Strauss AI Chief Targets 2x Returns as 80% of Enterprise POCs Stall

80%企业POC停滞之际，Strauss AI负责人瞄准2倍回报 企业AI项目往往在投入生产前停滞，除非组织将项目根植于严格的数据工程、针对具体岗位的技能提升，以及坚决弃用任何效率提升不足两倍工具的严苛规则。 Soloducho提出了一种企业AI双轨模式，在快速、低阻力的短期见效举措与直接挂钩损益增长的基础性长期项目之间取得平衡。他借用皮克斯电影《头脑特工队》（Inside Out）作为组织框架，通过自动取款机（ATM）和放射学等历史先例来缓解员工对被淘汰的恐惧，主张开展针对具体岗位的技能提升，使AI成为工作能力的放大器而非替代品。他告诫企业不要开展效率提升低于两倍的低收益项目，并警告称，若缺乏严格的数据工程和客观评估基准，80%的企业AI概念验证都会失败。他强调当前企业AI的落地仅处于“90分钟比赛的第20分钟”，呼吁领导者培养好奇心、谦逊态度、严格的问责机制，并构建持久的内部数据架构。 Strauss Group 首席 AI 官 Ilai Soloducho 指出，企业 AI 的核心目标是直接改善损益表（P&L）表现；通过短期“速赢”与长期核心项目的双轨模型推进，能有效破解约 80% 的 POC 无法投入生产的瓶颈。 (0:04) 面对员工对 AI 取代人力的恐惧，历史上的 ATM 和放射医学案例表明自动化能扩展服务能力并带来更高价值岗位；领导层应充当情绪“控制面板”，通过定制培训为员工打造提升效能的“超人战衣”。 (4:08) 为防范 80% 的 AI 支出沦为无意义噪音，企业应建立涵盖数据就绪度与 P&L 影响的严格决策指标，保留已有成熟流程，仅针对可实现至少 2 倍效率提升或 0 到 1 突破的场景立项。 (9:35) 企业级 AI 的稳固落地高度依赖数据清洗与管道治理等底层基础设施，这占据了 80% 至 90% 的实际工程量；必须运用黄金数据集与量化指标（evals）替代主观感觉，以此建立跨部门信任。 (13:13) 企业 AI 转型目前仅处于“90 分钟赛程的第 20 分钟”，管理者需保持好奇、谦逊与严格问责，重点强化组织自身的内部数据工程与技术底座，而非全面依赖外部供应商。 (17:34)

Strauss Group Chief AI Officer Ilai Soloducho argues that successful enterprise AI adoption depends on translating technical capability into measurable profit-and-loss impact, which requires navigating organizational psychology and unglamorous data engineering rather than chasing prototype hype. Dual-Track Delivery and the POC Gap (0:04) While global estimates project that artificial intelligence holds $4.4 trillion in economic potential and 88% of organizations have initiated adoption, roughly 80% of enterprise proof-of-concepts stall before reaching production. Soloducho attributes this divide to the operational chasm separating venture-funded video demos from production-grade enterprise software, emphasizing that bridging it requires solving organizational and human challenges rather than focusing solely on algorithms. To ensure AI produces measurable profit-and-loss (P&L) impact, Soloducho outlines a dual-track delivery model. Organizations must concurrently deploy "quick wins"—low-cost tools, such as $20 monthly licenses, that deliver rapid productivity in hours or days—alongside "big stones," which are six-month-plus transformations of core manufacturing and operational workflows designed for long-term structural returns. Managing Workforce Emotion and Automation Fears (4:08) Drawing an analogy to Pixar's Inside Out, Soloducho contends that leadership must act as an emotional control panel to guide employee fear, anxiety, and nostalgia toward productive agency. To address widespread anxiety that AI will replace human labor, he highlights historical parallels. The introduction of automated teller machines (ATMs) in the late 1960s and 1970s lowered branch operating costs, leading banks to open more branches and ultimately increasing total teller employment while shifting responsibilities to advisory tasks. Similarly, the 2010 rise of deep learning in radiology lowered diagnostic costs and expanded screening volume, increasing net demand for radiologists. To replace fear with agency, Soloducho argues organizations should treat technology as a "Superman suit" that amplifies human workers rather than displacing them. This approach requires customized, role-specific upskilling tailored for individual departments—such as legal, sales, or manufacturing—combined with mandatory hands-on experimentation on daily tasks to build employee confidence and leadership trust. Filtering Use Cases Against Business Impact (9:35) Legacy organizations frequently confront nostalgia that resists changing functional processes, alongside anxiety that drives business units to chase every conceivable use case. Soloducho warns that without rigorous governance, up to 80% of enterprise AI expenditure turns into impactless spending. Consequently, he cautions against implementing AI for marginal efficiency improvements of 1.1x or 1.2x, which rarely justify organizational disruption. Instead, initiatives must target at least a 2x operational gain or zero-to-one business expansion. Prioritization decisions must be validated through empirical data collected from factory floors and business units alongside external industry benchmarks. Soloducho frames technical choices around five primary criteria: internal data readiness, speed of delivering value, delivery feasibility relative to team bandwidth, direct P&L ROI, and solution fit balancing cost, customization, and workforce usability. Engineering Foundations and Rigorous Evaluation (13:13) Unmanaged enthusiasm for AI pilots can create blind spots when teams rely on subjective impressions. Soloducho advocates replacing anecdotal reactions with quantitative evaluation frameworks ("evals") built on curated "golden datasets" of 15 to 20 manually validated prompts to measure model accuracy mathematically before greenlighting production. To anchor strategy, he identifies trust as the critical objective, requiring cross-functional governance across IT, cybersecurity, plant operations, sales, HR, finance, and legal to protect data privacy and ensure compliance. Underpinning this operational model is foundational data architecture. Recalling his experience at Palantir Technologies, Soloducho notes that 80% to 90% of enterprise AI work consists of data cleansing, pipeline construction, ingestion, and governance. Because basic prompt engineering has strict limitations, scalable autonomous agents and retrieval-augmented generation (RAG) depend fundamentally on integrating curated internal data with fine-tuned foundation models. Soloducho concludes that enterprise AI is only at "minute 20 out of 90" of its overall adoption journey. He asserts that successful long-term navigation requires continuous learning, humility toward domain experts, strict human accountability for AI outputs, and building internal data infrastructure rather than relying entirely on third-party vendors.

