毫无意义。因此，核心问题需要从“AI 的 ROI 是什么”重构为：“在我们当前的战略节点上，究竟什么样的回报才真正重要？我们选择等待的代价（Cost of Waiting）又是什么？”董事会在面对这个问题时经常犯两种错误：要么资助了错误类型的回报，要么为了追求确定性而过度等待，最终错失良机。归根结底，“价值并不存在于技术本身；技术创造的是能力，而领导力和组织制度才决定了这种能力能否转化为真实的价值。” --- ### [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) 技术不是护城河：从 Dotcom 亲历看监管与判断力的价值技术优势并不天然等于生存能力。在互联网泡沫（Dotcom boom）时期，澳大利亚第一家成功 IPO 的金融科技公司拥有绝对的技术领先地位。然而，当时公司面临反洗钱（AML）法规的重大生存危机，若按原定方案执行，业务将直接“落地即死”。最终拯救公司的并非技术代码，而是团队通过政府关系沟通、游说、深入探讨产品重要性，推动了法规的合理制定；同时协助出口监管机构理解了澳大利亚历史上首次出现的“服务出口”概念。如今的 AI 大模型同样不能构成企业的核心护城河。最先进模型的头衔（Frontier Crown）频繁易主，更廉价的开源与替代模型紧随其后；尤其在亚太地区，中国正在发生的 AI 技术演进将直接颠覆许多美国当前吹捧的商业模式。如果一家企业的核心价值依附于某个大模型，实质上只是“从供应商那里租赁竞争护城河”：供应商随时会通过调整商业定位攫取他们想要的全部利润，只留给企业残羹剩饭。 - **真正的护城河**：来自组织的业务判断力、专属专有数据，以及唯有自身拥有的独特业务上下文（Context）。 - 这类核心资产无法在任何单纯的 ROI 数学模型中直接体现，但在计算 ROI 之前就已经决定了最终胜败。 --- ### [08:13](https://www.youtube.com/watch?v=cC8CR7B18A0&t=493s) 董事会关注的四种非纯财务回报在董事会的实际考量中，AI 项目的回报不能仅用利润或纯财务数字来衡量，而是涵盖四个核心维度： | 回报类型 | 核心内涵与业务价值 | |:--- |:--- | | **决策质量 (Decision Quality)** | 提升判断与决断的准确性，借此击败对手、赢得客户信赖、满足监管合规要求，确立企业在市场中立足的合法性。 | | **韧性 (Resilience)** | 增强企业在遭受压力与外部冲击时的承受与存活能力。在动荡市场中，具备更强韧性的组织才能生存下来。 | | **选择权 (Optionality)** | 让企业能够按自身节奏、在适当时机以较低成本采取行动。既能在风险可控范围内争做“先行者”，也能灵活应变。 | | **竞争优势 (Competitive Advantage)** | 分为两层：一层是跟上市场底线标准、避免自身成本基准过高；另一层是构建对手难以复制的独特能力（即超额收益 Alpha）。 | --- ### [11:17](https://www.youtube.com/watch?v=cC8CR7B18A0&t=677s) 企业四大原型与匹配的 AI 战略世界上不存在普适通用的 AI 战略。董事会下属各家公司的 AI 战略就像没有任何交集的独立圆圈。根据战略成熟度与市场环境，企业大致可以划分为四种原型： ```mermaid flowchart TD accTitle: 四大企业原型及其AI战略匹配 subgraph Preservers["守成者 (Preservers)"] P1["银行 / 医院 / 保险公司"] --> P2["护城河: 强监管与合规信任"] P2 --> P3["核心追求: 决策质量与韧性"] end subgraph Scalers["规模扩张者 (Scalers)"] S1["高增长平台型企业"] --> S2["将 AI 深度植入产品核心"] S2 --> S3["核心追求: 产品 Alpha 与市场制胜"] end subgraph Reinventors["重塑者 (Reinventors)"] R1["处于业务转型期的企业"] --> R2["探索新模式与能力迭代"] R2 --> R3["核心追求: 选择权与组织学习"] end subgraph Disruptors["颠覆者 (Disruptors)"] D1["AI 原生挑战者"] --> D2["无历史技术债与遗留包袱"] D2 --> D3["核心追求: 大胆创新与模式颠覆"] end ``` - **守成者（Preservers）**：如银行、医院、保险公司等强监管行业。其护城河本身就是监管牌照与信任，面临严格的客户与患者预期，无法在未充分测试风险前盲目创新。其追求的回报聚焦于**决策质量**与**组织韧性**。 - **规模扩张者（Scalers）**：高增长的平台型企业。他们不仅仅追求后台与日常运营的效率提升，而是通过将 AI 深度融入产品本身来击败市场竞争对手，追求业务上的 **Alpha**。 - **重塑者（Reinventors）**：处于业务转型中期的企业，通过 AI 项目获取**选择权（Optionality）**和**组织学习能力**。 - **颠覆者（Disruptors）**：AI 原生（AI-native）挑战者，没有任何历史技术债务、文化顾虑或沉重的监管包袱。如果“守成者”和“颠覆者”采取完全相同的 AI 押注，两者皆会失败：守成者承担了利益相关方无法接受的合规与信任风险，而颠覆者则在需要大胆激进时显得过于保守胆怯。面对同一种技术，不同原型的正确策略完全相反。 --- ### [14:33](https://www.youtube.com/watch?v=cC8CR7B18A0&t=873s) 董事会审批 AI 项目时的核心提问当 AI 提案递交到董事会审议时，董事们主要通过一系列底层逻辑进行审视： - **第一问：我们要解决的核心问题究竟是什么？** 大家都喜欢新奇好玩的工具，热衷于拥抱创新和启动变革项目，但董事会首先要剥离技术光环，明确该项目解决的实际业务痛点，而非单纯为了用 AI 而用 AI。 - **第二问：我们买入的究竟是何种形式的回报，它与企业的战略原型是否匹配？** 例如在与许多保险科技（InsurTech）公司沟通时，他们往往立足自身专业设想宏大的转型方案，但若其方案仅仅作用于轻微压低运营成本，却忽视了保险公司作为“守成者”在合规与风险控制上的根本诉求，便无法形成真正具备投资价值的商业论证。 ### [15:31](https://www.youtube.com/watch?v=cC8CR7B18A0&t=931s) 企业的推进路径抉择：实验、试点、规模化还是等待？对于董事会和审计委员会来说，单纯用 AI 去替代办公室里某个初级员工的常规琐事，在财务账目上往往是“微不足道”（immaterial）且价值最低的产出。如果现有员工已经能够妥善完成这些基础工作，这种优化根本谈不上革命性变革，董事会也不会在此类项目上投入太多精力。董事会真正会抛出的第三个核心问题是：“我们在此处的推进路径究竟是什么？我们是在实验、在运行试点、在进行规模化推广，还是在选择等待？” 在大型或风险规避型企业中，“等待”往往是一种极具战略意义的明智选择。很多时候，企业在现有技术栈和供应商生态中等待的，是一个经过严格审查、值得信赖的成熟供应商。企业宁愿选择一个来自高信任度供应商、具备清晰易懂成本模型且效果“足够接近最佳”（close enough）的方案，也不愿为了追求理论上更优的产出，去冒险采用低信任度供应商的方案，从而承担潜在风险或面临供应商绑定的困境。因此，选择走哪条路径、与谁同行，以及在整个推进过程中“人类判断力保留在何处”，是董事会审议时不可或缺的核心议题。 ```mermaid flowchart TD accTitle: 企业 AI 推进路径与供应商决策权衡 subgraph Pathway["推进路径选择 (Pathways)"] direction LR P1["实验探索 (Experimenting)"] P2["局部试点 (Piloting)"] P3["规模推广 (Scaling)"] P4["战略等待 (Waiting)"] end subgraph Selection["供应商与方案权衡"] direction TB A["成熟可信供应商 • 经过严格审查 • 成本模型清晰 • 方案效果『足够接近最佳』"] -->|优选：低风险/可控| C["董事会批准推进"] B["低信任度新兴供应商 • 理论产出更优 • 存在供应商锁定风险 • 治理与安全隐患"] -->|审慎/否决| D["面临搁置风险"] end Pathway --> Selection ``` - **路径定性**：明确项目到底处于概念验证（PoC）、业务试点，还是全组织规模化推广，避免阶段目标错位。 - **信任优于微弱的技术优势**：在大型受监管组织中，“可解释的商业模型与受信任的供应商”通常会击败“性能略高但风险不可控的初创方案”。 - **人机边界**：明确在业务流的哪一个具体节点必须保留人类的最终裁决与责任担当。 --- ### [17:00](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1020s) 衡量成功的指标与治理失败：AI 不会死于提案，而会死于执行董事会评估 AI 的第四个关键问题是：“除了生产力之外，我们如何衡量成功？以及在什么情况下我们会叫停项目？”立足于对未来两三年的行业观察，市场上极有可能出现一种趋势：许多此前信心满满对外宣称的 AI 项目，随后将经历一场“悄无声息的撤退”（quiet retreat）。这种撤退在财报上不一定表现为直接的大额资产减记，可能体现为业务整合，或者被外界解读为“AI 炒作泡沫破裂”以及“某些企业缺乏执行能力”。但这往往不是底层技术的溃败，而是一场彻头彻尾的“治理失败”（governance failure）。许多企业仅仅买入了 AI 的技术能力，却“跳过了最艰难的环节”——没有围绕新技术对实际工作流程进行根本性的重新设计。正如演讲者所强调的：“AI 不会死在提案 PPT 里，它会死在实际执行中（AI doesn't die in the pitch, it dies in the execution）。” - **预算失控与客户体验恶化**：缺乏流程再造的盲目上线往往直接导致支出超支与负面客户反馈。 - **资本化技术成本的减值损失**：作为审计委员会主席，当审视那些被 AI 迅速淘汰的旧技术资产时，资本化软件与系统计提的减值损失是会直接侵蚀企业底层净利润（bottom line）的真实数字。 - **明确叫停机制（Kill Switch）**：负责任的董事会必须预先设定触发项目终止的具体指标，而非任由项目在沉没成本中无休止消耗资源。 --- ### [18:30](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1110s) 无法被替代的生物学代价与“AI 洞察力” 探讨人机分工的核心，最终会落到“什么是人类所独有的”这一终极追问上。