Cushman & Wakefield Shifts AI Strategy to Outcome-Based Models

Cushman & Wakefield将AI战略转向基于成果的模式 企业级AI的投资回报率（ROI）需要用混合联合开发模式取代通用SaaS许可，并接受持续保持在80%的准确率，而非追求代价高昂的完美数据。 在本次炉边对话中，Colin Campbell与Chris Slovak详细介绍了Cushman & Wakefield如何从严格的“纯采购型”转变为与Unframe合作的混合联合开发模式，该模式具有风险共担的商业框架和前线部署工程团队的特点。他们批评了追求“Token最大化”与Copilot席位许可的做法，强调真正的ROI源于端到端工作流重塑以及业务价值挖掘。Campbell还主张合理设定数据准确率，而非一味追求100%的完美，并解释了接受持续保持在80%准确率的数据如何帮助拥有复杂历史包袱的传统企业加速AI落地执行。最后，Campbell分享了有关早期技术防护栏、前瞻性人才重构以及消除试点疲劳方面的经验教训。 Cushman & Wakefield重新评估了其传统的“纯采购型”模式，因为通用的现成SaaS无法解决商业地产运营数据的复杂性问题。 (0:04) 企业已经习惯于通过公有云提供商共同开发软件，这使得AI合作伙伴关系成为这种协作方式的自然延伸。 (0:04) 真正的战略性AI合作伙伴会通过围绕风险共担、利益共享和业务成果来构建商业模式，从而实现切身利益绑定（skin in the game）。 (4:35) 为了推进复杂的AI项目而不陷入停滞，企业应采用混合模式：内部团队负责交付快速见效的成果，前线部署的合作伙伴团队则攻坚繁重的工程任务。 (7:51) 评估企业AI的成功必须依据硬性ROI和价值挖掘，而不是Token使用量或席位许可普及率等替代指标。 (11:18)

Cushman & Wakefield transitioned from a traditional commercial software buyer to an outcome-aligned hybrid builder by partnering with Unframe, focusing on workflow redesign rather than superficial seat licensing. Colin Campbell and Chris Slovak argue that enterprise AI delivers value when organizations prioritize business analysis, trade unnecessary data perfection for speed, and avoid pilot proliferation. The Hybrid Build-and-Buy Model (0:04) Cushman & Wakefield historically operated as a cloud-native "buy shop," but the emergence of AI prompted a strategic shift. Colin Campbell, CIO of Recurring Revenue, explains that building entirely in-house was impractical given the firm's lack of engineering scale and 125-year history. Off-the-shelf software was equally insufficient to handle commercial real estate data complexities, such as air conditioning costs, cleaning logistics, and physical square footage. To bridge this gap, the firm adopted a hybrid model with AI partner Unframe. Aligned Incentives and Workload Distribution (4:35) Rather than relying on fixed fees or staff augmentation, Cushman & Wakefield and Unframe established a commercial framework based on shared risks and rewards. Unframe forward-deployed engineering teams directly into business units, sharing downside risk if projects failed and upside when measurable value materialized. Campbell notes that this dynamic builds organizational trust through momentum from substantial use cases rather than basic enterprise search tools. Work is split across two tracks: internal developers tackle rapid, high-engagement initiatives to maintain alignment with internal stakeholders, while Unframe handles complex architectural tasks. Chris Slovak highlights that this division lets Cushman & Wakefield retain focus on proprietary domain expertise while tapping scarce external engineering talent. Targeting Hard ROI Over Consumption Metrics (11:18) Campbell critiques using token usage or copilot seat deployments as proxies for AI success, questioning investments like spending $3 million on licenses without equivalent returns. He argues AI initiatives must prove value through reduced time-to-task, lower operational headcount or costs, or top-line revenue growth. Both speakers caution against an "efficiency trap" where isolated tasks within complex roles are automated without addressing the full workflow or protecting the human relationships tied to brand identity. Because code generation is increasingly automated, Slovak notes that the traditional software development lifecycle has flipped. Teams now dedicate up to 50% of project time to upfront design—determining business rationale, target users, and organizational dependencies—rather than active coding. Consistency, Pragmatic Accuracy, and Data Architecture (17:51) Campbell contends that "consistency beats accuracy" in many enterprise AI use cases. While functions like payroll require 100% precision, high-level forecasting and operational allocation can function effectively at 80% accuracy. Applying this principle, Cushman & Wakefield avoids lengthy master data management overhauls to create a single "golden record" across systems inherited from at least 37 acquisitions, choosing instead to utilize data where it resides to accelerate execution. Operational Guardrails and Curbing Pilot Fatigue (20:34) Reflecting on 18 months of implementation, Campbell emphasizes three operational lessons: deploying technical guardrails early, proactively adjusting workforce roles and talent strategies as duties evolve, and aggressively eliminating pilot fatigue. Campbell advises tech leaders to stop running dozens of concurrent vendor pilots and instead focus resources strictly on executing two high-value initiatives at a time. Campbell and Slovak conclude that successful enterprise AI adoption depends on front-loaded design, clear guardrails, and disciplined prioritization. Organizations must balance partner-driven technical execution with internal domain ownership, while accepting directional accuracy to avoid stalling in costly data centralization efforts.