施特劳斯集团（Strauss Group）首席人工智能官兼首席执行官幕僚长伊莱·索洛杜乔（Ilai Soloducho）指出，企业推进人工智能的核心并非追逐技术噱头，而是通过更优决策直接改善损益表（P&L）。他强调，尽管AI展现出巨大的经济潜力，但要跨越80%概念验证夭折的行业鸿沟，必须将双轨落地机制、组织情绪治理、严苛量化评估与底层数据工程紧密结合。 双轨推进：平衡速赢产出与核心业务重构 (0:04) 索洛杜乔认为，企业AI投资必须服务于可衡量的财务回报。虽然宏观预测显示AI具备4.4万亿美元的潜在经济价值，且88%的组织已启动相关探索，但约80%的企业概念验证（POC）无法成功跨入生产部署阶段。初创公司展示的Demo视频与企业级高可靠软件之间存在巨大的运营鸿沟，跨越障碍的关键不在于算法本身，而在于组织系统、员工心理与变革管理的协同。 为兼顾业务部门的耐心与长期战略，索洛杜乔提出了双轨推进模型：一方面以每人每月约20美元的低成本工具推动“速赢”（Quick Wins），在数小时或数天内释放生产力以维系团队信心；另一方面针对制造与供应链等核心环节启动实施周期超过半年的“大石块”（Big Stones）项目，旨在重构业务流程并获取长期的结构性投资回报，双轨最终均指向损益表的切实改善。 情绪治理：以历史先例化解岗位替代恐惧 (4:08) 索洛杜乔借鉴电影《头脑特工队》指出，管理层必须充当组织的情绪“控制面板”，主动识别并疏导员工在面对技术冲击时经历的恐惧、焦虑、怀旧与兴奋。针对最为普遍的“AI将取代人类”的恐惧，他列举了两项历史先例予以反驳：一是20世纪60至70年代自动柜员机（ATM）普及后，单店运营成本大幅降低推动银行加速扩建网点，柜员总就业人数反而上升，其职能也顺利转向高附加值的财富咨询；二是2010年深度学习进入放射影像学领域后，诊断成本下降激发了更大规模的筛查需求，放射科医生的净就业人数不降反升。 他强调企业不应寻求替代人力，而应为员工配备如同增强能力的“超人战袍”。具体落地策略包括推行高度定制化的岗位培训（针对制造、法务、销售或营销分别定制专属课程，摒弃泛泛的技术宣讲），并推行强制性的上手实操，通过在真实业务流中建立掌控感来化解恐惧，转化为对技术和管理层的信任。 严格决策准则：以两倍效能收益过滤无效支出 (9:35) 传统企业在技术革新时常受困于眷恋既有流程的“怀旧”阻力，索洛杜乔主张组织必须保持谦逊，不可强行将AI套用到所有问题上。对于运转良好的流程应当予以保留，而对于效率提升仅有1.1倍或1.2倍的边际优化方案应果断放弃，因为微小的收益根本无法抵消技术替换带来的流程冲击与管理成本。企业必须将资源聚焦于能带来至少2倍效率飞跃或“0到1”全新业务增长的场景，且优先级评定必须依赖一线调研与行业基准，而非内部政治角力。 针对各部门一拥而上导致80%预算沦为“无成效消耗”的焦虑现象，他确立了五项严苛的筛选标准：数据是否清洗合规且可用（数据就绪度）、能否在短期内证明业务价值（价值兑现周期）、技术负担是否匹配内部团队负荷（交付可行性）、是否能实质性提振营收或削减结构性成本（损益表ROI），以及方案是否平衡了预算、定制化与生产力提升（方案契合度）。 量化评估与底层基建：跨越概念验证陷阱 (13:13) 当AI应用初具雏形时，团队易陷入盲目狂热。索洛杜乔警示管理层必须摒弃“感觉不错”的主观臆断，推行严格的量化评估（evals）体系。通过构建包含15至20个具备高度代表性且经人工审核的“黄金数据集”（golden datasets），团队可以在投产前客观量化模型准确率（例如区分90%达标与60%失效），杜绝带病上线。同时，他将“信任”确立为AI战略的核心北极星，强调必须由IT、信息安全、工厂管理、一线销售、HR、财务及法务共同组建跨职能治理网络，确保数据隐私与合规底线。 索洛杜乔结合其在Palantir的经历指出，管理层往往执着于精美的前端可视化分析看板，但实际项目中80%至90%的重任都在于底层数据工程，包括数据清洗、管道搭建、摄取与权限治理。他指出通用基础模型与简单的提示词工程存在明显的能力上限，要实现真正企业级的自主智能体与检索增强生成（RAG），必须依靠精细治理的企业内部专有数据与基础模型的深度结合。 领导力准则与长期竞赛：打造企业自主能力 (17:34) 索洛杜乔归纳了驾驭企业AI转型所需的五项核心领导力准则：一是终身学习与好奇心，以应对底层假设每3至6个月便面临淘汰的现实；二是谦逊态度，尊重垂直业务专家的行业洞察；三是严格问责，使用者必须对最终结果负全责，不可借口“由AI生成”而推卸对业务事实的解释责任；四是始终锚定核心业务价值；五是理性设定预期。 他借用足球比赛打比方，指出当前的企业AI转型仅仅处在“90分钟全场比赛的第20分钟”，未来还可能面临加时赛与点球决胜。面对这场长期长跑，企业应当把重心放在内化核心技术能力、构建自身数据管道与工程架构上，而非一味依赖外部承包商与第三方供应商，以扎实的底层实力迎接长期演进。 索洛杜乔为传统企业在AI时代规避概念验证失败、达成切实损益回报提供了高度结构化的实战框架。然而，其设定的两倍效能收益门槛与重度依赖数据底层工程的前提，对数字化基础较弱或资源受限的企业而言意味着极高的执行门槛。在基础模型与智能体架构快速迭代的周期中，企业如何在追求架构自主可控与借助外部标准化方案降低试错成本之间找到动态平衡，仍是各行业领导者面临的核心挑战。

## Background 该视频由频道 Ai4 于 2026 年 8 月 18 日发布在 YouTube 平台上。视频记录了施特劳斯集团（Strauss Group）首席 AI 官兼集团 CEO 幕僚长 Ilai Soloducho 的主题演讲，深入探讨了在食品与饮料等大型行业中，AI 规模化落地面临的核心瓶颈，并从组织心理学、流程重构与投资回报率（ROI）等维度分享了 AI 领导力实践框架。 ## TL;DR Ilai Soloducho 指出 AI 战略必须坚守损益表（P&L）与 ROI 导向，平衡短期速赢与长期战略重构，破解高达 80% 的 POC 无法规模化投产的困境。他借用电影《头脑特工队》的隐喻，提出通过定制化技能提升化解恐惧、设定 2X 提升标准破除怀旧盲目、以及建立严密评测体系（Evals）规避无效投入。同时，他强调数据清洗与管道治理等底层脏活累活是企业构建自主 AI 能力的核心基石，呼吁领导者以终身学习和高度担当迎接这场刚刚进行到“第 20 分钟”的长期转型。 ## Atmosphere & style 整场演讲基调务实、冷静且兼具战略高度，讲者依托在 Google X 和 Palantir 的硬核技术及落地经验，褪去了行业普遍的浮躁炒作，直切企业转型的本质痛点。借助电影角色与世界杯足球赛等生动隐喻，演讲在理性拆解复杂技术架构与数据战略的同时，赋予了组织变革深刻的人文与心理洞察。整体氛围沉稳而振奋人心，传递出一种不盲从、重实效、勇于承担落地责任的清醒领导者气场。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) AI 领导者的核心职责是将技术关注点牢牢锁定在损益表（P&L）的影响与投资回报率（ROI）上，而非盲从技术炒作。 - [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) 企业推进 AI 落地应采取双轨制：兼顾能在数天内见效的速赢项目（Quick Wins），以及耗时 6 个月以上重构生产与运营的长远大石（Big Stones）。 - [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) 全球 80% 的 AI 概念验证（POC）在规模化时失败，真正瓶颈通常不在于技术本身，而在于企业组织机制、流程与心理认知系统的滞后。 - [04:08](https://www.youtube.com/watch?v=RhnVyGMiYac&t=248s) 历史表明技术自动化往往会扩张服务规模而非单纯减少岗位（如 ATM 普及反而促使银行柜员增多并转向高价值理财顾问）。 - [04:08](https://www.youtube.com/watch?v=RhnVyGMiYac&t=248s) 破除员工对被 AI 替代的恐惧需要建立“赋能战甲”思维，通过定制化职能技能提升与动手实践培训让员工切身体验效率跃迁。 - [09:35](https://www.youtube.com/watch?v=RhnVyGMiYac&t=575s) 并非所有场景都需 AI 介入：应敬畏运作良好的旧流程，只有在能实现至少 2 倍（2X）质变提升的痛点领域才值得重塑。 - [09:35](https://www.youtube.com/watch?v=RhnVyGMiYac&t=575s) 盲目跟风和焦虑会导致高达 80% 的 AI 投入沦为无业务影响的无效开支，必须基于数据就绪度、见效时间与明确 ROI 建立准入标尺与淘汰机制。 - [13:18](https://www.youtube.com/watch?v=RhnVyGMiYac&t=798s) 防范试点阶段的主观臆断盲区，必须建立包含黄金数据集（Golden Datasets）的量化评测体系（Evals），作为进入生产环境的硬性门槛。 - [14:39](https://www.youtube.com/watch?v=RhnVyGMiYac&t=879s) 企业级 AI 战略的北极星是信任（Trust），需要 IT、安全、一线运营、财务及法务合规等跨职能部门的深度协同。 - [14:39](https://www.youtube.com/watch?v=RhnVyGMiYac&t=879s) 酷炫的可视化报表只是表象，构建 RAG、模型微调与智能体能力的 80% 至 90% 核心功夫在于底层数据管道与数据清洗治理这些“脏活累活”。 - [17:34](https://www.youtube.com/watch?v=RhnVyGMiYac&t=1054s) 组织成员必须对 AI 输出结果承担兜底责任（Accountability），坚决杜绝用 AI 自动生成汇报却对数据一无所知的现象。 - [17:34](https://www.youtube.com/watch?v=RhnVyGMiYac&t=1054s) 当前的 AI 变革马拉松才刚进行到“第 20 分钟”，企业切忌过度依赖外部外包，必须尽早锤炼并沉淀内部的 AI 核心组织肌肉与基础设施。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) 个人背景与 AI 落地战略核心：聚焦损益表（P&L） 讲者 Ilai Soloducho 开场介绍了自己的技术与管理背景。他目前担任食品与饮料行业巨头施特劳斯集团（Strauss Group）的首席 AI 官（Chief AI Officer）兼集团 CEO 幕僚长（Chief of Staff to Group CEO）。在此之前，他曾在 Google X（Moonshot Factory）负责全公司的 AI 代码生成能力开发，随后团队并入 Google Labs 主攻 AI 智能体（AI Agents）；早年他 7 岁自学软件开发，12 岁创办社交网络创业公司，提前完成计算机科学学业，服役于以色列情报部门并供职于以色列总理办公室；此后在 Palantir 担任数据与解决方案专家超过 5 年，服务于西方政府核心部门与《财富》500 强企业，并拥有哈佛大学肯尼迪学院硕士学位，专精于大型机构与政府的技术赋能。 