作为受监管机构的董事会董事，演讲者一针见血地指出：在整个董事会职责中，AI 唯一做不到的事情就是“坐牢”（the only thing AI can't do is go to jail）。人类社会始终存在一种本能需求——必须有一个具体的自然人承担法律与公众责任，在广场上接受审判并承担最终的“生物学牺牲”（biological sacrifice）。更深层次的人类核心价值，并非单纯的智慧、同理心或判断力，因为 AI 正在这些维度快速追赶人类。人类真正的核心壁垒在于**AI 洞察力（AI Discernment）**。这是一种宏观维度的综合决策能力，涵盖了理解组织所追求的回报本质、提出切中企业命脉的根本问题、洞悉何时应当果断等待，以及清醒界定哪些核心控制权必须牢牢掌握在人类手中。 - **法律与治理的终极兜底**：算法无法承担民事赔偿与刑事责任，受监管业务的法律主体与最终合规责任永远绑定在人类董事身上。 - **AI 洞察力（AI Discernment）的内涵**： - 洞察超越单纯效率提升的多元回报形式； - 在技术狂热中保持克制，明确知晓何时选择“战略性等待”； - 牢牢守住必须由人类行使裁决权与道德判断的业务节点。 - **总结与致意**：演讲者建议参会者将“掌握并部署 AI 洞察力”作为最核心的收获，随后向全场致谢，并表示在启程返回澳大利亚悉尼之前，乐意在现场留步与参会者做进一步交流。 ## Notable quotes > [03:22](https://www.youtube.com/watch?v=cC8CR7B18A0&t=202s) 价值并不存在于技术本身；技术创造的是能力，而领导力和组织制度才决定了这种能力能否转化为真实的价值。 — Leah Fricke > [03:22](https://www.youtube.com/watch?v=cC8CR7B18A0&t=202s) 在我们当前的战略节点上，究竟什么样的回报才真正重要？我们选择等待的代价又是什么？ — Leah Fricke > [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) 如果一家企业的核心价值依附于某个大模型，实质上只是从供应商那里租赁竞争护城河。 — Leah Fricke > [05:01](https://www.youtube.com/watch?v=cC8CR7B18A0&t=301s) 真正的护城河：来自组织的业务判断力、专属专有数据，以及唯有自身拥有的独特业务上下文。 — Leah Fricke > [14:33](https://www.youtube.com/watch?v=cC8CR7B18A0&t=873s) 我们要解决的核心问题究竟是什么？ — Leah Fricke > [15:31](https://www.youtube.com/watch?v=cC8CR7B18A0&t=931s) 单纯用 AI 去替代办公室里某个初级员工的常规琐事，在财务账目上往往是微不足道且价值最低的产出。 — Leah Fricke > [15:31](https://www.youtube.com/watch?v=cC8CR7B18A0&t=931s) 我们在此处的推进路径究竟是什么？我们是在实验、在运行试点、在进行规模化推广，还是在选择等待？ — Leah Fricke > [17:00](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1020s) 除了生产力之外，我们如何衡量成功？以及在什么情况下我们会叫停项目？ — Leah Fricke > [17:00](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1020s) AI 不会死在提案 PPT 里，它会死在实际执行中（AI doesn't die in the pitch, it dies in the execution）。 — Leah Fricke > [18:30](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1110s) 在我的工作中，AI 唯一做不到的事情就是坐牢（In my job, the only thing AI can't do is go to jail）。 — Leah Fricke > [18:30](https://www.youtube.com/watch?v=cC8CR7B18A0&t=1110s) 人类社会始终存在一种本能需求——必须有一个具体的自然人承担法律与公众责任，在广场上接受审判并承担最终的生物学牺牲。 — Leah Fricke ## Mentioned Leah Fricke, Ai4, 澳大利亚 (Australia), 悉尼 (Sydney), 互联网泡沫 (Dotcom Boom), 反洗钱法规 (AML), 保险科技 (InsurTech), AI 洞察力 (AI Discernment) ## Video description Leah Fricke - Board Director/Chair Audit Committee, Various Companies (Australia) In my job, the only thing AI can't do is go to jail. Accountability, judgment, the human on the hook for the call: that's what boards are really weighing when they decide whether to fund AI, and it's