Cushman & Wakefield的Colin Campbell与Unframe的Chris Slovak指出，大型传统企业的AI转型必须打破纯采购或纯自研的思维定势，转向风险与收益共担的“混合共建”模式，通过端到端流程重构追求硬性业务回报，并以“一致性优于精确度”的数据务实观加快落地步伐。 告别纯采购与纯自研：风险共担的混合共建 (0:04) Campbell表示，拥有125年历史的Cushman & Wakefield在过去近20年间一直是严格的“云原生纯采购”企业，内部缺乏大规模软件工程开发团队。然而，商业地产领域蕴含着极高的数据复杂度，从大型宴会厅的空调能耗、照明、实际面积到保洁物流与座椅排布等变量，均直接决定人员配置与空间优化模型，通用型现成SaaS根本无法满足需求；若全盘转向内部自研，又会受制于人才瓶颈和交付周期。为此，企业选择与Unframe结成深度战略伙伴。Slovak补充认为，正如绝大多数企业依赖AWS或Azure云基础设施而非自建机房一样，企业级AI开发本质上也是一种生态合作。 双方打破了传统的固定费用或人力外包（staff augmentation）模式，引入深度嵌入业务部门的前置部署工程团队（Forward-Deployed Engineering Teams），并实行“利益共享、风险共担”的商业机制。当项目创造可衡量的商业价值时双方共享收益，若未达预期则共同承担损失。Campbell指出，通过率先攻坚实质性业务痛点而非浅层的内部搜索工具，配合对成本和利润的深度透明化，合作双方得以建立高度互信，并形成加速推进后续复杂项目的“滚雪球效应”。 击破效率陷阱：聚焦底层ROI与端到端工作流重塑 (11:18) 针对企业界盲目追求Token消耗量或软件席位发放数量的现状，Campbell提出了尖锐批评。他直言，如果一家企业斥资300万美元全员铺开Copilot席位，却没有产生与之匹配的净收益，这种投资便毫无实质意义。真正的AI价值必须直接锚定在硬性ROI上：具体表现为任务耗时的缩减、运营人数与人力成本的压缩，以及企业顶层营收的直接增长。他同时提醒企业警惕“效率陷阱”，即切勿把ROI寄托于对混乱现有流程中两三个孤立任务的自动化上——例如仅自动化客服人员的碎片化打字环节，却完全忽视了其接听电话、维系客户关系与处理异常案例的综合职责。 Slovak对此表示认同，指出盲目分发软件席位以自动化零散杂务无异于“给猪抹糖霜充当蛋糕”。双方强调，企业必须推动端到端的工作流重塑，严格划定并保护人类擅长的领域专业知识、客户关系维系以及品牌声誉管理，将AI定位为赋能工具而非简单的替代机制，避免因粗糙的自动化交互（如体验不佳的航司改签机器人）而损害企业品牌形象。 开发周期倒置与“一致性优于精确度”的数据务实观 (16:41) 随着AI自动生成代码让开发耗时与成本大幅下降，传统软件开发流程被彻底重构。Slovak分析指出，过去企业通常将25%的时间用于设计、50%用于编写代码、25%至30%用于测试部署；而在如今的AI项目中，高达50%的项目周期被前置分配至设计阶段，用于在动工前严格论证ROI衡量指标、实际使用人群以及伴随技术而来的组织依赖性重构。这要求传统企业适应开发前置的新节奏，摆脱对旧有Jira工作流与Epic管理模式的机械依赖。 在数据架构层面，Campbell提出了“一致性优于精确度（consistency beats accuracy）”的观点。他以Zillow房产估值作比：用户在初探社区时只需方向性的估算而非绝对数值；尽管薪资发放或CNN广播姓名必须达到100%精确度，但在面对高层业务预测和资源配置时，80%的统计一致性已足以支撑有效决策。对于经历过至少37次并购、积累了复杂历史数据的Cushman & Wakefield而言，刻意追求90%乃至100%的边际精度成本极其高昂。相比耗时耗资进行大规模主数据管理（MDM）以建立单一“黄金记录”，让数据保留在原位、利用统计置信度进行匹配，能让大型老牌企业以更快的速度交付成果。 组织落地反思：尽早确立技术护栏与终结试验疲劳 (20:34) 回顾过去18个月推进AI转型的历程，Campbell总结了企业IT管理层的关键落地经验。首先，必须在早期建立底层的技术防护规则（guardrails），他将其比喻为汽车的倒车摄像头——在未配备前可能难以察觉其必要性，但一旦启用便成为无法割舍的安全基石。其次，企业必须从第一天起就启动人才战略与岗位重构，正视AI对各层级技术团队职能的深刻改变，及时开展人员技能转型与角色调配。 最后，Campbell告诫企业必须坚决规避“试验疲劳（pilot fatigue）”。在早期技术热潮中，外部供应商推销的各类试用项目极易导致企业同时开展数十个并行的概念验证，这不仅稀释了技术团队的精力，也造成组织专注度的涣散。Campbell认为自身当时本应更早关停那些杂乱的试用项目，将全部资源严格聚焦于扎实做好两项核心首发任务，确保集中力量形成突破。 该对话呈现了传统重资产企业在AI落地中抛弃概念炒作、转向工程实用主义的典型路径。然而，在免除全面数据清洗、接受“80%精确度”的前提下，企业未来如何平衡潜在的累积误差与合规风险，并将其稳妥推广至更高风险的自主决策场景，仍是有待长期检验的现实课题。

## Background 该视频由 YouTube 频道 Ai4 于 2026 年 8 月 22 日发布，记录了高纬环球（Cushman & Wakefield）经常性收入业务 CIO Colin Campbell 与 Unframe 全球现场 CTO Chris Slovak 的炉边对谈。内容围绕企业在自研、外购与协同共建之间的战略抉择、告别算力消耗陷阱转向真实价值提取，以及传统 SaaS 格局在 AI 浪潮下的重塑展开。 ## TL;DR 对话深入探讨了拥有 125 年历史的商业地产巨头高纬环球如何通过与 Unframe 建立利益绑定的“协同共建（Co-Build）”模式打破纯自研与纯外购的两难困境。嘉宾指出，企业应摆脱“代币狂热”与浅层自动化陷阱，将 50% 的研发精力前置于业务设计与流程重塑，并在方向性决策中接受“一致性优于绝对准确性”以终结沉重的数据治理执念。Colin 还基于 18 个月的实战经验，强调了前置安全护栏、加速人才转型以及果断叫停泛滥概念验证（PoC）的重要性。 ## Atmosphere & style 整场对谈呈现出务实、坦诚且极具一线实战感的基调，完全剥离了泛泛的营销话术与 AI 炒作。两位嘉宾展现出高度默契与平等的研讨姿态，在直面百年老企技术债务、算力浪费与组织转型的阵痛时言辞犀利、毫不避讳。听感上像是一场资深技术操盘手之间的复盘推演，既有打破行业惯性思维的交锋，又充满了在复杂商业现实中淬炼出的从容与定力。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=4s) 高纬环球作为拥有 125 年历史且以纯外购 SaaS 为主的商业地产巨头，在生成式 AI 时代打破了非建即买的二元困境，开辟出与外部伙伴协同共建（Co-Build）的新路径。 - [00:04](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=4s) 商业地产对能耗、空间利用等微观运营数据高度依赖，开箱即用的通用 SaaS 难以满足定制需求，推动企业寻求能深度参与战略制定的共建伙伴。 - [04:35](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=275s) 共建模式的核心是风险共担与利益绑定（Skin in the game），Unframe 通过派出前置部署团队并绑定实际业务成效，打破了传统的固定工时或算力消耗计费。 - [08:05](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=485s) 企业应采用双轨混合交付模式：内部团队负责快速见效场景与采纳度验证，外部伙伴承担复杂且消耗精力的底层工程，避免内部研发陷入单点死锁。 - [11:18](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=678s) 企业应警惕只囤积算力与席位的“代币沉迷（Token Maxing）”，将 AI 的 ROI 考核回归到缩短周期、降低成本或拉动增量营收的“价值最大化（Value Maxing）”上。 - [11:18](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=678s) 过度追求自动化边缘任务会坠入“效率陷阱”，真正的破局点在于业务流程的重构与专业人才的能力增强，而非单纯用 AI 替换零散人工职责。 - [14:39](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=879s) 随着写代码边际成本趋零，AI 时代的研发重心发生转移，团队需将 50% 的工时投入到前期问题定义、业务流程梳理和商业价值设计中。 - [16:41](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1001s) 在方向性决策和资源预测等高容错场景中，“稳定一致性”往往优于追求绝对准确性，企业无需为从 80% 提升到 100% 支付不成比例的高昂工程代价。 - [18:40](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1120s) 百年老企无需为 AI 落地强行推行耗时数年的集中化主数据清洗（MDM），让数据保留在原处并依赖统计确定性快速交付业务价值更具资本效率。 - [20:33](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1233s) AI 安全与合规护栏如汽车倒车影像般不可或缺，必须在项目启动初期就前置建立，以防范潜在系统性风险。 - [20:33](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1233s) AI 转型中人才体系的重塑必须与技术构建同步启动，应尽早重新定义技术岗位职责并协助团队实现技能转型。 - [20:33](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1233s) 企业应坚决叫停泛滥无序的“概念验证（PoC/Pilot）狂热”，集中全公司资源将两件核心高价值场景做到极致并迅速推广。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=4s) 自研、外购还是协同共建：企业 AI 战略的新路径 Chris Slovak（Unframe 全球现场 CTO）与 Colin Campbell（高纬环球 Cushman & Wakefield 经常性收入业务 CIO 兼数字化体验与全球数据/AI 战略负责人）展开了一场关于企业落地 AI 真实挑战的对话。长期以来，高纬环球作为一家拥有 125 年历史的商业地产巨头，在 IT 策略上一直是一家彻底的“外购（Buy Shop）”企业，几乎全面依托 SaaS 产品并保持云原生架构。然而，随着生成式 AI 的爆发，企业无法直接将摆锤从“完全外购”猛烈摆动到“完全自研（Build）”，否则过去的技术资产与历史积累将被全盘推翻。 商业地产本质上是一个极度依赖数据的行业。从大型宴会厅的空调能耗、照明开销、空间利用率，到保洁排班与桌椅摆放，所有细节都是决定人员配置、空间优化和商业价值主张的关键数据点。现成的开箱即用（Off-the-shelf）SaaS 软件已无法满足这些高度定制化的深度业务场景；但另一方面，由于长期以外购为主，高纬环球内部并未储备庞大的自研工程团队。面对这种两难，Colin 指出企业需要的既不是传统意义上按人头计费的人力外包（Staff Augmentation），也不是泛泛的咨询公司，而是一个能够共同制定战略、伴随团队成长并挖掘实质价值的共建伙伴。 ```mermaid flowchart TD accTitle: Build vs Buy vs Co-Build Decision Flow subgraph Traditional["传统二元选择困境"] A["纯自研 (Build)<br/>工程资源严重受限 / 成本高昂"] B["纯外购 (Buy)<br/>SaaS 泛滥 / 无法满足深度定制数据需求"] end subgraph CoBuild["协同共建混合模式 (Co-Build Hybrid)"] C["核心业务专长与快速见效场景<br/>(内部团队主导)"] D["复杂高负荷工程系统<br/>(共建伙伴驻场赋能)"] end A -.->|演进破局| CoBuild B -.->|演进破局| CoBuild ``` - **云端构建的本质**：当现场被问及有谁在开发软件时使用云提供商时，Chris 指出实际上绝大多数企业已经在借助外部伙伴的基础架构与软件蓝图进行开发，AI 时代的共建不过是这一合作模式的自然延伸。 - **PropTech（地产科技）现状**：行业内充斥着大量碎片化的微型点状解决方案（Micro Point Solutions），企业面临着要么艰难集成、要么全面重构替代的抉择。 --- ### [04:35](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=275s) 风险共担与信任建立：共建合作的商业模式重塑 在探讨如何挑选 AI 合作伙伴时，Colin 强调最核心的考量是“利益绑定与风险共担（Skin in the game）”。如果企业投入了大量时间、精力和资本，外部伙伴就必须在商业机制上承担对等风险。Unframe 通过派出前置部署团队（Forward Deployed Teams），直接深入高纬环球的业务端与技术端，并建立了一种以实际结果为导向（Outcome-driven）的商业结构，彻底打破了传统的固定费用（Fixed fee）或纯算力消耗（Consumption-based）计费模式。 这种模式实现了“成则共享收益，败则共同承担损失”。双方在成本核算、价值衡量和风险上限上保持完全透明，消除了传统甲乙方之间防范“供应商过度榨取收入”的对立心理。通过交付第一个真正具有实质业务价值的用例（而非“花 10 秒就能做出来的企业级搜索小工具”），项目形成了滚雪球般的势头与信任基础。这种在交付速度、坦诚沟通以及共同跨越波折中所建立的真实信任，使得双方能够以极高的效率推进后续更大规模、更高难度的复杂项目。 - **新型商业契约**：摒弃单一的固定咨询费，采用基于业务成效分享的商业框架，使合作伙伴的行动动机与企业最终 ROI 完全对齐。 - **动量效应（Snowball Effect）**：首个硬核用例的成功落地是建立组织内部信心与跨公司信任的关键拐点。 --- ### [08:05](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=485s) 混合交付模式：最大化内部能力与复杂工程解耦 针对处于迷茫期、不知该全盘自研还是外购的 CIO 们，Colin 给出的建议十分明确：“采用混合模式（Go hybrid）”。除了极少数拥有无限工程预算的顶尖科技公司外，绝大多数企业必须面对工程资源有限的现实。高纬环球通过解耦不同复杂度的场景，构建了双轨并行的交付体系： | 交付通道 | 负责主体 | 适用场景与业务目标 | |:--- |:--- |:--- | | **内部敏捷交付** | 内部开发与业务团队 | 中短期快速见效（Quick wins）场景、高频业务互动与采纳度验证，用于激发内部动量与团队能力提升。 | | **外部协同共建** | Unframe 前置工程团队 | 极其消耗底层工程精力、架构高度复杂的长周期项目，防止内部核心开发力量被单一项目全面锁定。 | 这一策略避免了工程团队陷入“单点死锁”，从而能够兼顾业务响应速度与全员采纳度。同时，Colin 与 Chris 一致认为，企业必须牢牢守住自身的行业领域知识（Domain Expertise），打造属于自己的“商标级核心资产与秘方软件（Trademark Software）”，而将底层系统化工程赋能交给专业伙伴。 --- ### [11:18](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=678s) 摆脱“代币沉迷”与“效率陷阱”：以真实 ROI 驱动价值提取 谈到行业内普遍存在的“代币消耗狂热（Token Maxing）”——即企业囤积了大量算力配额与席位授权却不见实际回报的现象，Colin 提出了坚定的“价值最大化（Value Maxing）”主张。他直言不讳地指出，如果一家公司花费 300 万美元为全员购买 AI 席位 License，却未能带来对等的 300 万美元实质业务收益，那么这项投资就是毫无意义的。衡量 AI 价值的标准必须回归硬性指标：**是缩短了周期？削减了成本？还是直接拉动了营收增长？** ``` 理想的 AI 价值模型：投入 $1,000,000 成本 ==> 拉动 $10,000,000 新业务增量营收 ``` 与此同时，企业极易坠入“效率陷阱（The Trap of Efficiency）”。Colin 警示，试图通过优化混乱低效的边缘任务来榨取省钱空间往往得不偿失。一个岗位可能只有 2 到 3 项核心任务容易被 AI 优化，其余十多项涉及人际沟通、客服关怀和复杂协作的职责无法被简单线性替代。过度聚焦于自动化微小任务就像是“给猪抹糖霜硬装蛋糕（Putting a little icing on the pig to make it look like a cake）”，根本没有解决核心商业问题。真正的破局点在于业务流程的重新设计（Redesign）以及对专业人员能力的增强放大（Augmentation）。 - **品牌认知与人类交互**：自动化不能牺牲品牌体验。简单的退改签聊天机器人若无法妥善解决问题，反而会损害客户信任；AI 的设计必须保留甚至凸显企业独特的专业壁垒与服务温度。 --- ### [14:39](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=879s) 从自动化到工作流重构：AI 时代开发周期的根本转变 随着代码生成工具的大规模普及，编写代码的边际成本正在迅速趋近于零。Chris 与 Colin 探讨认为，当前的技术瓶颈已经从“我们能不能把它做出来”彻底转变为“我们到底应该做什么”。这迫使研发团队必须重回老派的商业分析（Business Analyst）与问题定义阶段。 根据高纬环球内部的基准研究，现代软件与 AI 应用的开发周期分布已经发生了剧烈的结构性重塑： ```mermaid flowchart LR accTitle: Development Cycle Shift subgraph TraditionalCycle["传统软件研发周期"] direction LR T1["前期设计<br/>25%"] --> T2["编码构建<br/>50%"] --> T3["测试与发布<br/>25%"] end subgraph ModernCycle["AI 时代新型研发周期"] direction LR M1["问题定义与业务剖析<br/>50%"] --> M2["自动化编码与构建<br/>大幅压缩"] --> M3["校验与采纳部署<br/>敏捷推进"] end ``` 在大部分编写与基础测试工作被自动化取代后，决定系统成败的关键，在于前期耗费 50% 的时间去深入业务腹地，精准厘清真正的痛点、数据流向以及重塑工作流后的商业增量价值。 ### [16:41](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1001s) 研发重心的迁移：设计阶段与“一致性大于准确性”的争论 在探讨 AI 项目交付与落地时，Chris Slovak 与 Colin Campbell 谈到了软件研发流程的根本性颠覆。Colin 指出，在 Cushman & Wakefield 最成功的项目中，团队有整整 50% 的时间都花在“设计阶段”（Design Phase）——明确商业投资回报率（ROI）、确定最终使用者是谁，以及梳理需要同步改变的业务上下游依赖关系。在传统的敏捷开发模式中，团队习惯了围绕 Jira 工作流和 Epic 展开马拉松式的编码，但现在的现实是：“构建（Building）实际上变成了最快的一环。” 