在谈及 AI 领导力的核心哲学时，Ilai 强调 AI 的最终目的是辅助组织做出更优质的决策并交付更卓越的业务成果。然而，在铺天盖地的技术炒作（hype）中，AI 领导者必须将目光牢牢锁定在**损益表（P&L）的影响与投资回报率（ROI）**上。他将推动业务变革的举措划分为两大类： 1. **速赢项目（Quick Wins）**：无需耗费天价的“尖端火箭科技”，有时仅仅是一个价值 20 美元的软件订阅许可，就能在数小时或数天内释放员工生产力； 2. **长远大石（Big Stones）**：涉及运营和制造流程深度重构等重大项目，通常需要 6 个月以上才能显现成效。AI 绝非一朝一夕能削减 10% 成本或拉升 15% 营收的“魔法”，领导者必须双轨并行，兼顾短期成效与长期战略演进。 - 宏观潜力巨大：全球 AI 潜在经济价值达 4.4 万亿美元，88% 的组织已开始探索或引入 AI； - 规模化落地鸿沟：高达 80% 的概念验证（POC）在尝试规模化推广（Scale）时宣告失败； - 演示与生产的脱节：大量初创公司在 Demo 和宣传视频中表现亮眼，但真正部署到大型企业复杂生产环境时却难以支撑； - 核心破局点：AI 成功落地的瓶颈往往不是技术本身，而是企业组织机制、流程与心理认知系统的建设。 --- ### [04:08](https://www.youtube.com/watch?v=RhnVyGMiYac&t=248s) 借《头脑特工队》拆解 AI 变革：组织情绪控制台与“恐惧（Fear）”应对 为了生动阐明组织在面对技术剧变时的心理状态，Ilai 借用了皮克斯动画电影《头脑特工队》（Inside Out）中控制台与各种情绪拟人化角色的隐喻。在企业推进 AI 转型过程中，员工必然会产生恐惧、焦虑、怀旧、兴奋等复杂情绪；AI 领导者的职责就如同大脑控制台的操作者，既要接纳并理解这些情绪的自然存在，又要合理调控，引导组织跨越心理障碍。 在处理第一种常见情绪——**恐惧（Fear，即对“AI 是否会取代我们”的担忧）**时，Ilai 列举了技术史上两个经典的范例来破除恐慌： - **20 世纪 60 年代末至 70 年代初的自动取款机（ATM）**：当时舆论普遍预测银行柜员将被全面淘汰，但实际结果是单家分行的运营成本大幅下降，银行开设了更多分支机构，柜员总数不降反升，且职能转向财富管理和投资顾问等高附加值领域； - **2010 年代机器学习进入放射科（Radiology）**：尽管外界曾断言放射科医生即将失业，但由于检查效率提升、成本降低与筛查需求普及，放射科医生的实际雇佣数量反而进一步增长。 ```mermaid flowchart LR accTitle: 技术演进与岗位转型机制 subgraph S1["技术引入"] A["引入自动化工具<br>(如 ATM / 影像 ML)"] --> B["运营成本大幅降低"] end subgraph S2["规模与职能演进"] B --> C["网点/服务供给规模化扩张"] C --> D["释放人力转向高价值业务<br>(理财顾问 / 深度诊疗)"] end ``` 针对组织内部的恐惧，破局关键在于**赋能而非替代**的思维转变（Mindset Shift）： - **“超人战袍”视角**：将 AI 视为赋予员工超能力的战甲，使员工成为更强大的专业人士； - **定制化技能提升（Upskilling）**：拒绝泛泛而谈的通用培训，针对制造、法务、销售、市场等不同职能定制具体的实操课程； - **亲自动手的实践式学习（Practical Hands-on Learning）**：让员工在真实业务场景中把玩工具，亲身体验效率跃迁； - **涟漪效应（Ripple Effect）**：当员工感受到管理层对 AI 建设的长期承诺并用上先进工具时，其工作成就感与创新归属感将呈指数级提升。 --- ### [09:35](https://www.youtube.com/watch?v=RhnVyGMiYac&t=575s) “怀旧（Nostalgia）”与“焦虑（Anxiety）”：敬畏旧流程与杜绝无效投入 在分析**怀旧（Nostalgia）**情绪时，Ilai 提醒道：“并非所有问题都需要用 AI 来解决。”变革者必须保持谦逊，充分尊重企业既有的业务流程与行之有效的解决方式，切忌在组织内部无节制地强推 AI。领导者需要明确划定三类边界： - **保留（Preserve）**：对运作良好、高度稳定的流程保持敬畏，不要盲目触碰； - **淘汰与重塑（Discard）**：如果 AI 只能带来 1.1 或 1.2 倍的微小改善，往往不值得投入巨大精力；只有能够实现至少 **2X（两倍以上）质变**、切实解决核心痛点的领域，才值得打破旧机制建立新范式； - **从 0 到 1 创新（Innovate）**：在能够推动业务加速增长的全新赛道上探索 AI 原生能力。 - **决策数据源**：避免凭直觉拍脑袋或受内部政治驱动，需深度融合内部调研（深入车间工厂与一线业务单元）与外部对标（行业最佳实践及专业咨询机构洞察）。 随后，针对**焦虑（Anxiety）**这一体现为“组织因害怕落后而盲目争抢各种 AI 场景”的情绪，Ilai 敲响了警钟：盲目跟风会导致高达 **80% 的投入沦为毫无实际业务影响的无效开支（Impactless Spending）**。 | 决策考量维度 | 核心评估问题 | |:--- |:--- | | **数据就绪度（Data Readiness）** | 内部数据是否经过治理、清洗并具备 API/向量化接入条件？ | | **价值实现时间（Time to Value）** | 项目属于即时见效的速赢，还是长期投入的大石？ | | **交付难度（Delivery Easiness）** | 现有技术栈与组织协同能否支撑稳定部署？ | | **投资回报率（Clear ROI）** | 是否能直接体现在损益表（降低成本或增加收入）上？ | | **定制与预算匹配（Fit & Cost）** | 在庞大的开源/商业方案中，预算与定制化程度是否最优？ | 领导者必须建立清晰透明的量化准入标准，同时梳理出务实的“方案选型菜单”，既明确“应该聚焦什么”，更坚决确立“**坚决不做什么**”。 --- ### [13:18](https://www.youtube.com/watch?v=RhnVyGMiYac&t=798s) “喜悦（Joy）”与盲区防范：建立量化评估机制（Evals） 当 AI 战略与业务工作流完美契合、团队效率大幅跃升时，组织便迎来了**喜悦（Joy）**时刻。庆祝速赢成果固然重要，但狂热之下极易滋生盲区。许多团队在 Pilot（试点）阶段依靠主观直觉评价模型，轻率给出“这个模型很棒”或“那个工具很差”的模糊结论，导致项目在推向生产环境时频频翻车。 为了确保系统稳定可靠，必须建立严密的**评测基准（Evaluations / Evals）**： - **构建黄金数据集（Golden Datasets）**：人工精心设计并标注 15 到 20 个极具代表性的高难度 Prompt 及标准答案输出； - **质量量化追踪**：通过指标体系将模型表现量化（如准确率 90% vs 60%），用硬性数据代替主观臆断； - **生产准入标尺**：只有通过量化 Evals 验证的方案，才具备从实验室 Demo 走向大规模生产（Production）的资格。 --- ### [14:39](https://www.youtube.com/watch?v=RhnVyGMiYac&t=879s) 核心基石：构筑“信任（Trust）”与夯实底层数据资产 Ilai 指出，《头脑特工队》中虽然没有**信任（Trust）**这个角色，但信任却是整个企业 AI 战略的北极星。AI 转型绝非单打独斗的个人表演，而是涵盖 IT、网络安全、一线工厂、销售、HR、财务及法务隐私合规团队的全局协同战役。企业必须密切追踪实际业务价值转化（而不仅仅是表面调用量），严格坚守法律合规底线，确保所有利益相关方步调一致。 ```mermaid flowchart TD accTitle: 企业级 AI 转型协同架构 subgraph Foundation["底层核心资产"] D1["底层数据治理与管道建设<br>(占比 80-90% 的脏活累活)"] end subgraph TechCore["AI 技术赋能层"] D1 --> T1["企业私有数据检索 (RAG)"] D1 --> T2["专业模型微调 (Fine-tuning)"] T1 & T2 --> T3["自主智能体协同 (AI Agents)"] end subgraph GovAlign["组织治理与信任护栏 (Trust)"] G1["业务价值 ROI 考核"] G2["持续量化评测 (Evals)"] G3["法务、合规与数据隐私"] G4["跨部门协同 (IT/安全/制造/销售)"] end TechCore --> GovAlign ``` 在数据战略方面，Ilai 重申了《经济学人》（The Economist）2017 年提出的论断——“数据是新时代的石油”： - **报表光鲜背后的基石**：结合在 Palantir 的实战经历，许多高管总想要酷炫的可视化仪表盘（Fancy Dashboard）或精美分析报告，但这些只是浮在水面上的表象；真正耗费 80% 至 90% 精力且不可或缺的，是底层数据管道（Pipelines）与数据清洗治理的“脏活累活”； - **告别纯原生大模型限制**：企业若想摆脱通用大模型（Vanilla LLMs）和基础提示词工程（Prompt Engineering）的浅层能力，深入推进检索增强生成（RAG）、特定领域微调（Fine-tuning）乃至复杂 AI Agent，必须将经过互联网海量语料预训练的基础大模型与企业内部安全可靠的高质量数据深度融合。 --- ### [17:34](https://www.youtube.com/watch?v=RhnVyGMiYac&t=1054s) 领导力核心价值观与长期主义：马拉松的第 20 分钟 在演讲尾声，Ilai 向在场的 AI 领导者分享了指导长期转型的核心价值观： 1. **持续的好奇心（Curiosity）与终身学习**：AI 领域发展极快，当前看似先进的工具与模型，在 3 到 6 个月内就可能被颠覆淘汰； 2. **谦逊（Humbleness）**：清醒认识到认知的局限性，不盲从技术狂热； 3. **担当与责任心（Accountability）**：领导者不仅要对实施过程负责，更要对最终业务结果负全责。Ilai 举例批评了某些员工用 AI 自动生成 PPT 汇报却对页面上的数据一问三不知的现象，强调工具使用者必须对 AI 生成内容的真实性与准确性承担兜底责任； 4. **使命导向与预期管理（Mission-Oriented & Expectation Setting）**：管理好全员预期，认识到 AI 是一场旷日持久的组织进化。 他回忆道，自己在 2021 年加入 Google X 时就曾笃定预言 AI 即将改变世界，当时身边的朋友多半不以为然；直到 2022 年底 ChatGPT 横空出世，全球才骤然陷入狂热。