Ipsos Chief Data Scientist Warns AI Time Savings Risk Vanity ROI

Is Your Insights Team Capturing the Phantom ROI of AI? With Ipsos — Ai4

Ai4

Corporate AI investments fail to generate hard financial returns unless leaders actively reinvest saved worker capacity into adjacent services, proprietary context capital, and reduced vendor dependence.

AI-assisted summary of Is Your Insights Team Capturing the Phantom ROI of AI? With Ipsos — original source. Check the original for full context.

Source published: 2026-08-19

Detailed notes

## Background This session was presented by Ai4 on YouTube on August 19, 2026, featuring Xufeng Wang, Partner & Chief Data Scientist at Ipsos. According to the page description, the presentation challenges analytics and data science leaders to move beyond conventional productivity metrics to measure the "phantom" returns of AI and connect them to hard financial outcomes.

## TL;DR Xufeng Wang argues that relying on "time saved" as a primary AI metric is a vanity exercise that fails to reflect true financial ROI and often leads to hidden human debt, such as organizational bottlenecks and surface-level understanding. To capture true AI value, organizations must actively manage and reinvest saved time rather than assuming it automatically yields profit. Using a 3D framework, Wang illustrates how teams can reinvest this time budget by expanding into adjacent service markets, building proprietary Context Capital, and reducing vendor overhead with internal AI tooling.

## Atmosphere & style The session carries an engaging, practical, and lightheartedly self-aware tone, opening with casual banter about drawing a lunchtime audience on the final day of a Las Vegas conference. Wang balances sharp strategic critique with pragmatic business insights, making complex organizational concepts accessible and grounded in real-world market research operations. The mood is candid and forward-looking, offering leaders a candid reality check on AI productivity hype paired with an empowering framework for driving genuine transformation.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=4s) Xufeng Wang oversees both client-facing AI/ML consulting and internal data lake consolidation of 50 years of survey data at Ipsos. - [02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) "Time saved" is the universal anchor metric for evaluating AI, but it is a soft vanity metric that struggles to appear on financial balance sheets. - [02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) Perceived AI velocity does not directly translate to true capacity gains, as employees often complete tasks quickly yet remain equally exhausted. - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) AI creates "phantom savings" by filling individualized skill gaps—helping engineers communicate better and enabling non-technical staff to build software. - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) Data scientists act as managers of their own autonomous AI workflows using tools like Claude and Gemini, leading to unstandardized, siloed processes. - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) Relying on quick code prototypes replaces traditional spec documents but risks blurring roles, inviting senior leadership interference, and creating technical debt. - [08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) Over-reliance on AI summaries creates "phantom understanding," where teams pass automated messages back and forth without genuinely comprehending the content. - [08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) Rapid AI execution shifts organizational bottlenecks away from production and directly onto slow corporate, IT, and operational decision-making processes. - [08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) Saved time is not an automatic financial return, but rather a temporal budget that must be deliberately captured and reinvested. - [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) The X-axis of reinvestment involves expanding into adjacent markets, such as using AI-augmented teams to compete against top-tier management consultancies. - [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) A company's historical, organic data assets gain significant competitive value in an era increasingly flooded with synthetic AI content. - [15:10](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=910s) The Y-axis focuses on capturing scattered institutional knowledge to build "Context Capital," an operational asset that can be scaled and monetized. - [15:10](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=910s) The Z-axis emphasizes reducing the "vendor tax" and administrative friction by building targeted internal AI tooling rather than buying external SaaS products.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=4s) The Unexpected Thursday Audience and the Context of Ipsos

Xufeng Wang, Partner and Chief Data Scientist at Ipsos, opens his presentation with lighthearted banter about his talk's schedule slot. Before traveling to Las Vegas, he complained to a friend that speaking on Thursday during lunchtime—the final day of the conference—meant nobody would show up. His friend jokingly reassured him to look on the bright side: many attendees would have already burned through their daily token limit by Thursday with nothing else to do, leaving them no choice but to sit in on his session. Beyond the humor, Wang highlights that fundamental operating dynamics across work, automation, data science, and insight extraction have fundamentally shifted.

Wang framing his perspective around the specific operational realities of Ipsos, a global market research giant with roughly 20,000 employees worldwide where survey research remains the core business. He oversees two distinct teams: a traditional data science and machine learning division delivering client-facing consulting projects globally, and an internal data management team focused on consolidating 50 years of historical survey data into a unified enterprise data lake. This dual viewpoint—balancing client consulting with internal asset management—informs his analysis of how organizations perceive and miscalculate the true returns on AI investments.

- **Speaker**: Xufeng Wang (Partner & Chief Data Scientist, Ipsos). - **Company scale**: ~20,000 global employees with 50 years of survey data assets. - **Dual team structure**: Client-facing AI/ML consulting and internal data lake management.

[02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) Time Saved as an Anchor and the Myth of Productivity

Looking at AI value across an enterprise often feels like a "curse" because stakeholders evaluate returns through wildly different lenses. To navigate this complexity, leaders need an initial pivot point—a baseline metric everyone can roughly agree upon. That starting anchor is almost universally "time saved." It is the core subject of water-cooler conversations, self-congratulatory mirror checks after clever ChatGPT prompts, and peer catch-ups. However, time savings is rarely discussed outside the closed doors of executive finance, primarily because leaders struggle to explain how saved hours actually convert into hard financial realities.

"All models are wrong, but some are useful," Wang notes, suggesting that even if time saved functions as a soft "vanity metric," it serves as a necessary anchor for before-and-after evaluations. The metric demands a two-step critique: first, determining *how* time was saved and whether that saving is genuine or merely a mental perception; second, identifying how to capture and reinvest that newly freed capacity. Many workers experience early LLM breakthroughs feeling as though they saved massive amounts of time, yet still end their workdays completely exhausted—proving that perceived velocity does not automatically equal actual capacity gains. Converting soft perceived efficiency into hard ROI that fits on a financial spreadsheet requires rigorous scrutiny.

- **Primary anchor**: "Time saved" serves as the initial universal metric, despite being a soft vanity metric. - **The finance gap**: Soft productivity gains struggle to translate directly onto corporate accounting balance sheets. - **The fatigue paradox**: Users report rapid AI task completion while experiencing equal or greater end-of-day exhaustion.