这种范式转变要求技术团队彻底打破旧有思维，将精力集中在解决用户采纳（Adoption）的核心命题上。 由此，两人展开了一场在 AI 领域颇具争议的讨论：“一致性（Consistency）是否优于绝对准确性（Accuracy）？” 行业内普遍在争论系统到底需要达到多高的精度，以及企业为了从 80% 提升到 90%、甚至从 90% 提升到 100% 所愿意付出的高昂代价。Colin 用房地产估价平台 Zillow 打比方：用户打开 Zillow 查看自家房屋估值时，这个数字并不需要 100% 精确，只要具有统计上的参考意义，就足以支撑用户做出“是否该去那个社区看房”的决策。 ```mermaid flowchart TD accTitle: 业务场景中准确性与一致性的权衡决策模型 subgraph Deterministic["确定性场景（要求 100% 绝对准确）"] A["薪酬核算 / 财务结算"] B["公开媒体报道与实名发布"] end subgraph Directional["方向性场景（要求 ~80% 稳定一致性）"] C["战略预测与资源分配"] D["市场趋势探测 / 选址决策"] end Deterministic -->|代价高昂 / 必须人工校验| E["严格的集中式治理与 Golden Record"] Directional -->|追求速度与 ROI| F["就地利用数据 / 接受容差加速落地"] ``` Colin 进一步区分了确定性业务与方向性决策的差异：如果处理的是工资单（Payroll），或者像新闻媒体在直播屏幕上打出人名，准确度必须是 100%；但如果是用于资源分配、趋势预测的方向性选择，由于现实商业生态中本就充满动态变量，80% 的稳定一致性就足以推动决策，企业根本没必要为追求虚无的绝对精确而停滞不前。 - **时间分配的巨变**：最成功的 AI 落地项目中，50% 的工时消耗在前期业务设计、ROI 测算与流程依赖梳理上，而非实际写代码。 - **准确性的边际成本**：企业必须评估从 80% 到 100% 精确度所付出的高阶计算与工程成本是否匹配实际商业价值。 - **业务场景分层**： - *高容错/方向性场景*：如资源配置预测、房产估值参考，80% 一致性即可驱动业务。 - *零容错/确定性场景*：如薪资发放、权威事实发布，必须严格保证 100% 准确并经过校验。 --- ### [18:40](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1120s) 终结数据完美主义：百年老企的务实数据观与资本效率 接受“80% 一致性足以驱动业务前行”的理念，直接颠覆了企业传统的数据架构与治理思路。Chris Slovak 指出，如果不需要追求绝对完美，企业就完全没有必要开展动辄数年、耗资巨大的数据集中化清洗项目，也无需为了建模而强行把所有数据搬迁到底层数仓，而是可以“让数据保留在原处”（Leave it where it lies），在已知存在微小误差的前提下直接开展业务赋能。 ```mermaid flowchart LR accTitle: 传统数据治理与 AI 时代务实数据策略对比 subgraph Legacy["传统重型治理路线"] L1["分散在37+历史系统的数据"] --> L2["耗时数年的大型 MDM 项目"] L2 --> L3["追求完美唯一真实源 (Golden Record)"] L3 --> L4["滞后交付业务价值"] end subgraph Modern["现代敏捷 AI 路线"] M1["现状数据就地保留"] --> M2["统计确定性与概率匹配"] M2 --> M3["快速推向市场驱动业务变革"] end ``` Colin 对此深有感触，并坦言自己对那些成立仅 3 年、毫无历史包袱的“AI 原生公司”心存羡慕。相比之下，Cushman & Wakefield 是一家拥有 125 年历史、经历过至少 37 次兼并收购的跨国庞然大物。面对各分支机构复杂的异构系统，如果执着于搞重型的主数据管理（MDM）工程去追寻所谓的“单一真实数据源”（Golden Record），不仅消耗海量资本与人力，还会彻底拖慢交付节奏。相反，只要算法能够达到合理的统计确定性（Statistical degree of certainty），企业就应果断推进业务，利用节省下来的资源与周期跑出超高市场响应速度。 这一转变不仅是技术层面的取舍，更是与业务干系人沟通模式的重塑。技术负责人必须直截了当地告诉业务部门与终端用户：“我们并不追求完美无瑕，因为推动业务前进、拿到业务成果根本不需要完美。”对于长期崇尚数据精确度的工程师和数据专家而言，承认“适度即可”往往是最难克服的心理关卡。 - **企业包袱的现实考量**：拥有 125 年历史、完成过 37 次收购的大型企业无法复刻初创公司的轻量化路径，必须注重资本配置效率。 - **摆脱 MDM 执念**：放弃对“Golden Record”的教条式追求，转向以统计置信度为基础的快速匹配与交付。 - **干系人预期管理**：在开发初期向业务端明确交付标准，建立“追求可用商业成果而非理论完美”的共识。 --- ### [20:33](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1233s) 18个月AI实践的回顾反思：护栏、人才战略与终结概念验证狂热 在炉边谈话的尾声，Chris 询问 Colin：如果能给 18 个月前刚启动 AI 战略时的自己一句建议，那会是什么？Colin 总结了三条极具实战价值的教训： 首先是**安全护栏（Guardrails）的前置建立**。很多基础护栏看似是“理所当然的常识”，但实际运营中往往容易被忽视。Colin 将其比作汽车的倒车影像：“在拥有倒车影像之前，你从不知道自己有多需要它；可一旦你用上了，你就会感叹‘天哪，没有它我根本没法倒车’。”企业在起步期就必须将合规与安全防护网预置到位。 其次是**人才战略（Talent Strategy）的极速激活**。Colin 反思自己在转型初期过于偏向技术构建与混合方案，而未能同步并更早地推动人才体系重构。随着 AI 的介入，不仅技术领导者的工作职责被彻底改写，其团队下属的岗位定义和技能要求也发生了根本性重塑。未能及早帮助团队应对这一职业转型，是前期走过的弯路之一。 最后是**果断终止泛滥的“概念验证”（PoC / Pilot Sprawl）**。在 AI 热潮初期，供应商和内部团队倾向于对所有技术和场景都开启试点项目，导致大量 Pilot 遍地开花，却极难衡量真实商业价值，分散了组织的注意力。Colin 直言，如果重来一次，他会从第一天起就果断叫停所有无序的零散试点，集中全公司资源与注意力：“只挑两件关键的事情把它做到极致，然后迅速推进（We're going to do two things well and then move on）。” - **护栏机制**：AI 安全与合规护栏如同汽车倒车影像，属于不可或缺的底层刚需。 - **人才重构滞后**：技术岗位的本质已变，人才战略必须与技术架构并行动员与重塑。 - **聚焦突破**：摒弃泛滥无序的 Pilot，集中火力攻克 2 个高价值核心场景，以此作为撬动全局变革的支点。 ## Notable quotes > [08:05](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=485s) Go hybrid. — Colin Campbell > [11:18](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=678s) Putting a little icing on the pig to make it look like a cake — Colin Campbell > [16:41](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1001s) Building is actually the fastest part of it. — Colin Campbell > [18:40](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1120s) Leave it where it lies. — Chris Slovak > [18:40](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1120s) We're not shooting for perfection, because you don't need perfection to drive the business outcome. — Colin Campbell > [20:33](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=1233s) We're going to do two things well and then move on. — Colin Campbell > [04:35](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=275s) Skin in the game — Colin Campbell > [11:18](https://www.youtube.com/watch?v=_HfjPtjPHUQ&t=678s) Token Maxing — Colin Campbell ## Mentioned Chris Slovak, Colin Campbell, Unframe, Cushman & Wakefield, Ai4, Zillow, Jira, PropTech ## Video description Collin Campbell - Business Information Officer / Chief Information Officer - Recurring Revenue, Cushman & Wakefield