Companies Slash Costs With GenAI as Leaders Warn on Skill Decline

企业借助生成式AI大幅降本，领袖警示技能退化风险 要从生成式AI中获得可衡量的回报，需将模型直接嵌入企业核心工作流，但过度依赖可能导致以长期的认知能力退化换取短期的生产力提升。 来自多个行业的与会嘉宾概述了生成式AI在企业中的实际应用，展示了多项切实成果，例如将宣传视频制作成本从10万美元降至3,000美元，并实现全机构20%至26%的生产力提升。发言者强调，要获得切实的投资回报率，必须解决底层数据架构挑战，将工具原生嵌入日常现有工作流程，并在实施前设定严格的KPI指标。在详述高价值机遇（如实现医疗领域数千亿美元收入周期管理的自动化，以及将软件原型开发周期缩短至48小时）的同时，讨论也提醒要警惕供应商的过度承诺和独立的仪表板类应用。小组讨论最后针对人的层面展开了紧迫探讨，警告称尽管AI能让员工成为广泛领域的“超级创造者”，但组织必须积极防范认知能力退化与思维懈怠。 生成式 AI 能大幅削减创意资产的制作成本，例如 AREA15 仅花费 3,000 美元便制作出原本需要 10 万美元传统代理预算的广播级宣传视频。 (2:41) 美国医疗系统每年存在高达 1 万亿美元的行政浪费，其中仅收入周期管理（RCM）就占据了 2,000 亿美元可自动化的工作流市场空间。 (6:50) 企业在 2025 年面临的最大风险是按兵不动与表面化使用；实现真正的 ROI 需要将生成式模型深度嵌入到核心业务系统与工作流中，而非局限于单独标签页的聊天机器人。 (8:10) 高价值 AI 项目需要内部推动者、迭代试错包容度以及在启动前由财务团队对齐预设的 KPI 承诺；Gregory FCA 通过开展 102 场培训培养“超级创作者”，带动生产力提升 20%–26% 并使客户流失率下降 34%。 (22:31) 在定制软件工程中，生成式 AI 将需求文档（PRD）编写压缩至 8 小时，并在初步沟通后 48 小时内即可交付可交互的原型。 (27:58)

Achieving measurable returns on generative AI requires transitioning from superficial chatbot prompts to native workflow integration, disciplined governance, and clean data architectures. Panelists across real estate, healthcare, software, and public relations demonstrate significant operational efficiencies while cautioning that automation must not erode essential human critical thinking. Immediate Cost Reductions in Production and Scoping (2:41) Generative AI delivers immediate financial returns when applied to targeted creative and technical workflows. Winston Fisher reports that AREA15 produced a broadcast-grade sci-fi promotional video for $3,000 using generative tools, compared to an estimated agency cost exceeding $100,000 with conventional live-action and CGI methods. In operational marketing, AREA15 collapsed the formatting of event assets for social media from a three-hour manual task to a 30-second automated process. Furthermore, requiring internal teams to conduct preliminary research with advanced reasoning models saved Fisher Brothers approximately $1 million in third-party consulting fees. In software engineering, Konstantin Tsybulko of Dualboot Partners emphasizes speed of execution as an economic differentiator. By embedding generative models into early-stage scoping, Dualboot compressed the preparation of Project Requirements Documents from several business days to eight hours, while turning initial discovery calls into functional clickable prototypes within 48 hours. Data Cleanliness and Native Workflow Integration (19:04) Panelists caution that enterprise AI deployments regularly fail when organizations lack structured underlying data or isolate tools outside daily work patterns. Tsybulko notes that standalone applications isolated in browser tabs go unused; in one manufacturing deployment, an underperforming AI dashboard only gained traction after the company repaired underlying data defects and embedded the model directly into its existing Salesforce interface. He advises that vendors must mandate AI readiness audits to clean client data before deployment, arguing that any initiative must have an ROI clear enough to summarize to a CFO in one sentence. Winston Fisher similarly identifies clean data warehouses as a prerequisite for enterprise success, recounting the failure of an overpromised AI paralegal tool that could not reliably process leases and contracts. Fisher Brothers is currently investing in a structured data lake—termed the 'Golden Join'—to unify fragmented records and prepare long-term architecture for proprietary AI applications. Disciplined Governance and Strategic Healthcare Frameworks (29:11) In high-stakes, regulated environments, realization of value depends on strict problem selection and formal governance. Dr. Alvin Liu highlights a massive structural inefficiency in United States healthcare, where $1 trillion of the nation's $4.9 trillion annual spend (~18% of GDP) constitutes administrative waste. Roughly $200 billion is consumed by Revenue Cycle Management—coding, claims, and prior authorizations—making billing automation a premier target for near-term operational ROI. To safeguard clinical efficacy and capital, Johns Hopkins Medicine requires external AI vendors to secure an internal employee sponsor and pass centralized governance reviews. Johns Hopkins evaluates initiatives against a three-pillar matrix: clinical outcomes, operational efficiency, and financial return. Viable deployments must generally show positive returns in at least two of the three categories, with complete exceptions to financial viability reserved strictly for mission-driven care in rare diseases. Liu stresses establishing agreed-upon ROI metrics with financial executives before pilot launches, securing upfront commitments to fund enterprise scale if benchmarks are achieved. Workforce Evolution Versus the Risk of Cognitive Decline (33:24) To capture human ROI, Gregory FCA conducted 102 AI training sessions over two years, cultivating 'super creators' who offload grammatical mechanics to focus on strategy and client relationships. Founder Greg Matusky links this initiative to a 20% to 26% sustained productivity lift across a $24 million base, alongside a 34% drop in client attrition and a 26% improvement in staff retention. He argues that automation enables lateral thinking and broader coordination, comparing the shift away from routine drafting to a surgeon directing da Vinci robotics. Conversely, Fisher warns that over-reliance on AI risks intellectual complacency and cognitive decline, insisting that staff cannot outsource the foundational understanding of a business problem. Drawing on public policy experience, Fisher notes that workers must master underlying disciplines before automating them, as theoretical transitions into high-level strategy often fail across broader workforces lacking specialized training. Tsybulko reinforces this concern by noting that roughly 70% of automated conference messages he receives are sloppy and ineffective, illustrating that uncritical AI adoption degrades professional output without active human oversight. While generative AI yields substantial savings in asset generation and transactional processing, leadership teams must navigate organizational resistance, poor data infrastructure, and vendor exaggeration. The enduring challenge lies in capturing operational speed without compromising foundational workforce skills and critical analytical judgment.

企业要通过生成式AI获取最大投资回报率（ROI），关键在于越过表层的聊天交互工具，将模型深入嵌入业务工作流以降低直接成本和压缩交付周期，但实现规模化价值的前提是完成底层数据架构清洗并建立严谨的治理准入。跨行业高管同时提醒，在AI重塑工作范式并催生高效生产力的过程中，组织必须警惕因过度依赖工具而引发的基础认知退化与战略理解空心化。 业务场景中的成本压缩与生产力释放 (2:41) 多位高管证实，生成式AI最立竿见影的ROI首先体现在创意资产制作与标准化流程的高效压缩上。Fisher Brothers兼AREA15首席执行官Winston Fisher透露，其利用生成式AI工具为AREA15二期扩建制作的高质感科幻风格宣传片仅花费3,000美元，而采用传统代理商拍摄与CGI特效的预算预估在100,000美元以上；在日常营销中，原先需耗时3小时的艺人多媒体物料排版流程被AI缩短至30秒。此外，其团队通过高阶大模型推理辅助前期商业调研，直接规避了约100万美元的外部咨询顾问费用。在公关传播领域，Gregory FCA首席执行官Greg Matusky指出，通过两年来持续开展的102次内部周度培训，公司在2400万美元的营收基盘上实现了20%至26%的持续生产力增长，客户流失率下降34%，员工留任率提升26%。 在软件工程与医疗运营场景中，技术对周期的压缩同样显著。Dualboot Partners首席营收官Konstantin Tsybulko表示，借助AI工程工具，其团队将项目需求文档（PRD）编制时间缩短至8小时，从初步调研到交付可点击原型的周期被压缩至48小时。约翰斯·霍普金斯大学医学院的Alvin Liu博士则指出，美国每年约4.9万亿美元（约占GDP的18%）的医疗总支出中，无益于患者治疗的行政浪费高达约1万亿美元（占25%），其中单是收入周期管理（RCM）的报销和编码环节就耗费约2000亿美元，通过AI实现RCM标准化重构代表着医疗领域最具确定性的高ROI机会之一。 数据架构清洗与原生工作流嵌入门槛 (19:04) 针对落地阻力，Tsybulko认为2025年企业面临的最大商业风险是组织的迟疑与拖延，但多数领导者仅止步于浅层探讨，未能在核心系统内进行深层流程再造。他强调“孤立存在于浏览器标签页中的工具不会被使用”，并提出评估准则：“如果你无法在一句话内向CFO解释AI项目的ROI，那么你的方向大概率存在问题。”他分享案例指出，一家制造客户曾因底层数据质量欠佳导致AI仪表板输出高置信度的错误结论而遭弃用，直至完成数据清洗并将交互原生嵌入销售日常使用的Salesforce环境中，系统才真正发挥效用，因此技术供应商在部署前推行AI就绪度审计（AI Readiness Audit）至关重要。 Fisher同样提醒，企业级AI落地必然面临阻力，多数机构目前仍处于底层数据治理的“第一局”。他结合自身经历指出，其机构曾尝试部署一款AI律师助理工具，最终因供应商在销售压力下过度承诺而宣告失败。企业必须大力投资于消除数据孤岛的结构化数据湖（即“Golden Join”），同时管理层必须将AI部署视作一个持续试错的渐进流程，并建立包容失败的组织机制。 机构治理与医疗三支柱评估模型 (15:48) 在医疗等高风险受监管行业，确立严密的评估与准入框架是实现有效ROI的核心保障。Liu博士系统阐述了约翰斯·霍普金斯大学医学院推行的三支柱评估体系，即临床ROI（改善健康与生活质量）、运营ROI（精简文书以缓解医务倦怠）与财务ROI（直接增收或节约成本）。常规部署方案必须在三个维度中至少满足两项（如临床加财务，或运营加财务）；唯有在极少数病例极为罕见的孤儿病领域，出于人道主义救治使命，院方才会豁免财务平衡的硬性要求。 为防范概念炒作，大型医疗体系正在推行中心化的AI治理规范。Liu博士介绍，霍普金斯要求所有外部AI商业提案必须首先在院内找到自愿作为内部担保人的正式员工，随后才进入严谨的尽职调查与论证流程；在启动试点前，团队必须自上而下倒推规划，让掌握预算的财务主管从第一天起就针对预设的KPI与量化标准达成共识，承诺一旦试点指标达成便直接拨付全面部署资金。Madison Air首席信息官Kristina Paschall亦补充，制造企业面对海量用例，必须严格依据KPI筛选极少数能带来直接产能和收入提升的项目，并将工具专门交由具备深厚本土业务理解的骨干使用，而非依赖技术弥补专业知识的不足。 超级创造者崛起与认知退化隐忧 (33:24) 生成式AI对专业知识工作者的重塑引发了关于认知演进的深层交锋。Matusky与Liu博士持相对乐观态度，认为将基础机械工作（如语法校对、常规代码编写或医疗缝合）剥离后，人类精力得以集中于高阶策略、同理心与横向联系。Matusky提出未来岗位不会被AI取代，而是会被熟练驾驭工具并实现海量优质交付的“超级创造者”所取代，如同NFL传奇四分卫Tom Brady凭借对整体阵型的宏观洞察主导赛场，专业人才应将焦点转向战略关系与定位。Liu博士指出，由于AI让获取纵向垂直知识的成本大幅下降，过去一个世纪对过度专业化的奖励将逐步让位于连接跨学科节点的“横向思考能力”。 然而，Fisher依据其在公共政策与多行业资产运营中的经验发出了严正警告，指出过度依赖AI极易导致认知退化与长期战略赤字。他强调，基础技能的反复磨炼是建立深层商业直觉不可逾越的必经之路（例如初级员工正是通过被批注红线来深刻理解逻辑与规范），“团队绝不能把对问题的本质理解外包给AI”。Tsybulko以峰会现场接收到约70%低劣、敷衍的自动化机器私信为例，佐证了当前的智力懒惰倾向。多位高管达成共识：当前AI应用面临的核心危险不在于技术本身，而在于从业者缺乏批判性审视与偷懒敷衍的盲目信任。 本场研讨表明，生成式AI在创意制作、软件研发与行政流程精简等领域已展现出显著的降本增效成果，但企业要实现从试验性探索向规模化收益的跃迁，仍受制于数据底座清洗、原生工作流集成以及严格的治理闭环。未解决的核心争论在于：在初级机械劳动被大面积自动化的背景下，企业与社会如何在享受即时生产力红利的同时，避免从业人员的基础认知训练断层与核心业务理解能力的持续退化。