one of the biggest reasons strong cases never get approved. Drawing on years spent on the boards where technology investments get approved, or quietly killed, I'll take you to the other side of the table and show you how directors actually decide, starting with a better question than "what's the ROI of AI?" I'll share a practical framework built around the four questions every board asks before it funds AI, and the forms of return, beyond productivity, that separate an investable case from a science project. Whether you build AI, sell it, or have to get it funded inside your own organisation, you'll leave knowing what the people who write the big cheques are really asking. [Source](https://www.youtube.com/watch?v=cC8CR7B18A0)

Source quotes

[3:22] Leah Fricke: 价值并不存在于技术本身；技术创造的是能力，而领导力和组织制度才决定了这种能力能否转化为真实的价值。 [3:22] Leah Fricke: 在我们当前的战略节点上，究竟什么样的回报才真正重要？我们选择等待的代价又是什么？ [5:01] Leah Fricke: 如果一家企业的核心价值依附于某个大模型，实质上只是从供应商那里租赁竞争护城河。 [5:01] Leah Fricke: 真正的护城河：来自组织的业务判断力、专属专有数据，以及唯有自身拥有的独特业务上下文。 [14:33] Leah Fricke: 我们要解决的核心问题究竟是什么？ [15:31] Leah Fricke: 单纯用 AI 去替代办公室里某个初级员工的常规琐事，在财务账目上往往是微不足道且价值最低的产出。 [15:31] Leah Fricke: 我们在此处的推进路径究竟是什么？我们是在实验、在运行试点、在进行规模化推广，还是在选择等待？ [17:00] Leah Fricke: 除了生产力之外，我们如何衡量成功？以及在什么情况下我们会叫停项目？ [17:00] Leah Fricke: AI 不会死在提案 PPT 里，它会死在实际执行中（AI doesn't die in the pitch, it dies in the execution）。 [18:30] Leah Fricke: 在我的工作中，AI 唯一做不到的事情就是坐牢（In my job, the only thing AI can't do is go to jail）。 [18:30] Leah Fricke: 人类社会始终存在一种本能需求——必须有一个具体的自然人承担法律与公众责任，在广场上接受审判并承担最终的生物学牺牲。

Source quotes

[0:04] Leah Fricke: the question is wrong. [2:04] Leah Fricke: technology creates capability and leadership and institutions determine if that capability is ever converted into any kind of value. [2:04] Leah Fricke: what return actually matters for us at this point in our strategy and what is the cost of waiting? [5:01] Leah Fricke: ChatGPT can tell you all about it [5:01] Leah Fricke: The model is not your moat. [5:01] Leah Fricke: scraps you are allowed to keep. [17:06] Leah Fricke: AI doesn't die in the pitch, it dies in the execution. [17:06] Leah Fricke: As a director, let me tell you the one thing, the only thing that AI can't do that I can as a director on a regulated board and that is go to jail. The human desire to have someone to whip in the town square to criticize and to send to jail... I am the biological sacrifice that will continue to exist for that.