Chris Slovak - Field CTO, Unframe Everyone has an AI strategy. Far fewer have results. In this candid fireside chat, Cushman & Wakefield CIO Colin Campbell and Unframe Field CTO Chris Slovak get into the realities of enterprise AI adoption across three of the most debated topics in the industry right now: build vs. buy vs. build together, token maxing, and the future of SaaS. Drawing on their firsthand experience as partners, they'll challenge the conventional wisdom that enterprises must choose between building and buying, make the case that the real ROI conversation has shifted from token consumption to value extraction, and share why the SaaS categories that defined the last decade are quietly crumbling. This isn't a product pitch; it's an honest conversation from two practitioners who've lived it, and a standing invitation for the room to tell them where they're wrong. [Source](https://www.youtube.com/watch?v=_HfjPtjPHUQ)

[8:05] Colin Campbell: Go hybrid. [11:18] Colin Campbell: Putting a little icing on the pig to make it look like a cake [16:41] Colin Campbell: Building is actually the fastest part of it. [18:40] Chris Slovak: Leave it where it lies. [18:40] Colin Campbell: We're not shooting for perfection, because you don't need perfection to drive the business outcome. [20:33] Colin Campbell: We're going to do two things well and then move on. [4:35] Colin Campbell: Skin in the game [11:18] Colin Campbell: Token Maxing

[11:18] Colin Campbell: If I spent $3 million handing out licenses to every individual in my company and I didn't have $3 million in benefit, why did I do it? What's the point? [11:18] Chris Slovak: putting a little icing on the pig to make it look like a cake [7:51] Colin Campbell: What do you want to be famous for? [16:31] Colin Campbell: consistency beats accuracy

[0:04] Thanks everybody for joining us and Colin,  thank you for joining us and being a partner [0:08] with on frame. My name is Chris Slovac. I'm global  field CTO of on frame and Colin, please introduce [0:13] yourself. Yeah. So, I'm Colin Campbell. I'm the  CIO over recurring revenue Cushman & Wakefield. [0:17] So, it's a little bit of a different job. It's  like a divisional CIO job. Except I see the entire [0:23] digital experience as well as the global data and  AI strategy for the firm. So, it's a little bit [0:29] of everything all wrapped up in the one. Well,  today we're going to just have a little chat. [0:35] Talk a little bit about the experience that Colin  had and his his decisioning to work with on frame. [0:40] We're probably going to talk a little bit about  some uh some interesting topics that maybe uh [0:45] hopefully will raise some eyebrows and  maybe even get some opinions started. So, um [0:49] I think I'm going to kick it off talking a little  bit about sort of the build versus buy journey. [0:55] Right? And everyone today is out here. There's  a bazillion vendors out in the in the hall. Um [1:00] and there's everything from full on consultancies  building whatever you want to buy it off the shelf [1:05] and and try to try to piece it together yourself.  Um so, Colin, for you, you know, how did you break [1:13] that down? How like what what was your skepticism  in in deciding how to partner in this sort of new [1:18] new space? Yeah. So, we were very interesting. So,  Cushman as a whole was pretty much universally a [1:24] buy shop. So, we were almost all SaaS-based  products. We've been cloud-native for gosh, [1:30] that feels like two decades even though that's not  possible. Uh but that's kind of how we started. [1:35] And then when you when AI started to pop up,  we were kind of sitting there reevaluating [1:38] our strategy. We realized that we couldn't go and  swing the pendulum quick completely the other way. [1:44] Cuz if we were just now on the other side  just doing complete buy or complete build, [1:48] what happened to everything else we just did for  the last, you know, 125 years? So, what we did [1:53] was we kind of kind of went into the middle. And  there's two problems that really struck us. One, [1:56] real estate is one of the most heavy data-driven  industries you can imagine. So, we're we're [2:01] sitting here in this giant ballroom uh as we look  around, but you got to think someone's sitting [2:05] there looking at the cost of air conditioning,  the cost of lighting, how much space, how many [2:09] people you need to clean this, how do who set  up the chairs. That's all data points that we [2:13] use to determine staffing models, optimizations,  value propositions, and so forth. So, that led us [2:19] to say, "Okay, that's not off the shelf anymore.  