[04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) The Phantom Return: Talent Enablement, Distributed Management, and Prototyping Debt

While technical efficiency gains—such as AI-assisted coding, agentic workflows, and process automation—are well understood, the human dimensions of AI efficiency remain under-analyzed. Wang identifies three subtle human dynamics that shape what he calls "phantom savings." First, AI enables talent across highly diverse and unexpected directions. For instance, technical engineers who traditionally struggle with communication leverage LLMs to articulate ideas more clearly, while non-technical team members use generative tools to experiment with building software. Rather than applying uniformly, AI fills individualized skill gaps across the organizational chart.

Second, the proliferation of cloud tools and advanced models like Claude and Gemini means nearly every data scientist now acts as a manager of their own autonomous AI workforce. Senior scientists orchestrate complex agent workflows, yet each manages their synthetic "team" using drastically different logic, creating an urgent need for organizational observability. Third, teams increasingly rely on "communication via prototype." Instead of writing lengthy emails or slide decks, engineers generate quick code prototypes in minutes to pitch ideas. However, this creates a double-edged sword: senior leaders get hooked on building prototypes themselves rather than handing work off to engineering teams, eroding the clear division of labor and generating technical debt.

- **Talent gap filling**: AI bridges distinct individual weaknesses, helping engineers communicate and non-coders build software. - **Personalized orchestration**: Senior data scientists leverage Claude and Gemini to build bespoke, unstandardized automation workflows. - **Prototype trap**: Quick code prototypes replace traditional specs, but lead to senior staff dictating product direction and accumulating technical debt. - "The deepest savings come from the recognition of the growth and diversification of talent enabled by AI—it's not only coming from installing tools, and it definitely is not coming from headcount reduction."

[08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) Organizational Debt, Phantom Understanding, and Decision Bottlenecks

Every human enablement feature of AI carries an underlying hidden cost—a new form of organizational debt. Beyond the accumulation of technical "AI slop," teams face severe human risks, most notably phantom understanding. Wang cautions against an over-reliance on AI where systems process information that humans no longer comprehend. He highlights comical yet troubling workplace scenarios where one engineer feeds a long, AI-generated message into ChatGPT, extracts the summary, and replies with another generated message without either human ever reading or grasping the core content. On the surface, the team appears to operate at lightning speed, but they become entirely dependent on model accuracy without acquiring underlying mastery.

Simultaneously, accelerated production cycles shift corporate bottlenecks directly onto decision-making processes. Because teams can build prototypes, iterate on proofs-of-concept, and plan projects far faster than before, work rapidly stalls while waiting for sign-offs from IT, operations, or senior leadership. Every moment spent waiting for a corporate decision compounds in cost much faster today than it did a year or two ago. Leaders must recognize that saved time is not a net return in itself—it is merely an operational budget that must be deliberately captured and managed.

- **AI Slop & Fake Comprehension**: Automated back-and-forth messaging creates an illusion of speed while eroding actual team understanding. - **The new bottleneck**: Rapid prototyping forces project delays upward into corporate, IT, and operational decision-making loops. - **Core rule**: Saved time is not a financial return; it represents a temporal budget created by balancing technical speed against human debt.

[11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) The Swimming Framework: Allocating the Time Budget Across X, Y, and Z Axes