Original transcript

[0:04] in a stream that's been focused on return on investment for AI, I probably have a [0:10] slightly different perspective. And I want to share that because I think for a lot of people [0:14] having conversations in enterprise around artificial intelligence when you're leading [0:19] those conversations with a language with metrics that your board don't trust or don't care about [0:29] you have a real stalling potential for anything that you want to do. Yeah. So if we think about [0:36] over the last few years, tens of billions of dollars have been um put towards AI projects and [0:44] the results are telling us some very different stories. So we've got major studies saying 95% [0:50] of enterprises have seen no measurable return and then other studies saying that 3/4 of enterprises [0:56] believe they're doing well. same technology and opposite answers. Most people are treating that [1:05] gap as an information problem. What I believe is that is telling us that the question is wrong. [1:17] The gap reflects that not all companies can achieve the same outcome with the same product [1:26] and the nuances around what will work and won't work are deeper than a headline number [1:31] from an executive or CEO survey can indicate. Wow. So I've spent my career last 10 years on [1:41] the board side of the conversation. I bring that board perspective to AI, [1:48] helping directors and execs to try to distinguish between AI that creates [1:54] actual enterprise value compared to AI that simply provides some productivity uplift. [2:04] In boardrooms, decisions about technology are not taken with the purity of a single ROI equation. [2:13] There are so many factors that boards need to keep in mind in any decision that is made. And [2:20] the problem that we deal with is that in all of our conversations about technology adoption, [2:27] transformation, change, technology is almost never our problem. [2:33] Bringing new technology doesn't solve the underlying challenge and difficulties that most [2:40] large technology transformation projects have had. The problem is the gap between what the technology [2:47] can do and what the organization can put to work and what it can govern. In the context of AI, [2:58] we're all very aware that the capability is racing ahead almost vertically and our [3:05] institutions are moving at human speed. There will always be choke points and those choke [3:11] points will often be human related. In that space between we are seeing value leaking out. [3:22] One of the measures boards are often focused on is the change load on an organization. You could come [3:31] to me for any of the boards that I sit on and say we have an AI tool, we have a capability, [3:36] we have a project plan that will add enormous value and depending on the change load that the [3:42] organization is currently bearing, that may be a project that will make sense or not make sense. [3:48] and the ROI that you have calculated and present to me in relation to that project is irrelevant [3:53] to me if the organization cannot actually digest that change at that time. The complexity of all [4:02] of the factors that boards need to keep in mind mean that that decision is a broader one for them. [4:11] So the reframe is not what is the return of AI, it's what return actually matters for [4:19] us at this point in our strategy and what is the cost of waiting and boards fail in [4:26] both directions on that question. We fund the wrong kind of return and occasionally [4:32] we wait too long for certainty and that certainty never comes. So my thesis is [4:39] that the value does not live in the technology. Technology creates capability and leadership and [4:48] institutions determine if that capability is ever converted into any kind of value. [5:01] I uh have had enough years on this planet to have worked through as an executive during the dotcom [5:09] boom. Some of you may recall that. If not, chat GPT can tell you all about it. Uh in that time, [5:16] I worked at an organization that had absolute technology superiority. First successful fintech [5:22] in Australia that IPOed. Uh I was employee number 13, lucky for some in this context. [5:29] We were faced with an existential risk that was a regulatory one. Anti-money laundering uh the [5:37] regulations that required would have meant that we could not have continued with the business closed [5:42] the door dead on arrival. It was only through government relationship management uh lobbying [5:51] technology discussions explaining the relevance and the importance of our product that meant that [5:57] those regulations were drafted in a manner that allowed the business to continue. In conjunction [6:02] with that, conversations around export value to our country and helping our export regulatory [6:11] body at the time understand for the first time in Australian history understand that you could [6:16] export a service. Those two factors are the core reasons that that entity survived. Nothing to do [6:25] with technology. That was about the capability, commitment and us