Ilai 用足球赛作比喻：“如果把这场变革比作一场 90 分钟的世界杯足球赛，**我们现在才刚刚走到第 20 分钟**，这甚至还不算加时赛和点球大战。”他呼吁企业切勿过度依赖外部供应商和外包承包商，而应趁早打造属于组织自身的内部 AI 核心肌肉与底层基础设施——做好组织准备、保持战略聚焦、狠抓落地执行与度量，以持续学习的姿态迎接这场漫长而激动人心的技术长跑。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) AI 的最终目的是辅助组织做出更优质的决策并交付更卓越的业务成果。 — Ilai Soloducho > [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) 有时仅仅是一个价值 20 美元的软件订阅许可，就能在数小时或数天内释放员工生产力。 — Ilai Soloducho > [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) AI 绝非一朝一夕能削减 10% 成本或拉升 15% 营收的“魔法”，领导者必须双轨并行，兼顾短期成效与长期战略演进。 — Ilai Soloducho > [00:04](https://www.youtube.com/watch?v=RhnVyGMiYac&t=4s) AI 成功落地的瓶颈往往不是技术本身，而是企业组织机制、流程与心理认知系统的建设。 — Ilai Soloducho > [09:35](https://www.youtube.com/watch?v=RhnVyGMiYac&t=575s) 并非所有问题都需要用 AI 来解决。 — Ilai Soloducho > [09:35](https://www.youtube.com/watch?v=RhnVyGMiYac&t=575s) 如果 AI 只能带来 1.1 或 1.2 倍的微小改善，往往不值得投入巨大精力；只有能够实现至少 2X（两倍以上）质变、切实解决核心痛点的领域，才值得打破旧机制建立新范式。 — Ilai Soloducho > [13:18](https://www.youtube.com/watch?v=RhnVyGMiYac&t=798s) 只有通过量化 Evals 验证的方案，才具备从实验室 Demo 走向大规模生产（Production）的资格。 — Ilai Soloducho > [14:39](https://www.youtube.com/watch?v=RhnVyGMiYac&t=879s) 数据是新时代的石油。 — 《经济学人》 > [14:39](https://www.youtube.com/watch?v=RhnVyGMiYac&t=879s) 真正耗费 80% 至 90% 精力且不可或缺的，是底层数据管道（Pipelines）与数据清洗治理的“脏活累活”。 — Ilai Soloducho > [17:34](https://www.youtube.com/watch?v=RhnVyGMiYac&t=1054s) 如果把这场变革比作一场 90 分钟的世界杯足球赛，我们现在才刚刚走到第 20 分钟，这甚至还不算加时赛和点球大战。 — Ilai Soloducho ## Mentioned Ilai Soloducho, 施特劳斯集团 (Strauss Group), Google X, Google Labs, Palantir, 哈佛大学肯尼迪学院 (Harvard Kennedy School), 《头脑特工队》 (Inside Out), 《经济学人》 (The Economist), ChatGPT ## Video description Ilai Soloducho - Chief AI Officer & Chief of Staff to Group CEO, Strauss Group With AI adoption at an all-time high, why does transformative business value remain so elusive at scale? While it is tempting to focus solely on the technology, I will explain why the true challenge of AI integration in the food and beverage sector lies in the human process and organizational psychology. This talk will illustrate how to navigate the complex emotional landscape of AI adoption—moving from fear and anxiety to joy and trust. I will share practical frameworks for upskilling teams, preserving what works, filtering out impactless noise, and ensuring every initiative delivers measurable ROI. Join me to learn how to prepare your organization for the new journey of AI leadership and turn early quick wins into long-term strategic evolution. [Source](https://www.youtube.com/watch?v=RhnVyGMiYac)

[0:04] Ilai Soloducho: AI 的最终目的是辅助组织做出更优质的决策并交付更卓越的业务成果。 [0:04] Ilai Soloducho: 有时仅仅是一个价值 20 美元的软件订阅许可，就能在数小时或数天内释放员工生产力。 [0:04] Ilai Soloducho: AI 绝非一朝一夕能削减 10% 成本或拉升 15% 营收的“魔法”，领导者必须双轨并行，兼顾短期成效与长期战略演进。 [0:04] Ilai Soloducho: AI 成功落地的瓶颈往往不是技术本身，而是企业组织机制、流程与心理认知系统的建设。 [9:35] Ilai Soloducho: 并非所有问题都需要用 AI 来解决。 [9:35] Ilai Soloducho: 如果 AI 只能带来 1.1 或 1.2 倍的微小改善，往往不值得投入巨大精力；只有能够实现至少 2X（两倍以上）质变、切实解决核心痛点的领域，才值得打破旧机制建立新范式。 [13:18] Ilai Soloducho: 只有通过量化 Evals 验证的方案，才具备从实验室 Demo 走向大规模生产（Production）的资格。 [14:39] 《经济学人》: 数据是新时代的石油。 [14:39] Ilai Soloducho: 真正耗费 80% 至 90% 精力且不可或缺的，是底层数据管道（Pipelines）与数据清洗治理的“脏活累活”。 [17:34] Ilai Soloducho: 如果把这场变革比作一场 90 分钟的世界杯足球赛，我们现在才刚刚走到第 20 分钟，这甚至还不算加时赛和点球大战。

[0:04] Ilai Soloducho: The fundamental mission of enterprise AI is not chasing technological hype, but using AI to make better decisions that directly impact the organization's profit and loss (P&L) statement. [0:04] Ilai Soloducho: A massive operational chasm separates a prototype demo from production-grade enterprise software. [4:08] Ilai Soloducho: AI will replace us. [4:08] Ilai Soloducho: Equip staff with a 'Superman suit'—technology that amplifies human capability rather than replacing the individual. [9:35] Ilai Soloducho: If an AI tool promises only a marginal efficiency gain (such as a 1.1x or 1.2x boost), it is rarely worth the organizational disruption and operational overhead. [9:35] Ilai Soloducho: Without strict governance, 80% of AI expenditure becomes 'impactless spending.' [13:13] Ilai Soloducho: Trust is the true North Star of your AI strategy. [13:13] Ilai Soloducho: 80% to 90% of the actual work resided in data engineering: data cleansing, pipeline construction, ingestion, and governance. [17:34] Ilai Soloducho: Using AI never absolves an individual from knowing the facts. [17:34] Ilai Soloducho: Enterprise adoption is only at the 'minute 20 out of 90' mark—excluding extra time and penalty shootouts.