To turn a soft "time-saved budget" into hard financial ROI, Wang introduces an intuitive 3D "swimming lesson" framework. Leaders must decide how to reinvest their capacity by looking along three spatial axes: the X-axis (expanding into adjacent swimming lanes), the Y-axis (reimagining the forward direction of travel), and the Z-axis (adjusting buoyancy and drag to stay lean). Reinvesting saved time requires playing aggressively rather than passively expecting productivity metrics to transform into financial returns on their own.

``` [Y-Axis: Forward Direction] Reimagine operations around "Context Capital" ^ | | [X-Axis: Adjacent Lanes] <---> [Z-Axis: Vertical Lift] Build new offerings around Reduce vendor tax & leverage solid core survey data custom internal AI tooling ```

On the X-axis, organizations build new revenue streams around their solid core business. For Ipsos, full-service survey research across 17 service lines forms the core foundation. By leveraging AI to build abstract software layers and augment team output, Ipsos can deploy smaller teams of experts to directly compete against elite management consultancies like McKinsey or BCG—entering market spaces historically out of reach. Additionally, AI makes low-cost experimentation and rapid failure possible, allowing teams to test new market concepts cheaply while unlocking the hidden value of proprietary data.

- **X-Axis (Adjacent Lanes)**: Expanding service reach into high-tier consulting (e.g., competing with McKinsey/BCG) via augmented expert capacity. - **Organic Human Voice**: The value of 50 years of trusted, organically collected survey data appreciates significantly in the age of synthetic AI content. - **Fail-fast experimentation**: Lowered technical barriers allow rapid, inexpensive testing of new business models around core assets.

[15:10](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=910s) Context Capital and Eliminating the Vendor Tax

Along the Y-axis, organizations must rebuild operational workflows around an invisible new asset: "Context Capital." Over decades, critical business knowledge, standard operating procedures, and institutional memory remain scattered across veteran employees' heads. Organizations must systematically capture this operational expertise into a central context layer or database. This contextual framework goes far beyond the simple prompt context fed into commercial LLMs—it represents encoded corporate domain knowledge that can be scaled, operationalized, and eventually monetized as a standalone capital asset.

Along the Z-axis, organizations increase buoyancy and reduce friction by reassessing their technology stacks and eliminating unnecessary external vendor tools. Wang acknowledges that while third-party tools are vital, onboarding new software incurs a heavy "vendor tax" through security audits, procurement paperwork, and integration overhead. With generative AI, internal teams can frequently build custom, tailored internal modules on top of existing infrastructure, trading vendor subscription costs and administrative delay for owned, internal tools.

- **Y-Axis (Context Capital)**: Digitizing and structuring scattered institutional knowledge into an operational context database that can be monetized. - **Z-Axis (Buoyancy & Friction)**: Reducing external "vendor tax" (procurement, IT audits, licensing) by building targeted internal AI tools. - **Strategic conclusion**: "Time saved is really a vanity AI ROI... use that as a pivot point. You have to earn the saving first, and then spend it to create something new to translate that into the hard return on investment."

## Notable quotes > [02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) All models are wrong, but some are useful — Xufeng Wang

> [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) The deepest savings come from the recognition of the growth and diversification of talent enabled by AI—it's not only coming from installing tools, and it definitely is not coming from headcount reduction. — Xufeng Wang

> [15:10](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=910s) Time saved is really a vanity AI ROI... use that as a pivot point. You have to earn the saving first, and then spend it to create something new to translate that into the hard return on investment. — Xufeng Wang

## Mentioned Ai4, Xufeng Wang, Ipsos, ChatGPT, Claude, Gemini, McKinsey, BCG

## Video description Xufeng Wang - Partner & Chief Data Scientist, Ipsos

What if the most valuable returns from AI are the ones your ROI model cannot see? Across analytics, data science, and data management, AI is creating new operating dynamics—changing how teams collaborate, how decisions get made, how products evolve, and how scarce talent is leveraged—yet much of that value remains hidden behind conventional productivity metrics. Grounded in lessons from leading data and analytics organizations, this session will challenge leaders to rethink how they define and measure AI value, and offer a practical framework for connecting these “phantom” returns to financial outcomes, organizational capacity, and sustained business impact.

[Source](https://www.youtube.com/watch?v=7s-uFCb5PPw)

中文简报

益普索首席数据科学家警告：AI节省时间恐沦为虚荣ROI 企业的AI投资难以产生硬性财务回报，除非领导者主动将员工节省下来的产能重新投入到相邻服务、专有上下文资本以及降低对供应商的依赖上。 Xufeng Wang认为，将“节省时间”作为衡量AI的主要指标是一种虚荣做法，不仅无法反映真实的财务ROI，还往往会导致组织瓶颈和理解浅层化等隐性人力负债。为了捕获AI的真正价值，企业必须主动管理并重新投资所节省的时间，而不是假定其会自动转化为利润。Wang通过一个3D框架阐述了团队应如何重新利用这笔时间预算：拓展至相邻服务市场、构建专有上下文资本（Context Capital），以及利用内部AI工具降低供应商开销。益普索（Ipsos）首席数据科学家王旭峰指出，“节省时间”虽然是评估 AI 的普遍基准，但本质上是一项虚荣指标；员工虽然感知到任务完成速度提升，但往往同样疲惫，且节约的时间难以直接转化为财务报表上的硬性 ROI。 (2:14) AI 带来了填补个体技能短板的“隐形回报”，例如帮助工程师改善沟通、协助非技术人员构建软件，但也导致数据科学家各自搭建未标准化的自主工作流，并因高管沉迷快速原型开发而产生技术债务与职责模糊。 (4:43) 过度依赖 AI 总结会催生“虚假理解”，即团队互传 AI 生成的摘要而无人真正掌握核心内容；同时，生产效率的剧增将组织瓶颈从执行端直接转移到缓慢的 IT 与运营决策审批环节。 (8:53) 企业需将节省的时间作为预算再投资到三维“游泳框架”中：在 X 轴向麦肯锡式高阶咨询等相邻赛道扩张，在 Y 轴将散落的组织经验转化为可复用的“语境资本”（Context Capital），在 Z 轴通过内部自建 AI 工具消除外部采购的“供应商税”。 (11:11)

Reading summary