That's now something else." So, you needed to [2:25] find a partner to be able to get there. The other  problem was, as I stated at the very beginning, [2:30] we were a buy shop. So, I didn't have a whole  lot of people building internally at the time. [2:35] Uh and so, I needed to find a partner that could  kind of not be a staff augmentation capability. I [2:40] didn't need to like, you know, generic staffing. I  needed a partner to come in and say, "Help me set [2:45] my strategy as I'm growing my team and creating  real value-driven opportunities in that space." Uh [2:52] weird side chance, uh we actually  met OnFrame uh through their founders [2:57] uh at a information security conference because  they originally came out of that industry. So, [3:02] that's how we started. Uh we'd let it into  basically we needed a partner to come with us [3:07] to do more than just doing staff augmentation.  I kind of find that interesting too. [3:12] When we get into that dichotomy, build versus buy,  um let me just quickly from the the crowd here, [3:19] who here uses a cloud provider when  they're developing their software? [3:26] Really, it's only that many? Everyone else has  servers sitting in their in their garages? Is [3:30] that really what we're we're seeing here? I bet  you it's more like 80%. And so, I would argue that [3:35] you're actually already building with a partner.  Right? Like those partners are coming in and [3:41] they're bringing software and you're developing  a blueprint on top of that, right, to deliver a [3:46] solution. So, I you know, this this motion of like  build versus buy, it's it's kind of interesting. [3:51] It's You're already doing a partnership and how do  how do we extend that concept into into AI, right? [3:58] 100% and then I would say as you think about the  extension of a partnership, so whether you're a [4:03] Microsoft or AWS or whoever, you they have their  you have better off, you have Azure, you have [4:09] everything else going along with it. So, how do  you get the best of what you're already spending, [4:13] then connect it in to extend into other  places? And that was really in real estate [4:17] just as crazy because proptech in general has  a ton of micro point solutions across the way [4:24] that you then either have to connect in or think  about a replacement strategy as well that goes [4:28] along with that. So Okay, so we're going to build  together. Yep. There's some risk in that for you. [4:35] Correct. What did you need to see from us?  Like what are you what are you looking for [4:39] in a partner? And I think this is actually  just a fantastic like just if everyone walks [4:43] out with two things, like there's probably  I can tell you how many consultancies here. [4:47] Find the one that works for you. Like that's just  the number one thing. You're going to have two [4:51] things that are going to determine that, at least  we did. One, I needed to have risk getting skin in [4:55] the game. They got to have some risk here. Like  it's not if I'm committing my time, my capital, [5:00] I need them to care, too. So, figure out what  that balance for you is. Very easy partnership [5:05] with them around how their forward deployed teams  basically came in and partnered with our business [5:10] users as well as our tech users and then you  created a commercial structure that allows the [5:14] risk to be more outcome driven than just fixed  fee or consumption based or something like that. [5:20] That way as we got more out of it, there was a  shared benefit and if we got nothing out of it, [5:24] there was a shared loss. That was a very very  interesting model at the time. Now it seems a [5:30] little bit more common. It's becoming more I think  we're I think we're leading the way on that. Yeah. [5:35] But uh that has actually transformed us because  again, it's not about low risk or no risk, [5:40] it's shared risk. Because if your  partner cares, you're they're going [5:44] to show up a little bit differently than  if it's just me paying them a fixed fee. [5:48] I I think too that when we share the risk and  we share the same goal, we're going to move [5:55] faster together and build trust really quickly.  