## Background 该视频由行业会议频道 Ai4 于 2025 年 8 月 27 日发布在 YouTube 平台。根据页面简介，本次圆桌论坛聚焦于生成式 AI 创造可衡量投资回报率（ROI）的真实商业案例，邀请了来自制造业、医疗、地产娱乐与公关等领域的企业高管与专家，共同探讨高价值场景的识别、ROI 衡量指标以及实现长期商业回报的最佳实践。 ## TL;DR 来自不同产业背景的领袖分享了生成式 AI 带来显著 ROI 的落地实践与方法论，涵盖 AREA15 仅花费 3,000 美元打造的高质量宣传片、全美医疗体系针对收入周期管理（RCM）万亿级浪费的治理，以及公关公司通过人机协同提升 26% 生产力并降低流失率的实操经验。嘉宾们达成共识：推动 AI 成功不仅依赖底层结构化数据与“以终为始”的严格 KPI 闭环设计，还必须将工具无缝内嵌至日常业务流，而非浮于表面。同时，圆桌警示了过度依赖 AI 可能引发的“认知退化”与理解能力外包的风险，呼吁从专精技能向横向战略思维转变，并妥善应对基层劳动力的技能转型与思维惰性挑战。 ## Atmosphere & style 整场圆桌讨论务实而犀利，彻底褪去了盲目的技术狂热与空泛的公关修辞，洋溢着一线决策者面对技术变革时的坦率与清醒。主持人以吉祥物小毛驴轻松破冰，而嘉宾们在随后的对话中展现了极高的互动张力与思维碰撞，既有展示 3,000 美元震撼视频时的自豪与兴奋，也有直面技术翻车、销售过度承诺以及大会官方 App 充斥垃圾信息时的辛辣吐槽。讨论逐步从具象的降本数字演进到对人类认知退化、精英主义偏差及基层劳动力转型的深刻关切，观感兼具思想厚度与实战启发。 ## Key takeaways - [04:38](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=278s) 生成式 AI 在创意制作与影视宣发管线中展现出惊人的降本效果，例如 AREA15 仅花费 3,000 美元即完成了传统制作需 10 万美元以上的超现实宣传大片，降本幅度超 97%。 - [02:46](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=166s) 底层数据基础设施是决定 AI 成败的关键瓶颈；缺乏高质量结构化数据与数据仓库支持，AI 落地只能停留在业务边缘。 - [06:50](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=410s) 全美每年 4.9 万亿美元的医疗开支中约 25% 属于行政管理浪费，其中规模达 2,000 亿美元的收入周期管理（RCM）是生成式 AI 实现可观 ROI 的万亿级突破口。 - [08:10](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=490s) 2025 年企业面临的最大商业风险并非投错 AI 场景，而是采取消极抵触和不作为，导致在底层工程和日常运营中被善用 AI 的竞争对手淘汰。 - [09:26](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=566s) 公关传媒等文字密集型行业引入 AI 能打破传统写作技能壁垒，将员工从排版、语法等机械性杂务中解放，专注于战略洞察与客户关系，并显著提升客户与员工留存率。 - [13:33](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=813s) 推动 AI 落地需要企业内部建立对技术演进的容错文化，并在底层重金打通不同业务数据湖之间的“黄金连接（The Golden Join）”，实现短期效率工具与长期数据底座并进。 - [15:48](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=948s) 高门槛与强监管行业推进 AI 需采取“以终为始”的试点设计，提前召集临床、业务、IT 与财务预算审批方锁定量化 KPI 与 ROI 算法，确保达标后能顺利全面落地。 - [19:04](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1144s) AI 应用必须深度内嵌到员工已有的日常工作流（如 CRM 系统）中，停留在独立浏览器标签页中的孤立工具注定会被团队遗忘与弃用。 - [22:31](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1351s) 企业 AI 变革本质是一场组织行为管理，通过高频系统性培训将员工打造成“超级创作者”，能有效消除失业恐惧并大幅减少交付中的返工与挫败感。 - [27:58](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1678s) 软件工程领域的 AI 评估公式聚焦于是否省钱、是否赚钱以及是否具备复合加速效应，如将 PRD 编制和可交互原型开发时间从数天缩短至数小时。 - [29:11](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1751s) 医疗机构衡量 AI 需综合临床、运营与财务三维 ROI 框架，作为使命驱动型组织，只要在临床改善与医护减负上效果显著且不致严重亏损，即便财务回报有限也应积极推进。 - [31:24](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1884s) 在聘请外部咨询机构前强制利用前沿 AI 的“深度研究”功能完成问题定义与框架拆解，能直接为企业削减上百万美元的外部顾问开销。 - [33:24](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2004s) 过度依赖 AI 会导致人类思维外包和认知退化，企业绝不能将对业务本质问题的理解外包给模型，否则眼前提效将转化为未来的战略赤字。 - [35:14](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2114s) 在单一领域获取知识的边际成本归零后，人类核心竞争力正在从传统的垂直专精转向全局战略、跨界连接的横向思维与批判性把控。 - [39:31](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2371s) 若基层劳动者缺乏扎实的批判性思维训练而单纯依赖 AI，加之滥用自动化生成低质内容，将加剧全社会的思维惰性与劳动力转型阵痛。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=4s) 圆桌嘉宾介绍与开场 圆桌主持人由 Madison Air 首席信息官（CIO）克里斯蒂娜·帕斯卡尔（Kristina Paschall）担任。帕斯卡尔在开场时介绍，Madison Air 是一家专注于空气流动与控制设备制造的企业，旗下拥有包括威斯康星州的 Broan-NuTone 以及位于列克星敦的 Big Ass Fans 等知名品牌。为了活跃气氛，她还展示了 Big Ass Fans 的吉祥物——一头名为“Fanny”的小毛驴压力玩具（stress donkey），并表示会后可与现场观众分享。本次圆桌会议的核心主题聚焦于“利用生成式 AI 获取最大化投资回报率（ROI）”。 随后，四位来自不同行业的重磅嘉宾依次介绍了自己的背景与业务切入点： * **温斯顿·费舍尔（Winston Fisher）**：纽约家族房地产投资与开发运营商 Fisher Brothers 的管理合伙人，同时也是位于拉斯维加斯的沉浸式娱乐综合体 AREA15 的首席执行官兼创始人（占地达 60 万平方英尺，紧邻永利渡假村）。 * **阿尔文·刘（Alvin Liu）**：约翰斯·霍普金斯大学医学院（Johns Hopkins Medicine）视网膜外科医生、人工智能特聘教授，并担任该院 AI 监督团队（AI Oversight Team）核心成员，统筹医学院内部所有 AI 相关重大战略与决策。 * **康斯坦丁·齐布尔科（Konstantin Tsybulko）**：Dualboot Partners 首席营收官（CRO），该机构致力于通过集成 Cursor、Figma、Copilot 等顶尖 AI 工具及自主研发的 AI 加速套件，为企业提供软件开发全流程加速，实现超过 50% 的研发效率提升。 * **格雷格·马图斯基（Greg Matusky）**：Gregory FCA 创始人兼首席执行官。该公司是全美前 40 大公共关系与传播机构之一，近期被行业协会评为全美“AI 赋能程度最高”的公关公司。 --- ### [02:46](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=166s) 生成式 AI 的价值奇点与企业落地的核心瓶颈 主持人帕斯卡尔抛出了首个核心议题：当前市场对 AI 充满狂热，但企业普遍存在“投错方向”的焦虑与恐惧；各位嘉宾究竟是在哪个关键时刻真正看清了 AI 的巨大价值？ 温斯顿·费舍尔率先作答。他回忆称，在 ChatGPT 刚刚问世时，自己便毫不犹豫地全身心投入，立志成为一名“尽可能专业的非技术背景 AI 专家”。在 Fisher Brothers 和 AREA15 内部，生成式 AI 最立竿见影、节省资金最多的领域当属“创意制作流程”。过去两年来，团队利用 AI 创作内容的速度显著加快、质量大幅提升，而成本却断崖式下跌。此外，在营销获客、租赁管理等流程冗余度高的“低垂果实”场景中，自动化工具也带来了显著价值。 然而，费舍尔也指出了当前企业推进 AI 时最常遭遇的结构性阵痛：如果底层的数据仓库（Data Warehouse）没有搭建完毕、缺乏高质量的结构化数据，AI 的实际应用就只能停留在边缘地带（on the margins）。他坦言，梳理并打通企业级底层数据是组织内部最艰难、最昂贵却又不得不做的一步，目前整个行业依然处于这场变革的“第一局”（first inning）。 --- ### [04:38](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=278s) 3,000 美元的宣传大片：AREA15 的创意降本实践 为了直观证明生成式 AI 带来的降本奇迹，费舍尔在现场播放了一段刚刚为拉斯维加斯 AREA15 第二期开业制作的宣传视频。 ```mermaid flowchart LR accTitle: Video Production Cost Comparison accDescription: Comparison between traditional video production and GenAI production for AREA15 subgraph Traditional["传统影视宣发管线"] T1["外包团队 / 拍摄布景 / CG特效"] --> T2["成本：$100,000+"] end subgraph GenAI["生成式 AI 制作管线"] A1["AI 剧本 / 视频生成 / 声音合成"] --> A2["成本：$3,000 (节省超 97%)"] end ``` 这段带有幽默科幻与新闻播报风格的短片虚构了拉斯维加斯南兰乔大道（South Rancho Drive）发生的一场“创意大熔解”事件： * 突发新闻主持人播报整座城市陷入超现实混乱，大批游客与本地人被不可抗拒地吸引到该街区； * 一架 40 年前在百慕大三角失踪的“空中航线 1515 航班”波音 747 客机突然降落在街区中央，主播甚至在播报中惊愕吐槽：“我刚才到底说了句什么鬼话？”； * 镜头中穿插着巨型艺术装置、餐饮体验与超现实沉浸式空间，旁白声称无法证实该区域是否属于地球文明。 费舍尔在视频播放完毕后揭晓了制作成本：**整部高质量视频的实际制作花费仅为 3,000 美元**。他指出，若按传统影视制作和特效管线，此类规格的视频制作成本至少在 100,000 美元以上。这种百倍级别的效率跃升和成本削减，正是生成式 AI 颠覆内容产业的标志性体现。 --- ### [06:50](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=410s) 医疗领域的万亿级浪费与收入周期管理（RCM）机遇 来自约翰斯·霍普金斯大学的阿尔文·刘教授将视角转向了美国体量庞大的医疗健康产业。他列举了一组震撼的数据：去年全美医疗支出高达 4.9 万亿美元，占全美 GDP 总量的约 18%，是美国经济的第一大板块。然而，这笔巨额开支中约有 **25% 属于纯粹的“行政管理浪费”**——即流程极其繁冗、耗时且重度依赖人工，却完全无法提升患者临床结局的无效流转。 刘教授指出，这构成了约 1 万亿美元的可寻址优化空间。其中，规模约 **2,000 亿美元的“收入周期管理”（Revenue Cycle Management, RCM）** 是生成式 AI 获得可观 ROI 的最大突破口： * **定义与痛点**：RCM 涵盖了医生、医院等医疗服务提供方与各大商业医疗保险机构之间繁琐的账单核对、报销申请、争议申诉与结算回款全流程； * **落地成效**：通过引入生成式 AI 解析非结构化病历并自动化匹配报销代码，大幅缩短了结算周期，消减了无意义的行政人力消耗。 --- ### [08:10](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=490s) 2025 年的最大风险：重构认知与企业级落地的迟滞 Dualboot Partners 的 CRO 康斯坦丁·齐布尔科接续讨论，对主持人的“试错恐惧”问题给出了反向思考。他强调，在 2025 年，企业面临的最大风险**根本不是投错场景，而是彻底对 AI 采取消极抵触和不作为**。 齐布尔科指出，当前企业在拥抱 AI 方面仍存在广泛的阻力与推诿。他直言自己并非悲观主义者，但基本商业现实是残酷的： > “如果你不用 AI 来武装和放大自己，你注定会被那些善用 AI 的人所取代——这对个人和企业而言皆是如此。” 他分析了当前市场中个人端与企业端的巨大断层：在 C 端，几乎所有人都在使用 ChatGPT、Gemini 或 Perplexity；但在 B 端，大多数企业依然处于“言多行少”的阶段，极少有公司真正将 AI 深度植入到软件工程底座、业务经营闭环与日常运营的骨髓之中。 --- ### [09:26](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=566s) 从“生存危机”到行业先锋：公关领域的 AI 赋能之路 Gregory FCA 创始人格雷格·马图斯基回顾了自己的 AI 启蒙时刻。早在四年半以前（远早于 ChatGPT 爆发），他在卡内基梅隆大学（CMU）攻读工程专业的儿子就向他发出预警，称其公关公司正面临“灭顶之灾”（existential threat）。当时身为全美前 40 大公关掌门人的马图斯基大为不解，儿子直接告诉他：“机器很快就能像你手下最顶尖的员工一样去写作、思考和创造。” 起初身为终身写作者的马图斯基对此嗤之以鼻，但随后他迅速启动研发，打造了公司专属的内部 AI 内容平台“Gladwell”（以作家马尔科姆·格拉德威尔命名）。平台早期非常简陋且充满幻觉（甚至胡编称公司的吉祥物是一只绿青蛙），但他敏锐察觉到了其底层的巨大威力。 当 ChatGPT 最终公开发布时，马图斯基将其称为“职业生涯中最快乐的一天”： * **打破技能壁垒**：过去数十年间，教导员工如何精准写作极其痛苦，因为文字表达需要在大脑中构建类似小提琴家或 NFL 四分卫般的复杂神经回路； * **回归核心价值**：AI 的接入消除了机械性的文字杂音——从牛津逗号（Oxford comma）的用法纠结、破折号（em dash）的争议到基础动词修饰，团队得以摆脱语法苦役，将全部精力聚焦于核心信息提炼、战略洞察与更高价值的客户服务； * **量化收益**：凭借这一先发优势，公司不仅荣膺全美最强 AI 赋能公关机构，还在客户留存率与员工留存率上实现了兼顾与突破。 ### [11:40](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=700s) 推动 AI 落地与转化的关键成功要素 主持人克里斯蒂娜·帕斯卡尔（Kristina Paschall）引导讨论切入企业领导层普遍面临的核心痛点：过去几年大家都在深入接触 AI，也见证了其巨大潜力，但并非所有 AI 项目都能带来回报。在与企业高管沟通时，究竟该如何建立信心、证明 AI 具备可衡量的价值？推动 AI 项目成功落地的核心要素究竟是什么？ 温斯顿·费舍尔（Winston Fisher）率先指出，首要任务是在组织内部找到对的“推动者”（Champion），同时坦然面对内部阻力与技术失败。他直言，如今许多 AI 厂商为了完成销售指标过度承诺，宣称“AI 无所不能、将颠覆一切”，但现实往往并非如此。例如，他们曾引入一套用于处理所有合同和租约的法律助理系统，结果实际效果非常糟糕。这并不意味着不该尝试，而是说明该厂商在销售压力下夸大了能力。费舍尔强调，他们的成功在于组织内部拥有一位深知“AI 是一个演进过程，而非最终结果”的业务领头人。团队必须建立一种认知：接受技术尚不完美、将失败视为迈向正确的必经之路，这是企业内部最关键的心态转变。 --- ### [13:33](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=813s) 从实验到生产：营销提效、清洁机器人与“黄金连接”数据湖 谈及如何跨越实验阶段并将 AI 真正投入生产环境，温斯顿·费舍尔分享了旗下实景娱乐综合体 AREA15 的实际案例。以往在举办活动时，团队需要将音乐艺术家提供的素材手动调整并转换为适配各种社交媒体的不同格式，整个流程枯燥乏味且长达 3 小时；而在上线 AI 标准化处理程序后，该流程直接缩短为 30 秒，团队成员激动得“几乎要落泪”。此外，在他们旗下的商业保洁公司中，AI 也被用于劳动密集型场景，通过“预测性排班”结合机器人技术来实现人力成本节约。尽管商用清洁机器人目前连“移动椅子”这种基础动作都无法在规模化场景中完成，但技术正在稳步推进。 费舍尔进一步指出，除了这些点状的效率提升，对企业最根本、最重要的长线工作是构建底层数据湖（Data Lake）与结构化数据。他将不同业务数据源之间的关联打通称为“黄金连接”（The Golden Join）。虽然构建这种统一数据底座耗资巨大、工程繁重，且目前尚未完全建成，但只有以此为基础，企业未来才能在其上构建专属的定制化 AI 系统。目前，他们正在并行推进短期的生产力工具应用与长期的底层数据底座建设。 - **AREA15 营销素材转换**：从耗时 3 小时的手工切图排版缩减至 30 秒自动化处理。 - **商业保洁自动化探索**：将预测性排班算法与地面清洁机器人结合，逐步降低基础人力依赖。 - **双轨并行策略**：前端快速落地立竿见影的效率工具，后端同步重金投入数据湖治理与“黄金连接”。 --- ### [15:48](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=948s) 医疗领域的 AI 治理：精准选题与“以终为始”的试点设计 约翰斯·霍普金斯大学医学院的刘艾尔文（Alvin Liu）教授从医疗场景出发，提出了 AI 成功的两大关键要素：精准识别真问题，以及设计严谨的试点项目（Pilots）。医疗关乎生命，属于高度受监管的行业，虽然医疗体系中存在大量痛点，但并非所有问题都能或应该用 AI 解决。首先需要兼具深厚 AI 技术与医疗临床背景的专家，从纷繁复杂的需求中筛选出 5 个真正值得且适合用 AI 解决的核心问题。 在试点设计上，刘教授主张“以终为始”（Start Backwards）的逆向推演方法：在启动试点前，就必须将未来全面落地所需的所有核心利益相关者召集到一张桌子上，尤其是掌握财务预算的审批人员。各方必须在第一天就预先确立 ROI 的计算方式与 KPI 指标，并达成契约——一旦试点在预设指标上取得成功并验证了正向 ROI，财务和业务部门必须承诺后续全面采购与推广。 ```mermaid flowchart TD accTitle: 医疗 AI 试点与治理评估流程 accDescription: 展示从内部发起人、AI 治理委员会评估到以终为始确立 KPI 与预算承诺的闭环流程 subgraph Phase1["第一阶段：需求准入与治理"] A["业务痛点筛选<br/>(聚焦少数真问题)"] --> B["寻找内部发起人<br/>(Internal Champion)"] B --> C["AI 治理委员会评估<br/>(AI Governance Team)"] end subgraph Phase2["第二阶段：以终为始的试点设计"] C --> D["全利益相关方就席<br/>(业务/临床/IT/财务)"] D --> E["预先锁定量化标准<br/>(明确 KPI 与 ROI 算法)"] E --> F["获得出资兜底承诺<br/>(达标即全面落地)"] end subgraph Phase3["第三阶段：闭环交付"] F --> G["开展试点验证"] G --> H{"预设 KPI 是否达标?"} H -- 是 --> I["全面预算拨付与推广部署"] H -- 否 --> J["及时止损并复盘迭代"] end ``` 针对全美大型医疗系统普遍兴起的“AI 治理团队”（AI Governance Team），刘教授介绍了约翰斯·霍普金斯大学的标准机制：针对蜂拥而至的 AI 医疗创业公司，医院设立了严格的准入流程。任何外部供应商在接受评估之前，必须先在霍普金斯内部找到认可其价值的“内部业务赞助人/推动者”，随后治理委员会通过多轮质询与辩论开展迭代式评估，杜绝盲目跟风引入无效工具。 --- ### [19:04](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1144s) 嵌入工作流与数据就绪度审计：避免在浏览器标签页中被遗忘 Dualboot Partners 首席营收官康斯坦丁·齐布尔科（Konstantin Tsybulko）提出了评估 AI 价值的极简法则：“如果你无法用一句话向 CFO 解释清楚 AI 项目背后的 ROI，那你大概率做错了。”他分享了一个制造企业的真实翻车案例：该客户斥资打造了一个视觉精美的 AI 仪表板，但员工根本不使用，管理层最初归咎于数据质量差。事实证明数据确实有误，导致 AI 自信地输出了一堆错误结论——“AI 能够极大地放大错误的数据或流程”。 齐布尔科指出，除了修复底层脏数据，更根本的解法在于改变交互载体。他们将该 AI 仪表板直接深度嵌入到员工每天使用的 Salesforce 系统中。他强调：“**停留在独立浏览器标签页中的应用程序是不会被使用的**（Apps that live in tabs do not get used）。”AI 必须深度内嵌到团队早已熟悉的原有工作流中，才能确保使用频次并发挥员工的主动协同价值。 