Xufeng Wang, Partner and Chief Data Scientist at Ipsos, argues that enterprise AI metrics centered on "time saved" are soft vanity figures that fail to translate directly into accounting balance sheets. He asserts that saved time is merely an operational budget that organizations must actively capture and reinvest into structural growth, institutional memory, and leaner tech stacks. The Vanity of Time Saved and the Fatigue Paradox (2:14) Across enterprises, "time saved" serves as the default anchor metric because it offers an accessible before-and-after baseline for individual tasks. However, Wang stresses that this perception rarely reflects hard financial realities. Despite reporting high velocity from generative tools, employees frequently finish their workdays fully exhausted, illustrating a paradox where perceived acceleration does not yield genuine surplus capacity. Converting soft efficiency into corporate profit requires scrutinizing how time was actually freed and whether that time was captured. Without deliberate financial translation, saved hours dissipate across ordinary routines rather than appearing on balance sheets. Talent Diversification, Phantom Comprehension, and Decision Bottlenecks (4:43) While AI helps bridge individual weaknesses—enabling technical engineers to communicate clearly and non-coders to assemble software—it introduces hidden organizational debt. Individual practitioners often manage disparate, unstandardized agentic workflows across models like Claude and Gemini, complicating organizational governance. Furthermore, rapid code prototyping increasingly replaces written specifications, luring senior leaders into product tinkering and accumulating technical debt. Wang also warns against "phantom understanding," where staff exchange AI-generated summaries without grasping the underlying substance. Concurrently, accelerated technical iteration shifts organizational friction upward: projects stall not at execution, but while awaiting approvals from corporate, IT, and legal decision-makers. The X-Axis: Adjacent Markets and Proprietary Core Assets (11:11) To convert the temporal budget into balance-sheet returns, Wang introduces a 3D framework. Along the horizontal X-axis, enterprises expand into adjacent market lanes by leveraging their established core competencies. For an enterprise like Ipsos—with 20,000 global employees and 17 service lines—generative AI allows smaller teams of augmented specialists to pursue complex consulting engagements historically dominated by top-tier management consultancies like McKinsey or BCG. This expansion relies on authentic proprietary assets. In an environment saturated with synthetic content, Wang notes that Ipsos's 50 years of organically gathered survey data increases in relative value, providing a defensible foundation for low-cost, fail-fast business experimentation. Context Capital and Eliminating the External Vendor Tax (15:10) Along the framework's forward (Y) and vertical (Z) axes, organizations must reshape internal infrastructure. The Y-axis focuses on capturing "Context Capital": systematically encoding scattered institutional memory, veteran workflows, and domain knowledge into an operational contextual database. Wang distinguishes this proprietary asset from basic prompt engineering, positioning it as an internal capital reserve that can be scaled and monetized. The Z-axis targets organizational friction by mitigating the "vendor tax." While commercial software has value, onboarding third-party tools generates administrative delays, security audits, and recurring license fees. Wang recommends utilizing internal data science capabilities to construct targeted, owned generative modules on top of existing architecture rather than continually purchasing outside software. Wang concludes that meaningful AI ROI does not originate from headcount reductions or basic software rollouts, but from deliberately spending earned time savings to create new capabilities. However, realizing these returns hinges on an organization's willingness to dismantle internal decision bottlenecks and systematically structure its institutional knowledge.

精炼阅读摘要

益普索（Ipsos）合伙人兼首席数据科学家Xufeng Wang指出，企业在评估人工智能投资时普遍依赖的“节省时间”本质上属于虚荣指标，并未直接转化为财务利润。他认为AI的深层价值并非来自削减员工编制，而是源于人才赋能形成的“幽灵收益”，企业必须主动将省下的时间预算重新配置到核心业务拓展、语境资本沉淀与内部工具研发中。时间锚点的局限与效率疲劳悖论 (2:14) 在跨国市场调研巨头Ipsos（全球约20,000名员工，拥有50年调研数据积累）的运营实践中，利益相关者对AI回报的审视视角差异极大。Wang指出，全行业普遍将“节省时间”作为评估AI成效的初始锚点，尽管这只是停留在日常闲聊中的软性“虚荣指标”，且极难被转化为财务报表上硬性的财务收益，管理层往往难以向财务部门证明这些省下的时间究竟产生了怎样的实际利润。更值得关注的是伴随AI应用出现的“疲劳悖论”。许多员工在日常工作中使用大语言模型后，主观上感觉处理任务的速度大幅提升，但在一天工作结束时依然感到精疲力竭。这种现象表明，主观感知到的执行提速并不等同于企业可用产能的真实增长。企业需要核实时间节省究竟是心理认知还是实际释放，并建立机制捕获富余产能，否则效率提升将无法兑现为切实的投资回报率。幽灵收益、技能填补与原型负债 (4:43) Wang将AI所带来的深层、非传统效率定义为“幽灵收益”，其核心在于组织内部的人力赋能而非单纯的技术流程自动化。AI并未均质化地改变岗位，而是差异化地补足个人短板：不善沟通的技术工程师得以借助大模型清晰表达观点，缺乏编程基础的业务人员则能利用生成式工具开发软件原型。与此同时，资深数据科学家借助Claude和Gemini等模型逐渐成为管理自主AI数字员工的“管理者”，但因个人调用逻辑各异，组织对流程可观测性的需求急剧上升。然而，这种赋能也带来了隐性的组织与技术成本。团队日益依赖“以原型代沟通”，即工程师用几分钟生成的代码原型替代了传统的长篇邮件和项目幻灯片。Wang指出这已演变成一把双刃剑：高级管理层容易沉迷于亲自搭建原型，而非将其交由工程团队规范开发，不仅侵蚀了原本清晰的分工界限，还造成了技术负债的大量累积。他强调，最深层的AI收益来自人才的成长与多元化赋能，而非依赖外部工具堆叠或削减员工规模。假性理解危机与决策瓶颈的转移 (8:53) 在代码与文档快速生成的表象之下，企业正面临“假性理解”的潜在风险。Wang警告称，过度依赖AI可能导致系统持续运转而人类却丧失理解能力的困局。职场中已经出现典型场景：员工将冗长的AI生成文本输入ChatGPT提炼要点，并根据摘要让AI生成回复；接收方同样将其交给AI处理，双方在没有一个真实人类阅读或掌握核心内容的情况下高效完成互动，不仅削弱了对底层业务的掌控，还形成了对模型准确性的盲目依赖。与此同时，AI提速使得整个企业的运营瓶颈发生了结构性转移。以往项目推进受制于前端开发与方案执行速度，如今团队能够在数分钟内完成构思与原型迭代，导致交付阻碍迅速向上游转移，卡在等待IT部门、运营团队或高管签批的环节。由于前端生产节奏成倍加快，停滞于决策环节的时间机会成本比以往更为高昂。Wang强调，节省下来的时间不是既得成果，而是一笔必须平衡技术速度与组织债务的“时间预算”。三维框架：语境资本与去供应商税 (11:11) 为了将软性的“时间预算”转化为财务报表上的硬性回报，Wang提出了以X、Y、Z三维空间为隐喻的“游泳框架”，要求管理者主动出击而非被动等待效率自发变现。在X轴（横向泳道）上，企业应依托扎实的核心业务拓展新收入流。以Ipsos为例，其依托17条核心业务线与50年积累的真实人类受访数据（在合成数据泛滥的时代，此类有机资产价值持续攀升），通过AI层放大专业团队产能，派出精干专家团队直接跨界竞争McKinsey或BCG等顶级管理咨询机构的市场，并以更低成本快速试错。在Y轴（行进方向）与Z轴（浮力与阻力）上，重构运营底层同样关键。Y轴要求企业将散落在资深员工头脑中的业务逻辑、操作规范与机构记忆，系统性沉淀为“语境资本”数据库。这种高度编码的企业专属领域知识远超通用LLM的简单提示词上下文，能直接转化为可规模化运作的核心资产。而在Z轴上，企业应控制外部“供应商税”，即引入第三方工具伴随的安全审计、采购流程与许可开支。内部团队可基于现有基础设施针对性构建定制化AI工具，减少摩擦并降低对外部供应商的依赖。 Wang总结指出，“节省时间”仅是AI投资回报评估中的一项虚荣指标与战略支点。企业必须认识到，时间节省必须先通过克服组织摩擦与隐性债务来“挣得”，随后必须被果断“花费”于开拓新型业务与沉淀核心资产中，否则效率红利将沦为虚幻的数字泡沫。如何在加速创新的同时打破管理层审批瓶颈并沉淀全公司级的语境资本，是组织能否兑现真实ROI的关键挑战。