And part of that trust building, right, is and [6:02] I've been somebody who's been selling  software and been the SaaS side of things, [6:05] enterprise sales for 20 years. Always on the tech  side. I'm a tech guy, not not just a sales person. [6:11] But there's always this weird dichotomy of sharing  the outcome because you're always worried that [6:16] your your you know, your vendor is going to just  try to like maximize their revenue off of you, [6:21] right? But in this world, when we're sharing risk  and we're sharing outcome, we we have to share [6:26] real cost and real value, right? It's a kind of  a weird It's a weird change that has to happen. [6:31] Correct. And it's open and transparent we value  think about costing and modeling within that as [6:35] well, right? If you you can think about how you  want to do caps or sharing and other opportunities [6:40] from risk perspective, but in addition to it, I  also think there's just a gravity or a snowball [6:46] that occurs. You get that first real good use  case. Not the tiny one, not the little enterprise [6:52] search thing that you did in 10 seconds, but  you get your first tangible use case. That That [6:57] creates a little momentum. That creates a little  bit of that like snowball effect, if you will, as [7:01] it as it rolls down the hill and the next one gets  a little bit easier, it's a little bit bigger, [7:05] it's a little bit more challenging, maybe a little  costlier. And then [snorts] you get another bit of [7:09] momentum that keeps going and going and going.  And I don't mean that from just a perspective of [7:13] with Unframe or anyone else. I think that also  applies to just things you self-build as well. [7:18] But that idea of how you get the trust after  that first one and how that speed, that delivery, [7:23] that honesty, and even the bumpiness with it  creates that authentic nature, that's really where [7:28] the value came in. Cuz it wasn't just the shared  commerciality, it was the shared like trust that [7:34] you built. And then you can move faster because we  can have open discussions about where our opinions [7:39] are coming in and where you want to go and like  we we actually work as partners. I think it's um [7:45] it's a unique um partnership. So, we  appreciate it. I guess on that front [7:51] there's probably a a bunch of CIOs out here right  now today that are like trying to make a decision. [7:57] Do they go full build? Do they go buy? Do they  partner? I mean, what would your like I guess what [8:01] would your advice be to them? You know, the ones  that are in the camp of I I don't know what to do [8:05] yet. Go hybrid. Uh very very simple. You're going  to have a finite amount of people unless you are [8:13] in in that special group that has unlimited  engineering capabilities. And if so, that's [8:17] fantastic. I don't have that. But for those of us  who are more closer to the middle of the road and [8:22] more in reality you're going to want to go hybrid.  You're going to have a finite internal development [8:26] team that you're going to want to maximize their  capacity to some level of medium and quick wins [8:31] because that builds that internal momentum and  capabilities across your team. The hybrid approach [8:36] I liked cuz it did two things for us. It gave  us speed to market across some core products. [8:41] Secondly, where I have things that would tie  up significant parts of my engineering team, [8:46] I don't want them to be only focused on one thing.  That minimizes business value, business impact, [8:50] go to market, and also kind of makes people feel  like they're being excluded, to be honest with you [8:55] cuz you got to you do have to eventually solve  for adoption, right? So, you want to make sure [8:59] that you're able to get to give attention as  it's needed, but also have the ability to solve [9:03] for both complex as well as kind of the longer  run cases. So, what we do today with Unframe, [9:09] they were much more in the early stages about  the AI acceleration for us. What we do today is [9:14] our very kind of engineering heavy, complex cases  that would take up the vast majority if not all [9:19] of my team I send to them. The quicker wins that  are doing high-level