面对克里斯蒂娜关于“供应商是否有责任在客户数据未就绪时踩刹车”的提问，齐布尔科表示这在他们的日常合作中非常普遍。Dualboot 在开展 AI 项目时，通常会以“AI 就绪度审计”（AI Readiness Audit）为起点，首先审查并清洗数据，确保项目从干净数据支撑的小切口切入，因为数据质量直接决定了后续的用户采纳率与项目成败。 --- ### [21:41](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1301s) 应对员工失业恐慌：公关传播公司的 AI 变革与布道 格雷戈里传播（Gregory FCA Communications）创始人兼 CEO 格雷格·马图斯基（Greg Matusky）分享了在知识服务型企业推行 AI 的独特挑战。当时各大行业报告普遍指出，公关与市场营销是受 AI 冲击最大、最可能发生大规模失业的第一大行业类别。当一家 140 人的公关公司老板突然宣布“我们要全面实施 AI”时，整个团队内部爆发了巨大的恐惧与阻力。 与其他企业由下而上或指派项目负责人不同，马图斯基在公司内部亲自扮演了“AI 布道者”的角色。他深知这项技术能为机构带来怎样的生产力跨越，但首要任务是向产生抵触情绪的员工做好沟通与预期管理，将 AI 重新定义为赋能员工的工具而非替代者，从而促使全员从恐惧防御转向主动拥抱。 ### [22:31](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1351s) 打造“超级创作者”：以员工职业发展驱动的 AI 组织变革 Greg Matusky 阐述了他推行 AI 时的核心理念：企业引入 AI 不应单纯为了追求生产力提升，而是要“为了员工的切身利益”，让他们的职业生涯更具生命力和竞争力。他反驳了“你的工作不会被 AI 取代，只会被懂 AI 的人取代”这种老生常谈，指出真正的竞争威胁来自于**“超级创作者”（Super Creator）**——即那些彻底驾驭了 AI 工具、能够以惊人速度持续产出海量高质量内容的人。为了消除员工对技术替代的恐惧，他向团队给出 100% 的承诺，帮助全员掌握这些工具以捍卫职业生涯。 这种转变本质上是一场**行为管理变革（Behavioral Management）**。公司营造了一种积极庆祝成功、包容失败的文化氛围，并在两年内开展了多达 102 场 AI 培训课程（几乎保持每周一次的高频节奏）。通过全员共享工作流与提示词经验，团队内涌现出一批新的 AI 布道者。由于创作者本身具备灵活的发散思维，而非程序员那种非黑即白的二元逻辑，团队很快适应了“同一个创意任务可以有 20 种不同 AI 求解路径”的不确定性，形成了全员主动拥抱 AI 的正向势头。 ```mermaid flowchart TD accTitle: AI推动员工赋能与企业价值增长闭环 subgraph Enablement["赋能与文化重塑"] A["高频培训 (2年102场)"] --> B["消除恐惧：承诺保卫职业生涯"] B --> C["从单一布道者到全员分享"] end subgraph Shift["工作重心的结构性转移"] C --> D["压缩繁琐杂务 (排版/校对)"] D --> E["聚焦核心产出与客户/媒介深度关系"] end subgraph Impact["可量化的业务价值"] E --> F["生产力提升 20%~26%"] E --> G["客户流失率下降 34%"] E --> H["员工留任率提升 26%"] end ``` - **培训规模**：两年内密集举办 102 场内部 AI 培训课程（平均每周 1 场）。 - **生产力跃升**：首年实现 13.3% 生产力增长（公司营收规模约 2400 万美元，直接增厚约 30% 底线利润），最新统计数据已进一步提升至 20% 至 26%。 - **客户与留任指标**：客户流失率（Client Attrition）大幅降低 34%，员工留任率（Employee Retention）提升 26%。 --- ### [25:59](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1559s) 衡量无形成果：告别“红线批注”与员工自尊心的重塑 在回答关于追踪指标的提问时，Greg Matusky 强调，公司虽然一直按月监控基础指标，但最深刻的变化在于**工时结构的根本性转移**。过去那些琐碎、枯燥且最容易导致员工心生离意并辞职的后勤杂务被大幅压缩，省下的时间被转移到真正能对客户产生冲击力的创意生产以及与媒体、客户的深度关系建立上。 更关键的是技术带来的**无形成果——专业自尊心（Self-esteem）**。在传播公关行业中，最打击员工士气的是收到客户退回满篇修改意见的“红线批注（Redlines）”。Greg 坦言，即使自己已经 64 岁，看到客户满纸红线依然会感到受挫与痛苦。而借助“人机协同”，团队在成稿阶段就能精准契合客户的语调与受众需求，交付成果的准确率大幅上升，错误率急剧下降。这种高胜率不仅超越了客户预期，更显著提升了员工的自我价值认同感与工作幸福感。 --- ### [27:58](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1678s) 软件开发领域的 ROI 公式：省钱、赚钱与复利加速 来自 Dualboot Partners 的 Konstantin Tsybulko 分享了在向软件客户销售 AI 赋能产品时评估 ROI 的核心逻辑。面对当前市场上充斥的炒作（AI 往往处于“过度被吹捧或严重未被充分利用”的两极状态），他们建立了一套简单清晰的衡量公式： 1. **是否省钱（Saving Money）**：降低交付成本与人力损耗； 2. **是否赚钱（Making Money）**：开拓新业务与收入来源； 3. **是否具备复合加速效应（Moving Faster in a Compounding Way）**：以极高效率缩短研发与销售周期。 为了使方案具备说服力，他们直接用自身业务作为实证：借助生成式 AI，原本需要耗时数天的产品需求文档（PRD）现在仅需 8 小时即可编制完成；而过去漫长的方案验证流程，现在只需通过一次客户沟通（Discovery Call），就能在 48 小时内快速生成可交互的点击式原型（Clickable Prototype）。这种将销售周期和售前验证效率提升数倍的实例，在客户沟通中极具说服力。 --- ### [29:11](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1751s) 医疗体系的三维 ROI 框架与非盈利使命 约翰斯·霍普金斯大学医学院的 Alvin Liu 教授系统阐述了医疗健康场景下评估 AI 项目的“三维 ROI 框架”： | ROI 维度 | 核心衡量目标 | 积极成效的具体体现 | |:--- |:--- |:--- | | **临床维度 (Clinical ROI)** | 患者预后与健康水平 | 疾病负担减轻、存活率提高、患者整体满意度与幸福感增强 | | **运营维度 (Operational ROI)** | 流程流转与文书效率 | 大幅压缩文书与病历整理负担，有效缓解医护人员的职业倦怠（Burnout） | | **财务维度 (Financial ROI)** | 成本节约与收益产出 | 直接财务回报、资源优化与资金收益 | Alvin Liu 提出，医疗机构在立项与落地 AI 时，理想状态是**在三个维度中至少达成两项正向回报**（例如：临床 + 财务，或运营 + 财务）。但他特别强调，医疗行业的本质是“使命驱动（Mission-driven）”，金钱并非唯一准则。在很多情况下，只要一项 AI 举措能够带来极为显著的临床和运营正向效益，即便无法带来巨额财务利润，只要立项方能证明长期看不会导致灾难性的严重亏损，项目依然应当推进。此外，在罕见病诊疗等特殊领域（例如全球仅有 10 例患者的罕见病救治），即便从商业模型上看注定是负财务 ROI，出于医学人道主义与社会价值，机构也必须坚定投入。 --- ### [31:24](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1884s) 穿透咨询成本：深度研究与 AI 的百万美元省钱法 AREA15 的 Winston Fisher 从自身业务出发，补充了企业削减外部开支层面的 ROI 实战路径。除了内部视频生成和运营流程自动化带来的“降本增效（Do more with less）”之外，他们当前最直接的巨大回报来自于**减少对外部管理咨询顾问（Consultants）的依赖**。 Winston 在企业内部推行了一项硬性规定：在聘请任何外部顾问或机构之前，团队必须先利用 ChatGPT 或 Gemini 等前沿工具的“深度研究（Deep Research）”功能对该项目或议题进行全面推演与架构拆解。尽管使用 AI 进行研究依然需要人类的批判性思维、精准的提示词工程以及专业常识来核验校准，但 AI 在极短时间内清晰界定问题、梳理背景并构建分析框架的能力，帮企业省去了大量前期的外部咨询支出。Winston 坚信，仅通过订阅最高级别的 AI 深度研究工具，就已经在无形中为公司节省了上百万美元的外部顾问费用。 - **立项前置规则**：严控外部顾问开销，要求团队必须先借助 AI 深度研究完成问题定义与框架构建。 - **复合收益结构**：通过自动化内部流程与自研深度调研，双向压缩内部工时与外部采购成本。 ### [33:24](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2004s) 认知退化危机与对业务本质理解的外包风险 Winston Fisher 指出，当人们过度依赖 AI 来替代自身的思考与学习技能时，正在经历一种实质上的“认知退化”（cognitive decline）。这不仅仅是将任务外包给 AI，而是在削弱人类大脑原有的能力。如果企业内部的员工无法真正精通业务、无法透彻理解所要解决的问题本质，那么眼前看似提升效率的短期收益（short-term gain），实际上会演变成未来巨大的“战略赤字”（strategic deficit）。Winston Fisher 强调，他不认为 AI 能够完全取代人类去深刻理解问题并执行解决的能力，因此他向团队敲响警钟：“团队绝不能将对问题的理解外包出去（my team cannot outsource the understanding of the problem）”，而这恰恰是使用 AI 时极易陷入的陷阱。 Kristina Paschall 对此表示完全赞同。她分享了 Madison Air 最近进行的一项针对 AI 就绪度（AI readiness）的 SWOT 分析，其中发现的核心优势之一，正是将 AI 部署给那些对行业本身拥有极其深刻理解的人才。谈及企业落地，Madison Air 在推进 AI 项目时严格划分优先级，重点押注在能够创造营收的机会以及能直接带动业务增长的内部生产力提升上。面对海量涌现的 AI 应用构想，公司借助关键绩效指标（KPI）体系对需求进行筛选，坚决贯彻“聚焦关键少数”（focus on the critical few）的策略。 ```mermaid flowchart TD accTitle: AI使用中的双刃剑效应 subgraph Risk["AI过度依赖的陷阱"] A["将核心思考外包给AI"] --> B["人类理解问题能力退化"] B --> C["形成战略赤字 (Strategic Deficit)"] end subgraph Solution["正确的高ROI实施路径"] D["行业专家深度理解业务"] --> E["AI赋能高价值场景"] E --> F["KPI量化筛选 (聚焦关键少数)"] F --> G["营收增长与长效生产力"] end ``` ### [35:14](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2114s) 大脑重塑与横向思维：专家专精转向全局战略 Greg Matusky 顺着认知退化的话题展开探讨。他引用了英国伦敦出租车司机的经典神经科学研究：司机为了在极为复杂混乱的城市街道中导航，必须通过极其严苛的资格考试（The Knowledge），这使得他们大脑中负责空间关系的海马体区域显著增长，但当他们退休后，这些神经连接和增长便逐渐消退。类似现象也出现在美式橄榄球（NFL）四分卫身上。Greg 拿传奇四分卫汤姆·布雷迪（Tom Brady）举例：布雷迪虽然身体身体素质有限，但他的大脑能在单个镜头定格中瞬间看清整场进攻布局、所有球员的位置和防守分工，这种卓越的全局视野使他成为历史上最伟大的球员。Greg 提问道，在职业生涯的成熟阶段，自己真的还需要纠结“牛津逗号”（Oxford comma）该放在哪里这类基础细节吗？人们或许在执行细节和后勤琐事上会出现认知退化，但只要将精力转向战略规划、人际连接与商业洞察，就能跨越这一退化阶段。 Alvin Liu 教授进一步从医学与社会演进的角度补充。过去百年来，社会一直极度奖励“超级专精”（hyperspecialize）的人才——无论是视网膜专家、公关专家还是房地产行家。然而随着 AI 的普及，在单一垂直领域获取深度知识的边际成本已变得极低。为了防止认知退化，人类必须转变思考方式，从“垂直专精”走向“横向思维”（lateral thinking），学会跨界连接点滴线索、洞悉全局森林。以医学外科手术为例：如果达芬奇（Da Vinci）手术机器人缝合得比人类医生的双手都要精准完美，那么外科医生究竟应该继续花时间磨练双手缝合技巧，还是应该将宝贵的脑力投入到手术方案规划、整体战略制定以及术后护理管理中？Alvin 认为，AI 带来的真正风险不在于工具本身，而在于“人的惰性”（laziness）——盲目相信机器给出的笃定答案。 - **伦敦出租车司机效应**：高强度空间记忆训练使大脑特定区域增大，但停止使用后相关神经连接显著缩减。 - **汤姆·布雷迪（Tom Brady）案例**：凭借对赛场全局的瞬间解析能力而非单纯的身体素质，登顶 NFL 历史巅峰，展现了从执行层向全局统筹层跃迁的价值。 - **达芬奇手术机器人启示**：当机器在精细化技术执行（如血管缝合）上超越人类时，医生的核心价值转向术前战略布局与术后综合管理。 ### [39:31](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2371s) 劳动力培训的现实挑战与垃圾信息乱象 Winston Fisher 结合自己的公共事务背景提出了清醒的警示。他曾担任前州长 Cuomo 治下的纽约市经济发展委员会联合主席，现任纽约与新泽西港务局（Port Authority of NY & NJ）董事，深谙经济发展与劳动力政策。他提醒大家，高谈阔论“思维升维”在顶层视角听起来很美，但在现实的劳动力发展（workforce development）中，很多基层劳动者并没有顶尖的教育背景，也无法像精英阶层那样使用 AI。对于这部分群体，基础的批判性思维训练必不可少。有时，理解一个逗号为什么放在那里，最扎实的学习方式恰恰是亲身体验被红笔修改打回的痛苦过程。若基础学习与认知强化被 AI 轻易取代，大量面临技术转型的劳动者将被边缘化，这绝非单纯的社会政策问题，而是事关社会经济根基的培训必修课。 Konstantin Tsybulko 紧接着直指行业内因 AI 导致的“思维惰性”乱象。他提到本次 Ai4 大会的官方 App： > “我不确定你们怎么样，但我收到的信息里大概有 70% 完全是垃圾（total crap）。” 大量使用 AI 自动生成的外联消息粗制滥造，发送者甚至连一秒钟的背景调研都不愿意做，既不校对文案，甚至连基本的称呼都写成冰冷的“Hi”而漏掉了名字“Hi Konstantin”。这清晰地暴露了当滥用 AI 且缺乏人类认真输入与校验时，AI 只会加速产出低质内容。 在全场热烈的讨论后，主持人向与会嘉宾及现场观众致谢，宣布本场圆桌论坛圆满结束。 ## Notable quotes > [08:10](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=490s) 如果你不用 AI 来武装和放大自己，你注定会被那些善用 AI 的人所取代——这对个人和企业而言皆是如此。 — 康斯坦丁·齐布尔科（Konstantin Tsybulko） > [11:40](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=700s) AI 是一个演进过程，而非最终结果 — 温斯顿·费舍尔（Winston Fisher） > [15:48](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=948s) 以终为始 — 阿尔文·刘（Alvin Liu） > [19:04](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1144s) 如果你无法用一句话向 CFO 解释清楚 AI 项目背后的 ROI，那你大概率做错了。 — 康斯坦丁·齐布尔科（Konstantin Tsybulko） > [19:04](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1144s) 停留在独立浏览器标签页中的应用程序是不会被使用的 — 康斯坦丁·齐布尔科（Konstantin Tsybulko） > [22:31](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=1351s) 你的工作不会被 AI 取代，只会被懂 AI 的人取代 — 格雷格·马图斯基（Greg Matusky） > [33:24](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2004s) 团队绝不能将对问题的理解外包出去 — 温斯顿·费舍尔（Winston Fisher） > [33:24](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2004s) 聚焦关键少数 — 克里斯蒂娜·帕斯卡尔（Kristina Paschall） > [39:31](https://www.youtube.com/watch?v=QOrvxhmcUUg&t=2371s) 我不确定你们怎么样，但我收到的信息里大概有 70% 完全是垃圾 — 康斯坦丁·齐布尔科（Konstantin Tsybulko） ## Mentioned Madison Air, Broan-NuTone, Big Ass Fans, Kristina Paschall, Winston Fisher, Fisher Brothers, AREA15, Alvin Liu, Johns Hopkins Medicine, Konstantin Tsybulko, Dualboot Partners, Cursor, Figma, Copilot, Greg Matusky, Gregory FCA, ChatGPT, Gemini, Perplexity, Malcolm Gladwell, Salesforce, Tom Brady, Da Vinci, Andrew Cuomo, The Port Authority of New York and New Jersey ## Video description Alvin Liu, Endowed Professor, AI Oversight Team, and Inaugural Director of the James P Gills Jr MD and Heather Gills Artificial Intelligence Innovation Center, Johns Hopkins Medicine Winston Fisher, Chief Executive Officer, AREA15 & Partner, Fisher Brothers, Fisher Brothers Greg Matusky, CEO & Founder, Gregory FCA Communications Konstantin Tsybulko, CRO, Dualboot Partners Kristina Paschall, CIO, Madison Air Abstracts This session will focus on analyzing real-world case studies and projects where generative AI has delivered a measurable return on investment (ROI). Panelists will share their experiences with specific AI deployments, discussing which projects demonstrated the most significant ROI and the factors that contributed to their success. Learn how to identify use cases with the highest potential for financial and operational impact, and gain insights into the metrics and approaches used to track AI-driven value. Panelists will also share best practices for optimizing generative AI initiatives to maximize ROI and create long-term business benefits. [Source](https://www.youtube.com/watch?v=QOrvxhmcUUg)

[8:10] 康斯坦丁·齐布尔科（Konstantin Tsybulko）: 如果你不用 AI 来武装和放大自己，你注定会被那些善用 AI 的人所取代——这对个人和企业而言皆是如此。 [11:40] 温斯顿·费舍尔（Winston Fisher）: AI 是一个演进过程，而非最终结果 [15:48] 阿尔文·刘（Alvin Liu）: 以终为始 [19:04] 康斯坦丁·齐布尔科（Konstantin Tsybulko）: 如果你无法用一句话向 CFO 解释清楚 AI 项目背后的 ROI，那你大概率做错了。 [19:04] 康斯坦丁·齐布尔科（Konstantin Tsybulko）: 停留在独立浏览器标签页中的应用程序是不会被使用的 [22:31] 格雷格·马图斯基（Greg Matusky）: 你的工作不会被 AI 取代，只会被懂 AI 的人取代 [33:24] 温斯顿·费舍尔（Winston Fisher）: 团队绝不能将对问题的理解外包出去 [33:24] 克里斯蒂娜·帕斯卡尔（Kristina Paschall）: 聚焦关键少数 [39:31] 康斯坦丁·齐布尔科（Konstantin Tsybulko）: 我不确定你们怎么样，但我收到的信息里大概有 70% 完全是垃圾