Detailed notes — zh

## Background 本视频由 Ai4 于 2026 年 8 月 19 日发布在 YouTube 平台上，内容为益普索（Ipsos）合伙人兼首席数据科学家王旭峰（Xufeng Wang）的主题演讲。根据页面简介，演讲旨在探讨 AI 在分析、数据科学和数据管理领域创造的新运作动态，并挑战传统 ROI 模型，提出将这些隐藏的“隐性回报（Phantom ROI）”转化为财务成果与业务影响力的实用框架。 ## TL;DR 演讲围绕企业 AI 投资中的“隐性回报（Phantom ROI）”展开，指出简单的“节省时间”往往只是虚荣指标，需将其视作“时间预算”重新投入到业务中。演讲分析了 AI 给团队技能分化、虚拟团队管理及原型沟通带来的隐性变化与风险，并揭示了组织债、盲目执行及决策瓶颈这三大新型债务。最后，演讲者提出了 X、Y、Z 三维转化模型，指导企业如何将节省的时间转化为硬性财务回报。 ## Atmosphere & style 整场演讲氛围幽默且务实，开场以周四午餐时间 Token 额度用光等自嘲笑话轻松破冰，迅速拉近与现场听众的距离。演讲者的表达风格清晰有力、条理分明，将深厚的工业界实战经验与形象的比喻（如三维游泳模型）巧妙结合。整体语调兼具洞察力与前瞻性，既直击企业应用 AI 时的实际痛点，又给听众带来充满启发与实操性的启发感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=4s) 益普索（Ipsos）基于外部客户咨询与内部 50 年调查数据浓缩成“数据湖”的双重视角，探讨 AI 投资带来的隐性回报（Phantom ROI）。 - [02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) 行业普遍将“节省时间”作为评估 AI 价值的基准，但这可能只是一种虚荣指标（Vanity Metric），关键在于如何将其转化为财务部门认可的硬 ROI。 - [02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) 评估“节省时间”需要厘清两个层面：一是是否真正缩短了工时，二是生成的“时间预算”如何转化为实打实的财务收益。 - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) AI 为员工补足技能短板的方向高度多样化，技术人员借其提升沟通，非技术人员借其搭建产品，管理者需深入观察个体的技能补足路径。 - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) 随着数据科学家利用 AI 自建自动化与工作流，形成了“全员管理者”现象，极需建立可观测性（Observability）来规范管理逻辑。 - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) “以原型代沟通”虽然提高了沟通效率，但过度沉迷于完善原型会导致技术债堆积，模糊团队职责分工。 - [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) 最深层的节省来自于识别 AI 赋能下人才的成长与多元化，而非安装了什么工具，更绝对不是来自于裁员（Headcount reduction）。 - [08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) 除了生成低质代码（AI Slop）的技术债外，组织层面还出现了组织与组织债务、盲目依赖 AI 导致无人理解底层逻辑的“盲目执行”债务。 - [08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) AI 加快了原型搭建与概念迭代，但卡在 IT、运营或管理层的决策环节形成了“决策债（Decision-making Debt）”，延迟决策代价昂贵。 - [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) 通过 X 轴（邻道拓展）将时间预算转化为硬 ROI：借助 AI 扩充服务边界以跨界与传统管理咨询公司竞争，并利用 AI 极低成本进行 PoC 试错。 - [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) 在 AI 大模型时代，企业长期积累的“有机人类声音”和高信任度一手数据资产战略价值大幅提升。 - [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) 通过 Y 轴（前进方向）构建“上下文资本（Context Capital）”：将散落的企业隐性知识与 SOP 编码为业务上下文数据库，打造可变现的知识资产。 - [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) 通过 Z 轴（精简与降低摩擦）：评估供应商税（Vendor Tax），借助 AI 利用内部工具自行改造，替代部分外部工具以降低运营成本。 - [17:52](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=1072s) 企业领导者必须采取积极进取的策略，将赚到的“时间预算”重新投入到新业务、上下文资本和流程精简中，才能真正实现 Hard ROI。 ## Detailed walkthrough 以下是根据视频内容整理的详细中文逐段记录与分析： ### [00:04](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=4s) 演讲背景与益普索（Ipsos）的业务视角演讲者王旭峰（Xufeng Wang，益普索合伙人兼首席数据科学家）在拉斯维加斯举办的 Ai4 大会上开始了演讲。他打趣道，在周四午餐时间能看到这么多观众令他惊喜——因为他来之前和朋友吃饭时还在抱怨这个糟糕的时间安排，而朋友则戏谑说“很多人到了周四 Token 额度都用光了，无事可做才会来听你的演讲”。抛开玩笑不谈，他指出当今的数据科学、自动化和数据挖掘方式已发生剧变，本次演讲旨在探讨他在益普索（Ipsos）观察到的“AI 投资隐性回报（Phantom ROI）”。为理解后续观点，王旭峰特别强调了益普索的背景与业务环境： * **核心业务**：益普索是一家全球拥有约 20,000 名员工的知名市场调研公司，传统问卷调查（Survey）是其核心业务。 * **团队构成**：他管理的部门包含两大团队——一是面向全球客户、处理传统数据科学与机器学习的咨询团队；二是负责数据管理的团队，将益普索 50 年积累的调查数据浓缩并提炼成单一的“数据湖（Data Lake）”，形成独一无二的数据资产。 * **双重視角**：演讲所分享的经验既涵盖外部客户咨询与服务，也包含内部数据管理与运营。 --- ### [02:14](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=134s) “节省时间”：被误解的虚荣指标与投资起点在企业层面评估 AI 价值时，不同角色对 AI 效果的认知截然不同。王旭峰提出，行业普遍将“节省时间”作为评估 AI 价值的基准点，大家常在茶水间交流或自我暗示“使用 ChatGPT 又省下了多少时间”，但唯独不敢在财务部门面前这么说，因为没人能向财务解释“节省的时间”如何转化为账面利润。尽管“节省时间”可能只是个虚荣指标（Vanity Metric）甚至是一种错觉，但它仍不失为一个有价值的锚定起点。评估“节省时间”需要分为两步逻辑： 1. **如何真正实现节省**：是确实缩短了工时，还是仅仅产生了干完活的错觉（例如一天下来使用了大模型却依然感到精疲力竭）？ 2. **如何预算并转化时间**：节省下来的时间即为“时间预算（Time Budget）”，如何将其转化为财务部门能够在 Excel 表格中看懂的“硬性 ROI（Hard ROI）”，是连接软性回报与硬性回报的关键。 --- ### [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) 人才赋能的分化与管理新挑战在探讨 AI 对团队和人才带来的隐性影响时，王旭峰总结了企业在人才技能、团队管理和原型沟通三个维度的变化与潜藏风险： * **人才技能赋能的多样化**：AI 帮助员工填补了各自的技能短板，但发展方向极其多样。例如，不擅长沟通的工程师开始借助 AI 更顺畅地撰写沟通邮件，而技术背景较弱的非技术人员则利用 AI 满足自己搭建产品的欲望。管理者必须深入观察个体，了解每个人究竟用 AI 补足了什么技能。 * **“全员管理者”现象**：在资深数据科学家利用 Gemini 或 Cloud 等工具自建工作流和自动化时，每个人其实都拥有一支“虚拟 AI 团队”。但由于缺乏正规的管理经验，不同人管理 AI 团队的逻辑截然不同，因此企业极需建立“可观测性（Observability）”。 * **“以原型代沟通”的双刃剑**：以往沟通想法需要写邮件、做 PPT，现在工程师倾向于花费几分钟直接用 AI 快速搭建一个产品原型（Prototype）。然而，资深技术领袖必须明白“何时止步”，如果自己沉迷于不断完善原型、主导产品方向，反而会导致技术债的堆积，模糊团队职责分工。 > **核心观点**：“最深层的节省来自于识别 AI 赋能下人才的成长与多元化，而非仅仅来自于安装了什么工具，更绝对不是来自于裁员（Headcount reduction）。” --- ### [08:53](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=533s) AI 带来的三大新形式债务除了通过 AI 生成低质代码（AI Slop）所带来的技术债外，组织层面还出现了三种极易被忽略的新型债务： * **组织与组织债务**：技术工具的普及掩盖了团队内部的沟通与协同问题，忽视人类员工的心理与组织动态会导致组织债加剧。 * **过度依赖导致盲目执行**：在市场调研和分析团队中，常出现“AI 懂了，但人没懂”的怪现象。比如一位工程师用 AI 生成了一长段文本发给同事，同事又直接丢进 ChatGPT 总结并回传——表面上看项目推进极快，但实际无人理解底层逻辑，整个团队沦为 AI 准确率的奴隶。 * **决策变成新瓶颈**：借助 AI 搭建原型、规划方案和迭代概念变得空前高效，但流程往往卡在 IT 部门、运营部门或管理层的决策环节。在当前节奏下，延迟决策或忽视决策带来的代价比一两年前要高昂得多，形成了严重的“决策债（Decision-making Debt）”。 --- ### [11:11](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=671s) 如何将“时间预算”转化为财务硬 ROI（三维游泳模型）既然节省下来的时间只是一笔“时间预算”，企业该如何将其转化为实打实的硬性财务回报？王旭峰借用“游泳”提出了 X、Y、Z 三维转化模型： ``` [Y轴：前进方向] 重构业务流程与上下文资本 ^ | | [X轴：邻道] <---+---> [Z轴：上下浮力] 围绕核心拓展精简工具与降低摩擦新业务与服务 ``` * **X 轴（邻道拓展：围绕坚实核心打造新业务）**： * **扩充服务边界**：以益普索为例，调研是其 50 年立足的坚实核心。借助 AI 的赋能，少数专家即可处理更多工作，这使得益普索有能力跨界与麦肯锡（McKinsey）、波士顿咨询（BCG）等顶级纯管理咨询公司竞争。 * **极低成本快速试错**：利用 AI 迅速进行概念验证（PoC），提倡“快速行动，快速失败”。 * **重估数据资产价值**：在 AI 大模型时代，企业长期积累的“有机人类声音（Organic Human Voice）”和高信任度一手数据，其战略与商业价值正在大幅提升。 * **Y 轴（前进方向：重构流程与构建“上下文资本”）**： * **上下文资本（Context Capital）**：企业内部散落着大量隐性知识（如老员工 10-20 年积累的标准作业程序 SOP）。企业应制定系统化策略，将其提炼并编码为“业务上下文数据库（Operational Contextual Layer）”。 * **知识变现**：这层上下文数据库是企业专属的知识资产，越早建立管理与增长策略，就越早能够将其商业化变现。 * **Z 轴（浮力与阻力：精简外部工具与降低摩擦）**： * **评估“供应商税（Vendor Tax）”**：引入外部供应商工具往往伴随着繁琐的 IT 审计、合同流程与入职成本。在适当场景下，企业可以借助 AI 利用内部现有工具自行改造开发，替代部分外部工具，从而减少摩擦、降低成本。 --- ### [17:52](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=1072s) 总结与建议王旭峰在结尾总结道，“节省时间”从某种意义上说是 AI 回报的一种虚荣指标，但它易于衡量，因此可以作为撬动变化的杠杆。企业领导者必须采取积极进取的策略（Play Aggressively），不能停留在“宣布省下了时间”的中途阶段，而必须将赚到（Earn）的时间重新投入到新业务、上下文资本和流程精简中，最终将其彻底转化为财务账面上硬生生的投资回报（Hard ROI）。最后，他祝愿大家能够明智地花掉自己的“时间预算”。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=4s) 很多人到了周四 Token 额度都用光了，无事可做才会来听你的演讲 — 王旭峰的朋友 > [04:43](https://www.youtube.com/watch?v=7s-uFCb5PPw&t=283s) 最深层的节省来自于识别 AI 赋能下人才的成长与多元化，而非仅仅来自于安装了什么工具，更绝对不是来自于裁员（Headcount reduction）。 — 王旭峰 ## Mentioned 王旭峰 (Xufeng Wang), 益普索 (Ipsos), Ai4, 拉斯维加斯 (Las Vegas), ChatGPT, Gemini, Claude, 麦肯锡 (McKinsey), 波士顿咨询 (BCG) ## Video description Xufeng Wang - Partner & Chief Data Scientist, Ipsos What if the most valuable returns from AI are the ones your ROI model cannot see? Across analytics, data science, and data management, AI is creating new operating dynamics—changing how teams collaborate, how decisions get made, how products evolve, and how scarce talent is leveraged—yet much of that value remains hidden behind conventional productivity metrics. Grounded in lessons from leading data and analytics organizations, this session will challenge leaders to rethink how they define and measure AI value, and offer a practical framework for connecting these “phantom” returns to financial outcomes, organizational capacity, and sustained business impact. [Source](https://www.youtube.com/watch?v=7s-uFCb5PPw)

Source quotes

[0:04] 王旭峰的朋友: 很多人到了周四 Token 额度都用光了，无事可做才会来听你的演讲 [4:43] 王旭峰: 最深层的节省来自于识别 AI 赋能下人才的成长与多元化，而非仅仅来自于安装了什么工具，更绝对不是来自于裁员（Headcount reduction）。

Source quotes

[2:14] Xufeng Wang: All models are wrong, but some are useful [4:43] Xufeng Wang: The deepest savings come from the recognition of the growth and diversification of talent enabled by AI—it's not only coming from installing tools, and it definitely is not coming from headcount reduction. [15:10] Xufeng Wang: Time saved is really a vanity AI ROI... use that as a pivot point. You have to earn the saving first, and then spend it to create something new to translate that into the hard return on investment.

Original transcript

[0:04] So, you know, before I jump into it, this is interesting. Like I said, it's more people than I [0:09] saw. Before I come to Vegas, um I had a lunch with a friend of mine. So, I complained to him a little [0:18] bit. Say, "Hey, I have a talk Thursday, the last day of the conference. You know, during lunchtime, [0:25] I don't think there's going to be too many people showing up." But this really surprised me. Um what [0:31] he told me after is hey look on bright side there will be a lot of people who spend their token [0:36] limit on Thursday has nothing to do and come to your session you know so good you know who you are [0:44] uh you don't have to raise your hand um but all jokes aside a lot of things are different today [0:50] isn't it right the way we work um the way we automate things the way we do our data science [0:56] stuff the the way we extract insight a lot of things had changed Now today I want to share with [1:03] you some of the uh learnings we learned along the way at IPSOS um in what I call the phantom return [1:11] of investment. Now just like when you're using a AI agent um the context matters. So Ipsos is [1:21] a market research company right survey is our core business and in my organization there are [1:28] two different teams there one is the traditional data science machine learning teams that's doing [1:34] client work right so we have different clients across the globe we got about 20,000 people uh [1:41] in the entire company lots of clients globally and we also have a second team that's working on data [1:49] management right think about 50 years of IPSO survey we concentrate that into one single data [1:56] leak leverage that and that's a unique data asset so the perspective I'm going to share today keep [2:03] that in mind right so that's the context that's my point of view both from a consulting client facing [2:10] point of view and also from an internal point of view but at the end of the day it's what I [2:14] see right now That is a curse in a way when you look at things at enterprise level it's a curse [2:23] because there are a lot of things you can apply AI to AI give you different effects and you know [2:30] different people sees the value of AI differently. So we need a starting point. We need a pivot, [2:38] right? So we need a point something you know we can roughly agree on to get started. And [2:44] you very often hear people say we are saving so much time with AI, right? So you hear that [2:52] in the cooler talks. You sometimes look in the mirror you tell yourself hey great by creatively [2:57] using chat GBT I save a lot of time today. You talk to your friends you talk to colleagues. [3:02] Perhaps the only place you don't talk about that is in front the doors with finance, right? Because [3:09] you just can't explain. Okay, saving so much time. But how does that actually translate into the [3:14] finance exhaustion? That's okay. That's something we're going to look at today. So it's a perception [3:21] very often. It's a perception. Right or wrong? Is that the best way to look at it? You know, [3:26] all models are wrong. Some are useful. At least this gave us a starting point. [3:33] If you anchor on the time saved even it's a vanity metric even it's a perception you have an anchor [3:40] point and then you look at the before and after before is how do you actually get time saved [3:47] right so there are two questions at play you know how is time saved and is it really saved or is it [3:54] just a simple perception right you definitely have days sometimes especially early on you know when [4:01] LM become available. Oh gosh, this is saving me so much time. I'm getting things done so quickly. But [4:06] then you still feel exhausted right throughout the day. Is your time really saved? Right? So [4:12] that requires a closer examination. Now after you have actually achieved the time saved, you have to [4:19] ask the next question. How do you actually save that spended time? That's your budget. And that [4:25] is the key step connecting the soft RPI uh ROI with the hard ROI. Finance understands translate [4:34] that into the Excel sheet. That's always the key. And we'll take a look at both steps today. [4:43] Okay, we're not going to talk too much today about the well-known aspect of saving time, right? So [4:52] the efficiency gain or the quality increase or all of these