[0:04] So, hello everyone. It's great to be here.  Thank you so much for joining this session. [0:09] Uh my name is Eli Soloduho. I'll say a  few words about myself before we begin. [0:15] Uh so, I'm the Chief AI Officer and the Chief of  Staff to the Group CEO at Strauss Group, a leading [0:22] company in the world of food and beverage.  Before Strauss, I worked in Google X, called [0:27] Moonshot Factory, where I focused on developing AI  for coding capabilities across Google. Uh and then [0:35] my team graduated to Google Labs, also focusing  on developing Google's AI agents and capabilities. [0:42] Uh my background is pretty technical. I was  a software development since I was 7 years [0:46] old at the startup I had at the age of 12, a  social network. I finished computer science [0:51] studies pretty early, and I served in the  Israeli intelligence, worked in Israel Prime [0:55] Minister's Office, and then I worked for more  than 5 years in Palantir, a big data company, [1:01] where I work with key stakeholders from Western  governments and Fortune 500 companies to help [1:07] them make better decisions based on data. Uh and  then I did my masters at Harvard Kennedy School, [1:13] uh where my focus was how to leverage technology  in large institutions and governments. Um so, [1:22] uh before we begin, I would like to say a few  words about the philosophy and the core the core [1:28] vision. In the end of the day, our goal is to make  better decision, use AI to make better decisions, [1:34] uh and deliver better outcomes across the entire  organization. But, what what does it actually [1:40] mean, deliver better outcomes? And in my view, the  focus should be on impact on P&L. There's a lot of [1:48] hype and buzz around AI, but I think it's really  important for you, as the leaders in the world of [1:54] AI, to focus on the things that really matter,  the things that really that move the needle. [2:00] And think what are the things that really can have  ROI and impact on our P&L. And usually I divide it [2:06] into two categories. The first one is what we  call quick wins. It's not always that we need [2:12] rocket science technologies to make AI in our  organization. Sometimes it could be a $20 license. [2:19] It could be something that can unleash the  organization to run faster and drive productivity. [2:26] And the second area is what I call big stones.  Those are things that for the long term. In the [2:31] end of the day, AI is not magic. It's not that  in a day we can reduce cost by 10% or increase [2:38] revenues by 15%. Those things take time. If we're  thinking about our world drive AI in operations in [2:45] manufacturing, those are things that are for the  long term. Usually can take more than 6 months. [2:51] Versus quick wins that we can see results even in  a matter of hours or days. And what I think that [2:57] you as future leaders in AI think it's that  we should do both of them simultaneously. [3:03] We should drive the quick wins but also think  for the long terms about the big stones. [3:09] So, I think that you know, all of you are aware  of if you have heard of the huge potential of AI, [3:14] right? I think it says $4.4 trillion of the  potential and we see 88% of the organization [3:21] actually adopting AI. But we also see  that 80% of the POCs fail to scale. [3:29] Right? So, we see organization using AI, but it's  different than using AI in a pilot or in a demo [3:37] versus having it in production. We see startups  that raise millions or even billions of dollars [3:43] and they look really well on the demo, on a  video. But in the end of the day, there's a huge [3:47] gap if we want to scale it in a in a in a large  organization. And today, you know, the easiest [3:54] thing for me would to focus on technology,  but today I want to focus on the process [3:59] of the system, of what's the right way to do it in  order to really see results in your organization. [4:08] So, how many of you, raise your hand  if you've seen the movie Inside Out? [4:13] Oh, wow. Quite a few. So, I will  for those of you who haven't seen it [4:19] This is Joy, coming to you live  in Riley's mind. Make some noise. [4:26] Sadness is in the house. It's  Anger. Let me [screaming] at 'em. [4:33] We got to get our mouth guard,  people. Fear. No, no, that's not ours. [4:37] Disgust. Glad to have her on our team. Our  little girl's growing up so fast. Ow. Sorry. [4:46] It should be nothing but smooth sailing from  Hello. I'm Anxiety. Where can I put my stuff? [4:52] A new emotion. sorry. We wanted to make such  a good first impression. What do you mean, we? [5:00] I'm Ennui. [gasps] Look at your hair. Oh,  yeah, not happening. That's on we. On what? [5:08] It's what you would call the boredom. What's  your name, big fellow? That's Embarrassment. [5:15] Welcome to headquarters, Embarrassment. Oh,  we're doing a fit. No, oh, no, go hide. Oh. [5:21] You got a real sweaty palm  there, buddy. [laughter] [5:26] By the way, it's a great movie. If you haven't  seen it with your kids or families, I would highly [5:30] recommend. So, the movie talks about that, you  know, kids have feelings, right? It's okay they [5:36] feel fear sometimes. It's okay they feel anxiety.  