[2:41] Winston Fisher: significantly cheaper, significantly faster, [and] significantly better. [2:41] Winston Fisher: absolutely seismic meltdown of creativity on South Rancho Drive [9:25] Greg Matusky: the most joyful day of my career. [11:40] Winston Fisher: Robots still can't move chairs [19:04] Konstantin Tsybulko: If you can't explain the ROI behind your AI initiative to your CFO in one sentence, most likely you're doing something wrong. [19:04] Konstantin Tsybulko: apps that live in tabs do not get used [22:31] Greg Matusky: your job's not going to be taken by AI, but by somebody who knows AI [33:24] Winston Fisher: my team cannot outsource the understanding of the problem and it's very easy to do that with AI. [35:14] Greg Matusky: If you're a surgeon and you need to have technique right with your left hand as well as your right hand, you develop that skill with your brain. But if you're doing it with a computer, if you're doing it with da Vinci or something like that and it's automated and that machine can suture better than you could ever suture with the left or right, what's better: that you continue to build your suture skills or that you spend your time planning the surgery in the strategy in the aftercare? [39:30] Winston Fisher: Sometimes actually the best way to know why a comma goes someplace is to get a red line back, right? There's actually power in that. So, I think there's a balance to making sure that AI does not supplement the basic learning and reinforcement, and then that you think about how that impacts all the people who are being displaced... the training is imperative for success in society.

[0:04] We'll do some introductions as we kick [0:06] off here. So, um I'll introduce myself [0:08] and then the panel um of guests here who [0:11] all bring wonderful and diverse [0:13] experiences to this uh panel today will [0:15] introduce themselves. I'm Christina [0:17] Pascal. I'm the CIO of a wonderful [0:20] company, Madison Air. We are a [0:23] manufacturer of um many different things [0:26] all related to the movement of air. Uh [0:29] some of our most known brands uh that [0:31] you might know are Bron Newton, a great [0:35] company based in Wisconsin, Bigs Fans, [0:38] another company based in uh Lexington. I [0:41] brought um something from Big Ass. This [0:43] is a not a stress ball, but this is a [0:46] stress donkey. And this is Vanny, our [0:49] mascot from Big Ass. I do have a couple [0:52] of these if you want to introduce [0:53] yourself afterwards. uh and and um and I [0:56] I might be able to share one with you. [0:58] So with that um we're talking about [1:00] generative AI and uh getting maximum ROI [1:03] from generative AI today. And I'll let [1:05] the panel introduce themselves. [1:08] >> Hi, my name is Winston Fiser. I am one [1:11] of the managing partners of my family [1:13] business Fischer Brothers headquartered [1:15] in New York. We're a diversified [1:18] real estate investment development [1:20] operator. I'm also the CEO founder of an [1:24] entertainment um company headquartered [1:26] in Las Vegas called Area 15 which is a [1:30] now 600,000 square foot entertainment [1:33] district uh parallel to the strip right [1:35] by the wind. [1:38] >> Hello. Hello everyone. My name is Alvin [1:40] Louu. I'm a practicing retinal surgeon [1:43] and endowed professor in artificial [1:45] intelligence and Johns Hopkins medicine. [1:47] In addition, um I am on a team called [1:50] the AI oversight team. That is a [1:53] leadership team with purview over all [1:55] things AI related uh at Johns Hopkins [1:57] Medicine. [1:59] Hi everyone, my name is Constantine [2:02] Sabulo. I'm chief revenue officer at [2:05] Dual Boot Partners. We are an AI powered [2:08] digital transformation company. Uh what [2:11] we do is we accelerate software [2:13] development process by means of using [2:15] the best-in-class AI tools like cursor, [2:19] Figma, C-Pilot as well as our own AI [2:22] accelerators to generate over 50% [2:24] efficiency gains. [2:26] I'm Greg Matusky. I'm the founder and [2:28] CEO of Gregory FCA. We are a public [2:31] relations agency which was recently [2:33] named the most AI enabled PR firm in the [2:36] country. We represent many AI firms and [2:39] happy to be here. [2:41] >> Thank you, Greg, Constantine, Elvin, and [2:44] Winston. Okay, the first question I [2:46] have, there's a lot of excitement around [2:49] AI. Uh, but there is a fear of investing [2:53] in the wrong places. So, when did it [2:56] really click in for you and what was the [2:59] clearest signal that you got that that [3:01] made you feel there's a lot of real [3:03] value here? I guess from our [3:05] perspective, it's almost like there was [3:07] that moment in time, right? ChatGpt gets [3:09] released and I was somebody who quickly [3:12] just jumped in. I mean, I I deeply [3:14] believe in the power of AI. Um, decided [3:18] to to sort of become the non- tech as [3:21] best you can AI expert. And so, right [3:24] away, we started embracing it inside [3:26] both Fiser and Area 15. We've seen a lot [3:29] of value uh creative process that has [3:33] probably been an area I'd say that's the [3:35] most tangible and saved us really the [3:38] most amount of money. We we've been able [3:39] to create things um [3:44] significantly cheaper, significantly [3:46] faster, significantly better um in the [3:49] last two years. So it's it's pretty [3:51] amazing. The rest of it I would say [3:54] we've also used it where um and you hear [3:57] this you from everybody like this the [3:59] redundant processes right where you can [4:01] automate something lowhanging fruit [4:04] we're doing it in marketing we're doing [4:06] it in leasing but um we also bump into a [4:10] lot of problems um where until you have [4:13] your data warehouse until you have your [4:15] structured data honestly it's still at [4:18] the margins right and I think that is a [4:20] a period that is a um for me at least [4:23] one of the hardest things inside an [4:25] organization and one of the most [4:27] expensive to actually get correct. So we [4:30] we get a lot of use from it and we also [4:34] have to I still feel we're in the first [4:36] inning. [4:36] >> Okay. Uh before we move on, I'm [4:39] interested in what the rest of the panel [4:41] has to say about this question, but to [4:44] make bring this to life, I hear you have [4:45] an example for us. Yeah. I mean, an [4:47] example of when I say we've saved money, [4:49] there's a video that we just released um [4:52] for the opening of our second phase, [4:53] Area 15 in Las Vegas. And so, we have [4:57] the video. It's on YouTube, but I wanted [4:59] to play it cuz you'll sort of see this [5:00] cost us $3,000. [5:04] >> A new kind of district is emerging just [5:06] parallel to the strip. [5:08] >> We're getting reports of absolutely [5:10] seismic meltdown of creativity on South [5:12] Rancho Drive. Well, John, I don't know [5:15] how Vegas will ever be the same after [5:17] this, much alone the world. [5:19] >> Thank you, Barbara. I've never seen [5:21] anything like it. [5:27] >> The number one immersion destination, [5:29] and no one knows who's really behind it. [5:31] >> Pandemonium ensues as locals and [5:34] travelers are inexplicably drawn to the [5:36] district. [5:36] >> A massive 747 suddenly appears in the [5:40] middle of the district. Grounded Skyways [5:42] Flight 1515 believed to have disappeared [5:45] 40 years ago in the Bermuda Triangle. [5:49] What the hell did I just say? [5:51] >> Restaurants, large scale art, immersive [5:52] experiences, all bursting with A15. [5:56] >> We cannot confirm or deny that this [5:58] district exists, nor that it is of [6:01] earthly origin. [6:05] >> Man, I don't know what that thing is, [6:06] but it ain't from here. Nuh-uh. [6:14] I love that video. But that cost us [6:17] $3,000. And the idea that we could go [6:21] create something like that was unheard [6:23] of. It just didn't exist. So I I don't [6:26] know what that would cost. I actually [6:27] think that'd probably be $100,000 if not [6:30] more. And that's just one example of [6:33] massive efficiency gain and massive cost [6:36] savings that we've been able to embrace [6:38] inside our company. So that's that part [6:40] is really exciting. And then there's [6:42] other process examples I could point to, [6:44] but that's where AI is honestly [6:47] mind-blowing. [6:48] >> Yeah. Remarkable. [6:50] >> Um, so when it comes to healthcare, uh, [6:52] last year we spent about $4.9 trillion [6:56] in this country on healthcare alone, [6:58] which is the biggest sector of the US [7:01] economy, about 80 18% of it. [7:04] Unfortunately, of this $4.9 trillion [7:07] dollar spent, about 25% of it can be [7:11] considered