things right. So through v coding [4:57] through the adaption of automation through various different ways agentic systems right [5:02] so from technical side you know the drill right so you know there are multiple ways to get time [5:07] save that's one thing we observe at the enterprise level. Now the second part in my opinion is let's [5:15] talk about the phantom saving a human right at the end of the day we are all humans and this [5:24] is where you can actually pay a little bit extra attention to observe your team to observe your [5:28] organization and from what I see there are three things I think often is not talked about enough [5:36] in my opinion the first thing is AI actually enables your team enables your talent on multiple [5:44] different factors. But the downside is they're all enabled in a very diverse direction. Now, [5:53] one good example is engineers are not very often good communicators. They want to communicate, [6:00] but sometimes it's just very inefficient through the use of AI. They find the communication to [6:06] be easier. You'll have team members like that, but that's not every single engineers. You have [6:12] nontechn people trying to use the AI to build some technical work because they're interested. [6:17] They want to be builders. So what we observed so far is the enablement of talent is real but [6:25] different people use AI to fill different kinds of skill gaps and that is happening across your [6:31] entire organization. Every single person who's using AI is doing that and you have to be mindful [6:38] and observe individually to understand what kind of talent skill gap is being filled. The second [6:45] thing is everyone has an AI team now right but not everybody has that management experience. [6:54] So in my team for example they're lead data scientists senior data scientists already [6:59] using cloud already using Gemini to orchestrate their own workflow automate part of their work [7:06] but when you actually look at the logic the way they automate that it's actually very different [7:14] it's as if they have their own team but different people manage their team very differently so I [7:20] recommend to have some observability right because everybody practically has become a manager. Now, [7:28] now the last thing what we seeing is can be effective but it's a double-edged sword again [7:34] is communication via pro prototype but know when to stop and where to stop. Before when you are [7:44] trying to communicate an idea you probably have to write an email, you have to do a presentation. [7:48] It's very time consuming and very often engineers are just they just don't want to do that. that's [7:53] not their strong suit. So instead they spend a few minutes build out the prototype. You've seen that [7:58] happens. That happens to your team. But you have to know where to stop right. So very often your [8:05] role especially you're a lead data engineer uh or data scientist you have to stop building keep [8:11] building the product yourself and give that to the team. That divisional responsibility [8:17] has to be very clear. And this is where we often see the senior folks, the lead folks, [8:22] they begin to build the product themselves and dictate the direction of where the product is [8:28] going, which is very often result in additional debt. But the key point here is the deepest [8:35] saving come from the recognition of the growth and diversification of talent enabled by AI. [8:42] It's not only coming from installing tools and it definitely is not coming from headcount reduction. [8:53] Now the other side of the coin is a new kind of debt. Right? So the three examples [9:01] I gave in the previous slide, they all have the downside, right? Color, red color. That's the new [9:06] kind of debt. It's not only the technical debt you're accumulating through AI slop. [9:12] But also there's a human factor there. Always remember we're all humans. Your team are human and [9:18] there's a new kind of debt with the organization aspect. The second thing which more often happens [9:26] in insights team is the over reliance on AI for everything. Now AI understands it but do you [9:35] understand it? Right? You probably seen funny situations like one engineer send the other [9:41] engineer a long AI generated text and the other one feed that into chat GBT and send whatever [9:48] comes out the back. That's not understanding. So your team may seem on the surface moving at [9:55] lightning speed but no one actually understands what's going on. You essentially become subjects [10:02] to the AI at the mercy of the accuracy, right? So AI understands it. Make sure your team also [10:08] understand. Now the third thing is increasingly from the organization point of view we're seeing [10:14] decision making becoming the new bottleneck. Building a product is much faster. Building a [10:22] prototype is even faster. You want to iterate on idea, you want to iterate on proof of concept, [10:27] you want to build something, you want to plan ahead before you actually build something, right? [10:32] But very often it stops at decision- making. decision making from it, decision making from ops, [10:38] decision making from you. So every moment you wait for that decision making, every moment you [10:45] ignored in making that decision making compound, it becomes more expensive today than let's say one [10:52] or two years ago. So be mindful of the decision making um um debt, right? So it is one of the [11:00] new bottlenecks. Now time saved is not a return. It's a budget by considering the technical human [11:11] factor. Okay, let's say you're successful in adapting this in your organization. What does [11:17] this mean? It doesn't stop here. You increase your uh vanity metric which is time saved. What [11:22] does that mean? You still need to translate that into the hard return of investment. So treat this [11:28] as a budget. Right? So the next question is how do we actually spend it? How do you actually turn [11:37] the