as an organization being able [6:31] to explain the importance of our strategy to our customers, to our country and to others. The [6:40] model, the AI model is not your moat. The frontier crown changes regularly. Cheaper models are [6:49] right behind us. For someone who comes from the Asia-Pacific region, what is happening in China [6:56] will decimate business models that are currently celebrated in the US. If your value is living in a [7:05] model, you're renting a competitive moat from a vendor and a vendor who will increasingly [7:14] choose to use that competitive positioning to achieve a commercial outcome where they retain [7:23] the portion of value that they want. And you will be left with the scraps you are allowed to keep. [7:33] Your real moat, your real competitive advantage is in the judgment of your organization, the data, [7:44] the context that only you hold. None of that shows up in any kind of ROI model mathematically. [7:54] There is so much that comes before the ROI calculation even starts that makes [8:01] it impossible for that to reflect where the value lies and where the return is attributable. [8:13] So I think about as a director four types of return from projects. It's [8:19] not a purely financial context. We're not talking about profit. We are talking about [8:27] decision quality. There is enormous potential enterprise value in some [8:33] organizations in an improvement in decision quality. If we think about uh the components [8:46] that we would value at a board level around any new project. Decision quality is an important [8:57] component of how we beat competitors, impress our customers, make our regulators comfortable, [9:05] and broadly help our stakeholders to understand why we deserve the right to exist in a competitive [9:14] market. Another component is resilience. holding up under stress and shock. Everyone who works in [9:25] a large enterprise, a startup, small to medium enterprise recognizes in the people they see [9:33] in their market and around them, there are organizations that fail to survive in times [9:38] of stress. Having additional resilience is a key factor that will mean you will be a survivor in [9:48] your market. Optionality. Optionality. The ability to move at your pace, at your time, to take the [10:00] the right time to move and to do it cheaply. to be the first mover if that's what's appropriate for [10:07] you as an organization and to do that within your risk envelope and the final component competitive [10:17] advantage and I think this is the bit where we need a distinction between the adoption of [10:24] technology tools and capabilities so that you are meeting the market you are competing in everyone's [10:31] doing it you need to do it you need to have a cost base that is not significantly higher than your [10:37] competitors in your market. Um the idea of staying relevant, staying competitive. The other component [10:45] of that is capabilities that your rivals can't easily copy. Uh so I I think about that as the [10:54] alpha. So everyone is doing the same thing and you need to be able to do the same thing as all [10:59] of your competitors and you are winning if you have that little bit extra whatever that alpha [11:05] is that they are not yet able to copy. Yeah. So there is no universal AI strategy and what is [11:17] an appropriate AI strategy depends entirely on who you are as an organization. The boards that [11:24] I am on, the various AI strategies that we have are if they were circles, it would be completely [11:31] separate circles and zero vin diagram overlap. In organizations, you need to think about the [11:39] types of organizations, where they are in their strategic journey, uh where their market maturity [11:47] is at, their context. I think about them in a very summarized not particularly nuanced way as four [11:53] different archetypes. So we have preservers. So preservers you know regulated markets I work in [12:01] um banking I work in health. So banks, hospitals, insurance companies, we tend to be preservers. [12:08] Our main moat is a regulatory one. Uh we have uh customer client patient expectations that really [12:16] constrain our ability to dive into entirely new innovation models without real testing of the [12:27] risks that our stakeholder groups are concerned about. Our return tends to be around decision [12:36] quality and resilience. that is what we are looking for in the tools that we're adopting. [12:44] Then we think about uh scalers. So high growth platforms um those who win by putting AI into [12:52] the product. So they are looking for the alpha in a product strategy that can beat the market. [13:01] They are not only interested in back office and um dayto-day efficiency metrics. Um their focus [13:13] is on how they win by using AI. We think about reinventors who might be mid transformation [13:21] and their return is optionality and learning. uh and the ultimate which I'm sure many of you have [13:29] um seen many examples of in the trade area and in conversations during the course of this week [13:36] the disruptors