It's okay they feel joy. But, the concept around [5:42] this movie is that we have this control panel that  you should control when you use fear, when you use [5:47] the excitement, when you use the nostalgia. And  I think that the same thing happens in AI when we [5:53] drive a transformation in organization. Our role  is to be that control panel, and it's okay that [5:59] the organization have different feelings. It's  okay they fear from AI. It's okay they are very [6:03] anxious about that, but we need to control that.  So, let's start with the first feeling, fear. [6:09] When the unexpected is staring you in the face,  fear is there to keep you safe. [screaming] With [6:15] speed and wisdom, fear assesses the situation  and settles your nerves in his own special way. [6:23] He's got this. Thank you, fear. Get to know  all So, I'm sure many of you are, you know, [6:31] hearing podcast and articles and and you know,  everybody's talking about will AI replace us? [6:37] And to answer this question, let me take you  back in history to the late '60s, early '70s, [6:43] when the ATMs, the automatic teller machines, were  introduced. So, instead of depositing checks or [6:49] withdrawing cash with a bank teller,  you would do it in a with a machine. [6:54] And all those and all the experts said back  then that, "Hey, now that we have the ATMs, [6:59] we don't need the bank tellers anymore,  and their number will be will decrease." [7:04] But in fact, what happened is that the number of  that bank tellers only increased. So, you know, [7:11] the number of branches increased, and it was  cheaper to open more branches, and the bank [7:16] tellers actually did other things, right? They did  wealth management, investment, and other stuff. [7:23] Another example is not so so far ago, 2010,  when machine learning models were introduced [7:30] in the world of radiology. So, also, people say,  "Hey, we don't need radiologists anymore because [7:35] we have all the machine learning models." But in  fact, the number of radiologists only increased. [7:41] Why? Because it was again was cheaper. You  could scale it more. More people could use [7:46] this re-examination of radiology and now  we see more radiologists than before. [7:52] So in order to confront the fear and to in order  to help your organization, so you would really [7:58] need to focus on mindset shift, right? We need to  let them feel empowered. And it's not that it's [8:04] going to replace us like, you know, like a better  version of themselves. Think about a Superman suit [8:10] that you're putting on yourself that give you  extra power. So this is what we think about [8:15] what we could give our employees when we think  about AI. Obviously upskilling them, giving them [8:21] all the knowledge and the literacy needed in  order to be able to deal with the world of AI. [8:27] And I think that it also should be also customized  to their needs. It's different when you see, for [8:32] example, a training or or an exercise about how  to use AI versus how to do it in manufacturing, [8:38] how to do it in legal, how to do it in sales,  how to do it in marketing. So this is also [8:43] very important and finally practical learning. All  the time get their hands dirty. Let them play with [8:49] it. Not just keeping in the theory, but when they  see the value, when they play it with their hands, [8:55] when they see real use cases, they could  actually feel the the big big potential of AI. [9:02] And what I believe this creates is the  ripple effect. So um I think that, you know, [9:07] when employees feel that that when they use  AI, it also give them the higher level of [9:12] satisfaction of their of their work. They feel,  "Hey, I'm working in a in a in a in a place that [9:18] give me all those cutting edge technologies and  I feel that I'm working in an innovative place." [9:23] So I think it's also really important  for them to feel satisfied in their work [9:27] and also see that, "Hey, the the leadership  is actually committed to the process of AI." [9:35] Let's move to nostalgia. Oh, remember when we all  finally came up to headquarters? That was like 30 [9:42] seconds ago, nostalgia. Yeah, those were the days.  