ad administrative waste, [7:13] meaning processes and workflows that are [7:16] very complicated, uh, time consuming and [7:19] labor intensive, but they really don't [7:21] add much to patient outcome. So, one [7:25] could argue that the addressable market [7:27] for waste really is a a$1 trillion [7:30] dollar market. And the good news is that [7:32] a lot of these processes could be made a [7:35] lot more efficient by gener AI and [7:38] specifically within this $1 trillion [7:41] segment about $200 billion are spent on [7:45] revenue cycle management essentially is [7:47] a process through which providers [7:50] doctors and hospitals interact with and [7:53] get paid by um medical um insurance [7:56] companies. So [7:59] specifically within healthcare in [8:01] addressing the waste I see a lot of [8:03] traction using Genai to do revenue cycle [8:07] management. [8:10] >> Christina you asked about the biggest [8:13] fear of investing in the wrong place. [8:17] >> Based on what I see I would actually [8:19] flip this fear a little bit. I think the [8:22] biggest risk in 2025 [8:25] is not investing in the wrong place but [8:29] not embracing AI and not adopting AI. We [8:32] see a huge resistance across the board. [8:35] We see a lot of push back on embracing [8:39] AI. I'm not really a doomer, but I do [8:42] believe that if you do not use AI to [8:45] amplify yourself, you're setting [8:47] yourself up for [8:50] to be replaced by someone who is. And [8:53] that is something that is applicable to [8:55] both people and companies. As far as [8:57] user adoption, I guess like everyone in [9:00] this room is using Chad GBT, Gemini or [9:03] Perplexity. As far as companies, I see a [9:07] lot of talking, but we do not see enough [9:11] companies trying to really embed AI deep [9:14] into how they operate, how they build [9:16] software, how they optimize internal [9:19] operations, um, and how they just, you [9:22] know, run business. [9:25] >> Great, Greg. Well, my aha moment was [9:28] almost four and a half years ago when my [9:30] son who's a engineer out of Carnegie [9:33] Melon told me that my firm faced an [9:36] existential threat and I didn't even [9:38] know what it was. And I said, "What [9:40] could that possibly be? We're one of the [9:42] 40th largest PR firms in the country." [9:45] And he said, "Machines will soon be able [9:47] to write, think, and create as well as [9:49] your best people." And I said, "Uh, [9:52] that's crazy. I'm an artist. I've been [9:55] at this all my life. I can't believe [9:57] that. So, well before Chat GPT, we built [10:00] our own internal uh AI content uh [10:04] platform. We called it Gladwell after [10:07] Malcolm Gladright. And it was kind of [10:10] like uh uh Edison when he said, "What [10:13] what good is a" and they said, "What [10:15] good is a light bulb? What good is a a [10:17] baby?" When it first used it, it was [10:20] very rudimentary. It hallucinated. It [10:21] said we had a mascot that was a green [10:23] frog. But even then I could see that it [10:27] had great promise. We spent the next six [10:29] months immersed in it once Chat GPT came [10:32] out which I call the most joyful day of [10:34] my career because I had struggled for [10:37] many years to teach other people how to [10:39] communicate better and how to write all [10:41] for not. It was a very difficult path. [10:43] If you're not willing to create that [10:45] neuro network in your mind like a [10:47] violinist or an NFL quarterback does, [10:51] writing comes difficult for most people. [10:53] And I saw here was an opportunity to get [10:56] rid of the data, separate it. That's the [10:58] Oxford comma, the M dashes. Oh my god, [11:01] the M dash controversy, the the verbs, [11:04] and really concentrate on the message [11:07] and really concentrate on higher value. [11:10] So, we did that and uh we've been [11:12] recently named the most AI enabled PR [11:15] firm in the country by our lead trade [11:16] association and we've seen incredible [11:18] productivity gains, which I guess we'll [11:20] go into later. If you want me to hold [11:21] those, [11:22] >> we will. I I will share those later. [11:24] Ours is a interesting business so we can [11:27] immediately see where productivity falls [11:29] and client retention, employee [11:31] retention. So I have a case study that I [11:33] think will be very you'll be very [11:34] impressed with. [11:36] >> Awesome. So I think we can all agree [11:38] that on the potential, right? That's why [11:40] we're we're all here. That's why we've [11:42] all been immersing ourselves the last [11:44] couple years. We can agree on the [11:45] potential. We can agree that we've seen [11:47] some real examples. Um, but not all AI [11:51] initiatives pay off. So, you know, [11:53] sitting with our leaders in our in our [11:55] companies or in in the areas of work, [11:58] how do we build confidence in those [12:00] leaders that some of these do really [12:02] have measurable value? And I want to ask [12:05] along those lines, what has been [12:07] critical for you all in executing your [12:10] initiatives successfully? So critical [12:13] success factors [12:17] >> you can go in line. [12:19] >> So for the most part it's finding the [12:22] right champion inside the organization [12:25] finding um there is resistance and [12:28] actually also there's failure and one of [12:32] the one of the challenges that I've also [12:34] seen is many companies come in today and [12:37] they promise you that AI is going to [12:39] transform everything. AI can do this, AI [12:41] can do that. And the truth is it can't, [12:43] right? Like we embraced a whole [12:45] parallegal system that would took all of [12:47] our contracts and our leases and put it [12:49] in and it was bad. Doesn't mean we [12:51] shouldn't do it, but it means that [12:53] program, that vendor was promised [12:56] probably, my guess is because they were [12:57] under pressure to get sales. So our [13:00] success was that we had a champion [13:03] inside the business who understood AI [13:06] was a process and not an end result. and [13:10] and that makes a huge difference because [13:12] right now [13:14] yes you should invest yes you should [13:16] embrace it but you also have to accept [13:18] it's not there and I know that that's [13:21] probably not what everybody and and so [13:23] getting the team that understands [13:26] failure is one step towards getting it [13:28] correct I actually think is one of the [13:30] most important things I've seen inside [13:32] our organization [13:33] >> have you have you been able to move [13:36] beyond obviously in this video you were [13:38] but in other instanc is move beyond [13:41] experimentation to actually to end up [13:43] with a a you know a productionized [13:46] product. [13:47] >> Yeah. I mean look we we used inside [13:48] marketing at area 15 when we were doing [13:51] an event and you would get assets from [13:52] an artist a music artist. It would be a [13:55] three-hour process to convert things put [13:58] it up convert format to different social [14:00] media formats. I I think the team was [14:03] crying when we instituted essentially an [14:05] AI program to standardize that and what [14:08] was three hours of of you know just [14:11] boring work became a 30-cond process [14:14] huge that's where then the next thing [14:17] works really well conversely like we're [14:20] also using I mean to me AI robots [14:22] they're they start is inside we have a [14:24] commercial cleaning company um really [14:27] looking at now and that's labor savings [14:30] how can we start to embrace [14:32] um predictive staffing as well as [14:34] robotics inside of a comp inside of a [14:37] business that's predominantly labor. [14:39] Right? We've started to get success [14:41] there. Now, it sounds good. Robots still [14:44] can't move chairs. I mean, know that [14:46] sounds pretty pretty rudimentary, but [14:47] when you're in an office, they don't [14:49] have the ability to do that um at a [14:51] commercial scale yet, but those things [14:54] are happening. So there we've gotten [14:56] labor savings around certain aspects um [14:58] using that and again predictive staff in [15:01] married with robotics really is [15:03] exciting. So there's all these things [15:05] that that start to show up um that we [15:08] are going full force. I still come back [15:11] to though um I think the biggest thing [15:13] for at least our businesses is the data [15:16] lake um structured data getting [15:18] everything put together in a way that we [15:21] can I call it the golden join but that [15:23] we can actually create connections among [15:25] the various data sources inside our [15:28] business that ultimately we can build [15:30] our own AI system over that we're not [15:32] there that that's over that is a [15:34] tremendous amount of work and expense [15:36] but I think for at least what I've seen [15:38] probably the most important thing we [15:39] could other than these little productive [15:41] productivity gains [15:42] >> and you're doing those in parallel and [15:44] >> we're doing them in parallel it's not [15:45] either or right [15:46] >> okay thank you [15:48] >> um I think there are two critical [15:50] factors first is identify identifying [15:54] the correct problem and the second one [15:56] is doing really good pilots so in [16:00] healthcare healthcare is highly [16:01] regulated and rightly so because [16:02] people's lives are at stake so there are [16:06] a lot of problems in healthcare that we [16:07] want to solve but not all problems s can [16:10] or should be solved by AI. So the first [16:12] step is really have someone with deep AI [16:15] expertise and also understand healthcare [16:18] and say these are the five problems then [16:21] that can and should be solved by AI and [16:23] that's focus on them and the second [16:25] issue um is you know doing good pilots [16:28] and we we're seeing a lot of pilots and [16:31] this is not just AI related pilots and [16:33] oftent times pilots fail because you [16:36] don't have the right people at the [16:39] beginning meaning I think One should [16:42] actually start backwards and imagine if [16:44] you were to have an AI project and if [16:47] you were to do full-scale implementation [16:51] first figure out who the stakeholders [16:53] are and make sure all the important [16:55] stakeholders are at the table at the [16:58] pilot phase from day one. What you also [17:01] need to ensure is everyone has [17:03] predefined KPIs and predefined ways to [17:07] calculate the ROI and you want to get [17:09] buy in from everyone saying that fine [17:12] like now before we start the pilot these [17:14] are the ROI we care about these are the [17:16] KPIs that we care about if the pilot is [17:20] indeed successful improving and [17:23] generating positive ROI and improving [17:26] the predetermined KPIs everyone at the [17:30] table here will be willing to implement [17:32] it afterwards. Uh and that's especially [17:35] important conversation for the team [17:37] that's controlling the financing. You [17:38] want to make sure those people at a [17:40] table from day one and they're willing [17:41] to pay for it if indeed the pilot turns [17:44] out to be successful. [17:45] >> Okay. And and uh is that being governed [17:48] by the sponsor of the project or in your [17:50] role governing AI across the [17:52] organization? Is that uh is that your [17:54] role? uh that is there's a movement [17:57] across very large health systems around [17:59] the country to stand up something called [18:01] the AI governance team essentially it's [18:05] a team of leaders across at the health [18:07] system level that governs really the [18:09] interaction between AI vendors and [18:11] health systems as you imagine you know [18:13] there are a lot of healthcare AI [18:14] companies and very similar to what [18:17] Winston said you know everyone's trying [18:19] to sell Johns Hopkins many different AI [18:21] products right so what we established at [18:24] Hopkins [18:25] is a standardized way to govern [18:27] interaction between AI vendors and Johns [18:29] Hopkins. And one of the important things [18:31] is before they even get evaluated, they [18:33] need to find an internal sponsor and [18:35] champion, someone who actually works at [18:38] Hopkins and thinks this could bring [18:40] value and then they go through, you [18:41] know, a bunch of questions with debate [18:43] and we and then is an iterative process. [18:46] >> Okay. Thank you. So I'm hearing strong [18:48] sponsors, the right people on the team, [18:51] clearly defined KPIs from the start and [18:53] a a clearly defined problem [18:55] >> and a commitment to pay for it if those [18:57] KPIs are met. [18:58] >> Yeah. It has to end up in the end in in [19:01] driving that value. [19:02] >> Yeah. [19:03] >> Thank you. [19:04] >> Ideally, you want to talk about that [19:06] from the very beginning. Like the the [19:07] rule that I apply is if you can't [19:10] explain the ROI behind your AI [19:13] initiative to your CFO in one sentence, [19:17] most likely you're you're doing [19:19] something wrong. Um I'm I'm going to try [19:22] to not repeat what's been already [19:24] shared. I very much agree with all with [19:26] all that. Probably just going to share [19:29] my own like our own anecdote. Uh we work [19:34] with a manufacturing company that [19:36] reached out to us. They have this [19:39] beautiful AI dashboard [19:42] that uh was not used by their team and [19:46] they blamed data and uh the data was [19:49] indeed a problem. We fixed the data and [19:53] uh that was actually a classic example [19:55] where they fed AI with wrong data and AI [19:59] very confidently [20:01] uh reused that wrong data and gave the [20:04] wrong output. So AI can amplify the [20:07] wrong data or the wrong thing very well. [20:10] But what we fixed in this case is we [20:14] embedded this dashboard into Salesforce [20:16] and that would be um a a strong [20:19] recommendation from from from my side [20:22] based on our experience. [20:24] Apps that live in tabs do not get used. [20:28] you need to create something and embed [20:30] it deeply into the workflow that your [20:33] team is already familiar with. This way [20:36] you not only gain visibility over your [20:39] AI initiative but you also get uh or you [20:44] make your team contribute to the success [20:47] of this initiative and that is [20:49] critically important if you want to [20:50] generate some ROI. [20:52] >> A question for you. Um so as somebody [20:54] who buys uh as as a role as CIO I buy [20:58] services from vendors. Do you think [21:00] being you know on the vendor side [21:02] yourself that you have a responsibility [21:05] to tell a company hey you're not you're [21:07] not ready yet to build this AI because [21:09] have you have you done the prerequisite [21:12] data uh cleansing? [21:13] >> We do this all the time. [21:15] >> Okay. [21:15] >> We do this all the time. Typically the [21:17] way we start our AI engagement is is [21:20] with the uh AI. so-called AI readiness [21:24] audit. And what we do first and [21:27] foremost, we review the data and we [21:29] clean it up and we make sure that [21:31] whatever we start with is reasonably [21:35] small and is backed by clean data [21:38] because that affects adoption big time. [21:41] Thank you. And Greg, uh, you know, you [21:44] have a wonderful story in your [21:45] organization about, um, some of the [21:47] success you've had and and what it's [21:49] done for your your company. So, we'd [21:51] love to hear about that. Well, imagine [21:54] that you have a company of 140 people [21:56] and they all read the same studies and [21:58] the one study shows that the one [22:00] industry that's going to lose the most [22:02] jobs due to AI is public relations. [22:06] The number one category in a lot of [22:08] surveys is marketing and public [22:10] relations. And then the owner of the [22:12] company says we're implementing AI. [22:16] So you could see there's a lot of [22:18] resistance. [22:20] And in my company, I was a little [22:21] different in that I was the evangelist. [22:23] I knew what this could bring to us. And [22:25] I had to sell that to my employees. [22:29] And I did that by making sure they [22:31] understood it wasn't about productivity [22:33] gains, which is great. [22:37] It's about being in it for them, making [22:40] their career more viable, whether [22:42] they're with me or they go somewhere [22:44] else in the world. assuring that the [22:46] industry has a future because we have [22:48] people who just don't know AI. That old [22:51] saying, oh, your job's not going to be [22:52] taken by somebody uh this thing with [22:54] somebody who knows AI, right? No, it's [22:57] going to be taken by a super creator, [22:58] somebody who's mastered it and can [23:00] produce tons of content, high quality [23:03] content at an unbelievable velocity. [23:06] That's what's going to safeguard your [23:08] job. That's what I tell my people. My [23:10] commitment 100% [23:12] is to have you master these tools so [23:14] that you'll always have a career and [23:17] that one approach sold it in and it's a [23:20] behavioral change. This is this is a [23:23] behavioral management issue, right? It's [23:25] very easy. You create an atmosphere in [23:27] there where you celebrate success. We [23:30] invest in training. We held 102 AI [23:33] training courses over two years. 