time into tangible hard return of investment? The way I look at it from our experience is you [11:46] treat this as a swimming lesson, right? So and you think about three dimensions. The direction [11:52] you're going, your neighboring swimming lanes and your um you know up and down, [11:59] right? So the drag the buoyancy right so in the x direction is your neighboring line you have to [12:06] ask yourself can I actually swim now enable me to swim in the neighboring line in the y direction [12:13] is the direction you're going you ask yourself the question can I reimagine the way I swim is [12:20] there new way does AI enable me to do new things there and on the z-axis you have to ask yourself [12:27] is there any opportunities to keep myself lean to give me more lift decrease the drag so I'll look [12:35] into each one of them in the neighboring direction the x direction in my opinion is new rot around a [12:45] solid core right think about our business IPSOS survey is our core right so that's we have 17 [12:54] service lines doing all kinds of surveys passive quantitative qualitative in the US for example [12:59] and in all other countries is full service survey that is our core but AI does enable us to build on [13:09] top of that core and there are two ways that's happening one is it actually become abstract [13:16] layer around our core offering make our core offering more adaptable with the client with other [13:22] solutions one good example would be traditionally we would not actually compete against um you know [13:31] McKenzie or BCG these uh uh pure consulting companies right but with AI actually lifting [13:39] us up making us more efficient um with fewer experts we're able to do more work we actually [13:46] suddenly have the capacity to compete to compete in addition to carrying our core competency which [13:54] is in serving in market research So these new enablement you have to be very mindful of and [14:03] the second thing is you can actually begin to experiment prove a concept very cheaply you [14:10] think there's a new direction you think there's a new opportunity go experimental with it right so [14:15] move fast and fail fast AI enables the team to do that then among all of these right so go back to [14:24] the solid core And this is not unique to Ipsos is not unique to a market research company. We have [14:31] 50 years of survey data and that data in the era of AI actually what we observe is is increasingly [14:40] become more valuable is organic human voice. It is organically collected trusted data. Right? So [14:50] if you have any data asset you own now may worth a reassessment the value of that is increasing with [15:00] the advent of AI and you can potentially enable new offerings around it as well. [15:10] Now in the Y direction, the direction you are swimming, what we seeing is the operations [15:19] needs to be rebuilt around the new asset in what we call the context capital. This is the [15:27] very invisible [clears throat] asset, invisible capital you have. So think about it. [snorts] You [15:33] got different teams operating on different things manually. You have certain standard [15:38] operation procedures. Maybe you have an employee there 20 years, 10 years, knows how to do things, [15:44] right? A lot of these knowledge are scattered around your organization. Now it's time to have [15:49] a systematic plan to put them together to become the contextual layer. This contextual layer isn't [15:56] really just limiting to the coding contextual layer you feed into cloud and JBT. This is the [16:02] operational contextual layer. How do you do business encoded and managed into a context [16:10] database? This becomes the capital and you can even take it one step further monetizing. Right? [16:18] This is the knowledge that your corporation owns and maintains and grows over time. So the [16:26] sooner you have a strategy to manage your contacts capital, the sooner you can begin to monetize it. [16:35] Now in the Z direction and this is just one of the many examples you can [16:40] uh decrease the friction inc increase your um buoyancy in our organization we find sometimes [16:50] actually you can eliminate the external toolings and is needed right now before my vendor friends [16:56] come here and argue with me if that's a good idea I just want to make it clear this is very [17:01] situation we don't want to get rid of all the vendors of course um but it's worth to think [17:07] about reassess all the tools we're currently using there's the vendor tax right so when [17:13] you're on boarding a new vendor on boarding a new technology there's always the effort of [17:18] you know onboarding the paperwork the IT audit and all of that versus maybe you already have [17:25] some internal tools that you can just utilize AI to you know repurpose it build it yourself [17:32] right so Having a really custommade module for yourself is always a trade-off. There's a time [17:38] saved through the vendor pass and there's time spent and time saved potentially also through [17:44] your internal dividend. So all of these things need to be assessed individually. [17:52] So overall that's it. Time saved is really a vanity AI return investment in certain ways. [18:02] But it's easy to algor. So use that as a pivot point. Use that as a lever. You have [18:08] to do two things to realize that you have to earn the saving first and then buy it [18:15] to create something new to translate that into the hard return of investment. And [18:21] when you're doing that from what we see and my advice is you want to play that aggressively. [18:27] You don't want to sit in the middle point say I achieved time save and directly translate that [18:32] into something. You can't you can't meaningfully do that. You have to play aggressively take it [18:38] all the way to translate that vanity metric all the way to the heart return of investment. So [18:44] with that I'd like to conclude my talk and I wish you go spend your time budget wisely. Thank you.