Nostalgia, you weren't supposed to be here yet. [9:49] You still have about 10 years, two graduations,  and a best friend's wedding before you're invited, [9:53] but I will keep you in the loop, I promise. So,  look, not everything needs to be solved with AI. [9:59] We need to respect the existing processes,  the existing ways that people do work, [10:06] and and solve their problems. And we need  also to be very humble and understand, "Hey, [10:11] what are the things that we need to  preserve?" Don't touch it. They work. [10:15] Don't don't do everything about it. Don't put AI  on every single place in the organization. What we [10:20] need to discard, what are those things that don't  work? Or what are the things that we could do at [10:25] least two X difference. If it's like 1.1 or 1.2,  like it's slightly better than it was in the past, [10:32] maybe it doesn't worth the effort, right? But  for those things that really move the needle, [10:37] we need to dis- that we need to discard the the  old way and bring a new way, and where can we [10:43] innovate? Where can we bring zero to one, where  we can bring new stuff to accelerate the growth [10:48] of the business? And you know, my advice is  use multiple data sources to have as many as [10:56] data points to make an informed decision. Not  just think about your intuition where to put AI [11:03] or about internal politics. Think about internal  research that you're doing in the organization, [11:08] working with people, going to the factories, going  to the business units, but also external research. [11:14] And that means what is the industry doing? What  is the benchmark? What are other players doing [11:20] right now to accelerate AI and also working  with consulting companies that have this [11:25] type of knowledge. So, I think the mix of two and  being very data-driven here is the key to success. [11:33] Anxiety. Orange? Who made the console orange? Do  I look orange? I didn't touch it. Orange is not [11:37] my color. Not me. Hello everybody. Oh my gosh, I  am just such a huge fan of yours and now here I am [11:45] meeting you face-to-face. [laughter] Okay, how can  I help? Um I can take notes, get coffee, manage [11:51] your calendar, walk your dog, carry your things,  watch you sleep. Wow, you have a lot of energy. [11:56] Maybe you could just stay in one place. Anything.  Just call my name and I'm here for you. Okay, [12:01] love that. And what was your name again? Oh, I'm  sorry. I can get ahead of myself. I'm Anxiety. [12:05] I'm one of Riley's new emotions and we are just  super jazzed to be here. Where can I put my stuff? [12:11] What do you mean we? So you know anxiety is when  you know we want to do everything, right? When [12:16] your organization they hear about AI, they want to  do all type of use case. Everybody wants to have [12:20] it. But wait, 80% is impactless spending. And what  is really important is what should we focus on and [12:29] even more important what would we not focus on.  And I think it's very important to have a clear [12:35] defined criteria of what we're doing. Is it about  data readiness? Is it about the time to value? [12:41] Is it about delivery easiness? Is it about ROI?  Is it a mix of them? But having a clear defined [12:47] criteria of how we make decisions and in the end  of the day we also want to have a menu of options, [12:53] right? Because we see I'm sure you're here  you're hearing from startups, you're hearing [12:57] from large company. There's so many solutions  out there. So we also should come very humble [13:03] and think about different different solution,  different options in order to really solve our [13:08] problems. What are the things that in terms of  budget, in terms of productivity, in terms of [13:13] customization really fit our needs here. Joy.  When everything in your life is going your way, [13:21] that's when Joy takes control. This is Joy. With  an unstoppable spirit and unbeatable energy, [13:29] Joy brings a smile to your face  and no, who cares? Let's party! [13:37] So, look. For me, Joy is what happens when  AI strategy align perfectly and workflows are [13:41] optimized. And you know, there's this excitement  in the organization. We can celebrate the quick [13:47] wins. Everybody is happy. But, we should also be  very cautious of the blind spot. So, to give you [13:55] an example, evals, evaluations, right? Building  those golden data sets. For example, having those [14:01] 15 or 20 prompts that we do them manually,  right? And we actually create the output, [14:07] and then we can quantify, we can quantify the  quality of AI model. Instead of just saying, "Hey, [14:13] this AI model is great, or this one is terrible,  or this one is the best based on our intuition." [14:20] We can actually use numbers. We can say, "Oh, this  solution or this model is 90% right, 60% right." [14:26] So, when we use those evals, this is crucial when  we move from pilot phase to production phase. [14:33] Having metrics and KPIs, and we'll talk about  it in a second. So, trust is actually something [14:39] that is not part of the movie. I thought it's  really important, so I added it and I created [14:44] a movie for a creature for trust, obviously using  AI. Console spins out of control. [music] One [14:50] emotion brings the ultimate mandate. Meet Trust,  the true North Star of your AI strategy. Okay, so [15:01] I don't know if the same as in the movie,  but try to did my best because I really [15:05] think trust matters. Okay, trust matters. It's  not a one-man show, it's not just about you, [15:11] it's about the whole organization. You have IT,  you have the security, you have the the people in [15:16] the in the in the factories, you have the sales  people, you You the HR, the finance, you have [15:23] a whole organization because AI it's not just  about technology. AI is not just about people. [15:28] AI is not about process. It's a mix of things.  