who are AI native challenges and they are unbburdened by tech debt legacy [13:47] caution that that regulatory expectation that organizations like insurers banks [13:54] healthcare um are constrained by. So a preserver and a disruptor making the identical bet are both [14:03] wrong. The preserver would be taking risks that would be unacceptable to its stakeholder group [14:09] and the disruptor would be being too timid when it should be bold. So the same technology has [14:18] opposite correct decisions. There's no universal AI strategy for the same reason that there is no [14:24] universal business strategy. So how does a board actually decide? So I want to just quickly run [14:33] through the four decisions that I think about uh when AI ideas um come to any of the rooms [14:42] that I sit in. So my first question for a lot of things that come to BS is like what is the [14:49] problem we are solving? Um everyone loves a new toy. Everyone loves to be innovative. We like [14:55] projects. We like the idea that we can transform our business. Um but the key question I want to [15:02] understand is what is the problem we're solving? Not just here's an idea where we can use AI. [15:08] The second question is what kind of return are we buying and is it matched to our archetype? [15:16] I have lovely conversations with insure tech organizations who have a sense of a transformation [15:24] opportunity in their area of expertise and often in insurance if that goes to my operational cost [15:31] of staff sitting in my office it is the lowest value outcome in my world. It is immaterial in [15:37] my accounts. I don't have a great deal of time for that. If it's being adequately performed by [15:42] our existing staff, it is not revolutionary for us. The third question we ask is, you know, what [15:50] is our pathway here? Are we experimenting? Are we running a pilot? Are we scaling or are we waiting? [16:00] And sometimes in larger enterprises with within our tech stack within our vendor group what we are [16:07] waiting for is a trusted vetted vendor to explain to us how they can get something close enough to [16:15] the best possible AI outcome and close enough trusted vendor uh an understandable cost model [16:25] in larger risk averse enterprises will generally win over something that has an arguably better [16:32] outcome from a lower trust vendor where there are some risks, there may be some vendor capture [16:39] challenges. Um, and so that path and who we want to walk down that path with is really important. [16:51] asking where the judgment stays human in that pathway is an essential component of our [16:59] conversation as well. Uh so our final question, how we'll measure success beyond productivity [17:06] and what would make us halt? So I want to uh put a stake in the ground around my perspective of where [17:17] we're going on this over the next couple of years. I think what we're going to see is a quiet retreat [17:23] from some AI programs that are announced with confidence. It won't necessarily look like a write [17:31] down. It might look like consolidation. It'll be reported of proof that hype was overblown. It [17:38] might be reported as certain organizations that don't have execution capability. Um, it won't [17:45] necessarily be a technology failure, but what it will probably be is a governance failure. So, [17:53] someone bought into the capability, they skipped the hard bit. They didn't redesign their work [17:58] around it. AI doesn't die in the pitch, it dies in the execution. If we look at organizations who've [18:07] announced that their budgets have been blown out, um that they've had bad customer outcomes. As an [18:14] audit committee chair, when I look at the numbers of um impairment on capitalized technology costs [18:23] for technology that is being replaced by AI, that tech debt is a real number that hits the [18:30] bottom line. So moving on to what stays human. As a director, let me tell you the one thing, [18:40] the only thing that AI can't do that I can as a director on a regulated board and that is go to [18:47] jail. The human desire to have someone to whip in the town square to criticize and to send to [18:57] jail. I and the biological sacrifice that will continue to exist for that. More importantly, [19:04] I think where we're going is the question around what still is human. And that is [19:12] not necessarily wisdom, empathy, judgment. AI is catching up with us on that. But what is [19:18] ultimately human is understanding the types of returns you're looking for. asking the questions [19:27] that are relevant to your organization, knowing when to wait, knowing what to keep human, and all [19:34] of that has a title, which is AI discernment. So, I encourage you to think about the most [19:42] important thing you take away from this conference is understanding how to deploy AI discernment. So, [19:50] thank you for your time. If any of you would like to have a conversation, I'm happy to hang around [19:54] for a few minutes before I go and pack my bags and head back to beautiful Sydney, Australia.