102. [23:36] That's every other week or every week. [23:38] Every week we had a training on AI and [23:42] we created more than just myself as an [23:44] evangelist. Then we spawned new [23:46] evangelists and we celebrated when they [23:49] were success and we shared with one [23:51] another and we were open to failures. I [23:53] think some panelists said you're going [23:55] to make mistakes because that's the [23:56] nature of the inconsistency of AI. [23:58] Fortunately, creators are very flexible [24:00] thinkers. They're not like coders [24:03] binary, right? So they understood that [24:05] you can have 20 different approaches to [24:07] an a creative assignment like that and [24:09] still be in good terms. So we got this [24:12] momentum of everyone sharing believing [24:15] this is the one way they could safeguard [24:17] their jobs. In the first year that we [24:20] had implemented it, we saw a 13.3% [24:23] increase in productivity which our [24:25] company will be and this is public about [24:27] $24 million. you know that drops you [24:30] know 30% the bottom line I just updated [24:33] that number it's a 20% productivity [24:36] increase [24:37] right but more importantly is we saw a [24:41] 34% decrease in client attrition and a [24:44] 26% increase in employee retention. [24:49] Why is that employee retention? [24:52] When your employees believe that they're [24:54] getting something they can't get [24:55] anywhere else and that you're really in [24:58] it for them, that sure the business has [25:01] to be successful. We have to see [25:03] creative, but I'm more con I'm more [25:05] committed to giving them the skills they [25:08] need in a new era that is unlike [25:10] anything we've ever seen in the history [25:12] of human communications. Folks, this is [25:14] it. [25:16] They stay with you and clients stay with [25:19] you when you surpass their expectations. [25:23] So no longer do you do a single [25:26] standalone content issue or no longer do [25:28] you leave it to vagarities to have the [25:31] right voice tone or audience. You're [25:34] dead on through the combination of [25:36] humans and machines. And so we've we've [25:40] cut client attrition. We've increased [25:42] employee retention. And all along we've [25:45] seen a 20 it's a actually a 26% [25:49] increase in productivity. [25:51] So that's a real case study that's from [25:54] the from the very guts of this as far as [25:57] revenue per employee. [25:59] >> Did you have to track any new metrics as [26:01] part of this or these were these were [26:02] the metrics that you were already [26:04] tracking? [26:04] >> Well, we've already we've always always [26:06] tracked these things [26:08] >> uh month by month, right? because we're [26:10] thoughtful people, but we do look within [26:13] it. And interestingly, where does the [26:15] shift come? Because we do track ours. [26:18] It's gone away from the trivial [26:21] logistical that gets people to [26:24] quit their jobs, more into the [26:27] production that has an impact in the [26:29] client, and then more in the [26:32] relationship with the client. So those [26:36] tasks that employees hate, they're [26:38] putting less hours into and it's still [26:40] getting done. Those which they aspire to [26:43] is where they're putting their time. And [26:45] then like everything else, in the end, [26:48] it's always about the relationship, [26:50] right? And that's where they're putting [26:51] more of their time in the relationship [26:53] with the media, in the relationship with [26:55] creative, in the relationship with their [26:57] client. And that's that's what AI allows [27:00] you to do on a on on a thoughtful basis [27:03] with numbers. Yeah, it's wonderful to [27:05] hear like that it's showing up the work [27:07] that you're doing with your organization [27:09] and the investment that you're putting [27:10] in your people is really showing up in [27:12] your you know this is um you know should [27:16] be obvious but it's showing up in in [27:18] your the measures um by which you you um [27:20] you track your book the tangible and the [27:23] intangible and one of the intangibles [27:25] that is not seen by AI is selfesteem [27:29] right if you can do your job more [27:31] effectively if you have less errors in [27:34] our business called red lines. You get [27:36] something from a client, it's redlined. [27:38] And I'm 64 years old and as much as I'd [27:40] like to say it doesn't still affect me, [27:42] it still hurts me, right? And if if you [27:45] can meet the client expectation at a [27:48] higher percentage, you gain self-esteem [27:51] as a professional and you feel better [27:53] about yourself. This is a tool that [27:54] allows you to do that like no other I've [27:56] ever seen. [27:58] >> Thank you, Constantine. You're your [28:00] business. So, you're you're building [28:01] tools that help the software industry. [28:03] So how are you talking about uh ROI with [28:06] u externally when uh when you're selling [28:08] the products that you make? [28:10] >> Essentially we are talking about three [28:12] things. We're talking about whether we [28:16] are saving money as a result of this [28:18] investment, whether we're making money [28:21] or if we're moving faster in a way that [28:24] compounds. That is a simple formula that [28:27] we that we apply when talking with [28:29] clients. And typically [28:33] we try to use our own business as an [28:36] example since I very much agree there is [28:39] a lot of hype around AI. It's either [28:41] like overhyped or underused. So we try [28:44] to tell them like hey we now produce [28:48] project requirements documents within [28:50] eight hours which took us like couple of [28:54] days before. We now can convert a [28:56] discovery call into a clickable [28:58] prototype within 48 hours. And that is [29:01] just an example of how we can optimize [29:03] our sales cycle. And that is an example [29:05] that you can apply to to your business. [29:08] And these things resonate very very [29:10] well. [29:11] >> Great. Thank you. And Dr. Lou, how how [29:13] are you talking about ROI within John [29:14] Hopkins? You mentioned a little bit [29:16] earlier, but we'd love to hear more [29:17] about what ROI looks like. [29:19] >> Sure. Um, so within medicine, I think [29:22] there are three big categories of ROI. [29:24] First one is clinical. Second one is [29:26] operational. Third one is financial. A [29:29] positive ROI in the clinical realm [29:31] really means a better outcome for [29:33] patients, right? They they live better, [29:35] they're less sick, they just happier [29:37] overall. When it comes to operational [29:39] ROI, a positive one means increase [29:43] efficiency for a lot of the paperwork. [29:46] And this is important to decrease staff [29:48] burnout. And the third one is financial, [29:50] which is pretty self-explanatory. [29:52] I think in order to implement um AI [29:56] initiatives and maximize their success, [30:01] ideally whatever you're proposing should [30:04] be able to demonstrate a positive ROI in [30:07] at least two out of the three [30:08] categories. So for example, a positive [30:11] positive clinical ROI plus financial a [30:14] positive operational ROI plus financial. [30:17] That's I think ideal. However, medicine [30:21] is a little bit different because we are [30:23] also missiondriven, right? Not [30:24] everything is about money. So there are [30:27] circumstances where you want to [30:28] implement something that you think can [30:31] generate a positive clinical and [30:33] operational ROI. In those situations, I [30:36] still think it's incumbent [30:38] um on the person who's proposing it to [30:42] at least demonstrate that in the long [30:44] run, it's not going to be a massively [30:46] negative ROI. [30:49] Now there are exceptions to the to this [30:51] exception which is there's certain areas [30:53] of medicine no matter how you frame it [30:56] is always going to be a negative [30:58] financial ROI for example um coming up [31:01] with um treatment for very very rare [31:04] diseases 10 people in the world for [31:06] example those are the rare exceptions [31:08] where as a society or as institution we [31:12] say this is important to us as humans [31:15] this is important to us so let's do it [31:18] even though um financially I'm not going [31:20] to generate a net um a posit ROI but I [31:22] would say those are the exceptions. [31:24] >> Yeah. Great. Great. Thank you. Anything [31:26] else Winston you want to mention around [31:28] ROI? [31:29] >> Well I I think everybody's said right. [31:31] We look at [31:33] staff efficiency you can sort of see the [31:36] video. So we measure that. The other [31:38] area that we're starting to go much [31:40] deeper into was also cons uh lack of [31:43] avoiding consultants frankly [31:46] >> and and what I I think we still need [31:48] consultants just to be very fair but um [31:52] I've just instituted a rule that before [31:54] we hire anybody we have to use deep [31:56] research to to research the project or [31:59] research the issue. Not that it's always [32:02] right. You could by the way you have to [32:03] have critical thinking. You need to [32:05] understand it. you have prompt [32:06] engineering, all of that stuff. But the [32:09] idea that you can start to use AI, you [32:12] know, weather chat, GPT, Gemini or [32:14] whoever to articulate a problem and and [32:19] think about it, deep research saves us [32:22] so much time and money, right? It helps [32:24] frame the issue and that is actually [32:25] massive savings. I am convinced and I [32:28] have not tracked it but that I have [32:30] saved a million dollars by subscribing [32:32] to the highest level of chat GBT for the [32:34] deep research. [32:37] >> So between things like the video things [32:39] like operationalizing [32:41] processing inside the company where you [32:43] can do more with less and then also [32:46] outside fees and starts to become pretty [32:49] pretty impactful. [32:50] >> Great. Thank you. Um, yeah, what I what [32:53] I heard from that one was a lot of uh [32:55] specific measurable KPIs, but also uh I [32:59] like the comments around the intangibles [33:01] as well and that um teams are still [33:04] having the conversation around really [33:06] the the human impact that this [33:07] technology can make and I'm glad that's [33:10] still a part of the conversation. Sorry, [33:12] I want to So one thing the other way [33:14] though is and I was just talking to [33:16] somebody about the research is that the [33:19] risk with AI inside your team also [33:22] though is that there is they're talking [33:24] about there's cognitive decline that [33:27] people experience when they're actually [33:29] using AI to supplement their thinking [33:31] skills and their learning skills. So [33:33] it's not just that they're actually [33:35] outsourcing it to AI. You're actually [33:38] creating cognitive decline. Now if that [33:40] is happening inside your business [33:42] because the people are not actually [33:45] proficient at understanding your [33:48] business and understanding the problem [33:50] what you for me one of my concerns is [33:53] what looks to be a short-term gain [33:56] actually creates an incredible strategic [33:58] uh deficit for us to go forward because [34:02] I actually don't think AI is going to be [34:04] able to replace human knowledge to be [34:08] able to execute and and understand the [34:11] problem. So I I I caution that's one [34:13] thing where I still feel very strongly [34:15] that [34:16] >> my team cannot outsource the [34:20] understanding of the problem and it's [34:22] very easy to do that with AI. [34:25] >> Agree. Yeah. Uh actually uh at my [34:28] organization we were just doing a SWAT [34:29] analysis on our AI readiness and one of [34:32] the strengths we noted was [34:36] um deploying AI but the potential of [34:38] deploying AI to those that really [34:39] understand the the industry very deeply. [34:43] Um I also at at our organization Madison [34:46] Air I would say like we're we are [34:47] definitely prioritizing initiatives um [34:50] revenue generating opportunities and [34:53] productivity enhancements internally [34:56] that then um that then drive uh [35:01] increased revenue. So we're able to [35:03] definitely use um KPI metrics um to [35:07] prioritize the initiatives because they [35:08] are voluminous and we're also trying to [35:10] focus on the critical few. [35:14] >> Yeah, I just going to address one thing [35:15] that Winston said about cognitive [35:17] decline. I think this was very important [35:19] and I consider this a lot. I think about [35:21] this a lot. Um, you know, there was a [35:23] study done on taxi cab drivers in in [35:26] Britain and they used to have to go [35:27] through a very intensive testing process [35:31] um to find because it's a very chaotic [35:33] laidout c uh city uh to find a location [35:37] and that one part of their brain it grew [35:41] the connections grew that was spatial [35:44] relationships right um parts of their [35:47] brains didn't grow and actually shrunk [35:50] because they were concentrated so much [35:52] on passing this test. When they retired, [35:55] they lost a lot of that growth and [35:57] connection. And I and I thought I [35:59] thought about that a lot and I it [36:01] applies to sport, too. NFL quarterbacks [36:04] show the same sort of uh experience and [36:06] brain chemistry by understanding the [36:09] play, understanding where everyone [36:11] should be. [36:13] But NFL quarterbacks careers are pretty [36:15] short with regard to the actual [36:17] technique of that. They go on to become [36:20] coaches and GMs at a very higher level. [36:24] And I always think, do I really need to [36:26] know at this point in my career about [36:28] the Oxford comma, where it goes? Do I [36:31] really need to know about all that goes [36:33] into the data of this stuff? Or do I [36:36] need to be uh Tom Brady, a man with very [36:39] limited physical capabilities, the [36:41] greatest quarterback of all time who [36:42] grew a brain that could see that could [36:44] understand entire football play just by [36:47] one frame, could tell you where everyone [36:50] was, where everyone's assignment was, [36:52] and that's what made him a g great [36:54] coach. So I guess my point is I do think [36:57] we're going to show cognitive decline in [36:59] a lot of the things that were perhaps [37:02] very I don't know logistical in what we [37:05] do but I think we can overcome that by [37:08] focusing on the strategy the [37:10] relationships the human connection that [37:13] really drive business forward that [37:15] that's my theory. I I think Greg has a [37:18] good point and I like to add to is I [37:21] think in the past hundred years or [37:22] societies have really rewarded people [37:25] who hypers specialize. So you're a [37:27] doctor, you're a retina specialist, you [37:29] do PR really well, you do real estate [37:31] really well. So really society rewards [37:35] specialization. [37:36] With AI the acquisition of knowledge is [37:39] relatively cheap now along one vertical. [37:42] You can go deep really quickly and [37:45] cheaply. What that means is I think I [37:48] hope to prevent this cognitive decline [37:51] we need to think differently and that is [37:54] to go lateral really. I mean to a [37:57] certain extent AI can do that too but I [37:59] think it elevates our thinking. Not only [38:02] we need to know our specialty really [38:04] well but now humans really should be [38:08] thinking differently which is really [38:09] connecting the dots understanding the [38:11] forest and be a lateral thinker. I think [38:14] that's where the next level should be [38:16] and in terms of training and workforce [38:19] development that's what we also need to [38:21] focus hey like it's so easy for you have [38:23] to have a good deep research in this re [38:26] real estate problem now but I wanted to [38:28] go one step further and think about the [38:30] adjacent vertical and and just to pull [38:33] an example from the medical field you [38:35] know if you're a surgeon and you need to [38:38] have technique right with your left hand [38:41] as well as your right hand you develop [38:43] that skill with your brain. But if [38:45] you're doing it with a computer, if [38:46] you're doing it with Da Vinci or [38:48] something like that and it's automated [38:50] and that machine can suture better than [38:52] you could ever suture with the left or [38:54] right, what's better that you continue [38:56] to build your suture skills or that you [38:58] spend your time planning the surgery in [39:01] the strategy in the afterare? [39:04] I mean, these are where I think our [39:06] brains will shift some. And I don't [39:08] necessarily think that's a bad thing. I [39:10] think the real risk is is laziness that [39:12] people will just assume that the machine [39:14] is right because it's so affirmative. [39:17] But I think if your if your goal is to [39:19] become a more competent professional, I [39:22] think it can relieve you of those tedium [39:24] and those logistical tasks and really [39:26] allow to focus on the strategic and the [39:28] vision. Sorry, it my other job I do um I [39:33] I was co-chair of New York City's [39:35] economic development [39:37] council for Governor Cuomo and now I'm [39:39] on the board of the Port Authority of [39:41] New York, New Jersey. So very steeped in [39:44] economic development policy. And so I I [39:46] do I would caution some of these [39:48] conversations. They sound good at a high [39:50] level, but when you are actually doing [39:52] workforce development, when you are [39:55] working with with people who do not have [39:57] access to the same educational levels, [39:59] who do not have access to AI the way [40:02] we're describing, you do have to [40:04] actually have training. You do need to [40:05] have critical thinking skills that are [40:07] that are reinforced because you're [40:10] seeing it today. So, I do think [40:11] cognitive decline [40:13] it I wouldn't gloss over it that you [40:16] just pick up another skill. you do, but [40:19] if you don't understand what's going on, [40:21] sometimes actually the best way to know [40:24] um why a comma goes someplace is to get [40:26] a red line back, right? There's actually [40:28] power in that. So, I I I think there's a [40:31] balance to h making sure that AI does [40:35] not supplement and and is and doesn't [40:38] supplement the basic learning and [40:41] reinforcement and then that you think [40:43] about how that impacts all the people [40:46] who are being displaced. I don't mean [40:47] that social policy. I mean that economic [40:49] policy like the training is imperative [40:53] for success in society. I very much [40:56] agree on the laziness point that you [40:58] have made. Um I guess you're all using [41:02] the AI4 app and I guess you are you're [41:06] all getting messages and I'm not sure [41:09] about you guys but like 70% of my [41:12] messages are total crap. [41:16] And like it's clear that someone did not [41:19] even take a second to do a little bit of [41:21] research, right? [41:22] >> A tiny bit of research or to like [41:25] proofread their message or to at least [41:28] like say hi Constantine instead of hi. [41:32] So, I think it has a lot to do with AI, [41:37] but also it has a lot to do with like [41:40] who AI is getting input from or if AI is [41:45] getting input from anyone, if AI is [41:47] being used in the first place. [41:50] >> Okay, I think we're going to have to [41:51] close it here. Um, thank you all. I I [41:55] feel really happy to have met you all in [41:57] the last few weeks. So in a happen [41:59] stance sort of way uh we we were able to [42:02] be on this panel together and it was [42:03] really interesting to meet you all and I [42:04] hope we can continue to to chat. Uh [42:07] thank you all for coming.