When we really want to do an an an transformation, [15:34] we need to have everyone, all the relevant  stakeholders aligned about it. Clear ROI. We [15:40] need to focus on things that matter and obviously  measure adoption. It's not just about the usage, [15:46] it's about the value that the organization is  getting. Legal and privacy. In the end of the day, [15:50] we need to be responsible, compliant, and  work very closely with our legal partners. [15:56] Constant evil, we talked about it a minute ago,  and strategy. What are we doing and what are [16:01] we not doing? Make sure everybody is aligned  about it. Let me say a few words about data. [16:08] Data is king or the new [clears throat] oil  as The Economist magazine defined it in 2017. [16:12] And in the world of AI, it's even more  crucial. So, let me person let me share [16:17] a personal story. When I work in Palantir, I  worked with leaders like you uh in New York, [16:22] in in in UK, in Switzerland, Brazil, Australia,  and others. And when I asked those leaders, "What [16:28] are your biggest problems?" So, most of them would  say, "Hey, I want a fancy dashboard or you know, [16:33] a fancy an analysis or you know, some In the end  of the day, this is the easy part. The dashboard, [16:39] the reporting, the analysis. The hard part, the 80  90% of the work, even though it's not sexy and not [16:45] everybody's talking about it, is data. And the  building the data transformation and pipelines. [16:51] And once the data is reliable and safe,  that makes the rest of the work much easier. [16:57] And same happens in the world of AI. If we want  to take our organization to the next level, [17:02] and not just, you know, use vanilla LLMs and be  limited to prompt engineering, we would want to [17:09] do fine-tuning, we would want to do rag retrieval  augmented generation. And for that, we would need [17:15] to leverage our internal data together combined  with the powerful LLMs that were trained on post [17:22] training on all the data in the in the internet  out there. So, data is something that is crucial [17:28] if we want to have agents and all the other things  that are very very important for our organization. [17:34] Um, a few values that I want to talk about.  First is curiosity. Even though that now you know [17:42] ChatGPT 5.6, so you know Fable or Mythos and  you're top continuous learning being all the [17:48] time curious about it is really important. This  world is moving very fast. In 3 to 6 months, what [17:55] we are talking right now may be not be relevant,  right? So, we need all the time keep learning. Uh, [18:01] humbleness, you know, we don't know everything.  Accountability. You should be accountable over [18:07] the process, but also over the results. I've seen  the other day someone presenting slides with AI [18:13] and I'm fine. I like people presenting using AI  to build their slides, but when I asked him, "Hey, [18:18] what's this number mean? Or what did you write  there?" He was not accountable for that. He didn't [18:23] know what the slides were. So, it's also important  to be accountable for the results of the AI. [18:29] Mission-oriented and of course setting  expectation. This is just the beginning. [18:34] Okay, when I joined X, I remember like  it was in 2021, I told my friends, "Hey, [18:39] this is going to change the world." I could see  behind the scenes how amazing this technology was. [18:44] And everybody said, "Come on, Eli. We heard  about AI." And then when ChatGPT emerged in 2022, [18:49] everybody said, "Oh, AI. This is so exciting."  And everybody is all the time talking about it. [18:55] So, even here I'm saying, "Hey, this is a long  journey." If you like soccer like me or you watch [19:00] the World Cup, I would say we're in minute 20 out  of 90 not including extra time and penalties. So, [19:06] this is a long long journey and I think it's very  important for you to build this internal muscle [19:11] in your organization and not just rely on vendors  and and other contractors so because we need to [19:17] prepare ourselves and the infrastructure  for the future. Prepare your org, focus, [19:23] execute and measure and keep learning and  you will enjoy the ride. Thank you very much.