business engagement, faster [9:25] interactions, we handle those internally. And that  hybrid approach really works well for us so that [9:30] I can kind of get the best of both scenarios  where I'm not leaving something on the table. [9:35] I mean, I I find that that's an  interesting thought process because [9:39] there's only so much talent that understands  how to build these systems. And you know, [9:44] the question is, do you do you try to consolidate  that talent into your into your business so that [9:50] you have to both be an expert in real estate and  commercial you know, operations for real estate. [9:57] And you have to be an expert in developing  a genetic systems or you know, do you focus [10:00] on being experts in real estate and have a couple  key people who know how to you know, to your point [10:05] build those fast wins. And then you have a partner  who's going to in this case us, right? And I'm [10:10] sure that there's other vendors out there, but  they're consolidating talent around building [10:13] these types of systems, right? Correct. It is.  And really focusing on your domain expertise. I [10:19] sat and had lunch with a couple gentlemen who work  in the insurance industry. Interesting parallels [10:23] around what between what we do and what they do.  Uh very heavy data-driven petabytes of data type [10:28] stuff. And it was really interesting in talking  to them in particular and they're trying to figure [10:32] out their own journey. I'm like, you can't live  give up your domain expertise. That is your asset. [10:37] That is your knowledge capital. Find someone who  allows that to shine. And that's really what I [10:42] I was trying to stress to them is that's both an  internal build as well as likely a hybrid model. [10:48] I think you're hitting on something which is like,  what do you want to be famous for? Yeah. I like [10:53] you should still even if you're working with a  partner I call it trademark software, right? That [10:58] your secret sauce you should still own, you  should still create, but you should be willing [11:03] and able to work with a partner who's open and  transparent to help you kind of carry that through [11:07] through through a enterprise level solution,  right? And sort of finding that balance is is a [11:12] is a is a real trick. Um I think that that can  only be done if you have the trust. Agreed. Um [11:18] So let's let's talk about okay, we've got we've  got a plan. You know, you've got you've got a [11:23] modality of working with a vendor. How  do you think about how this whole problem [11:28] value maximizing versus token maximizing, you  know, like this whole discussion right now of [11:36] I've got all this entitlement and compute. I've  got all this this this this entitlement and and [11:41] tokens, but I'm not actually generating any  ROI yet. Yeah, so this is where I think it [11:47] was two sessions ago, the five gentlemen  that were up here, they were really good. [11:50] But I have a little bit of a different view. And  I think that's what's the value of conferences, [11:54] you hear different things. And I want that hard  ROI. My project has to deal to generate revenue. [12:01] If I spent $3 million handing out  licenses to every individual in my company [12:06] and I didn't have $3 million in benefit why did  I do it? What's the point? I do think there is a [12:14] fundamental item that says as you're determining  where value sits, as you're creating something new [12:20] or optimizing something existing think about what  is the tangible benefit you're gauging against. [12:26] Is it reduction in time? Is it reduction  in head count? Is it growth in revenue? [12:32] And that's really what we did is that I would call  it value maxing if we're going to look at it from [12:38] a trendy buzzword thing that we're all going to  forget about in 6 months and be ashamed we said. [12:42] But from my perspective, it's really looking  at that ROI and say, "Hey, I did this. It cost [12:47] me a million bucks. You know what? That new  service, that new revenue just brought me 10." [12:52] That's where I go look at my KPI  and I say, "Check. Great job, team." [12:57] That's kind of how I think about it. The other  side of this, just to be as blunt as I can is [13:01] don't fall into the trap of efficiencies. Um that  was shameless plug here. I do have a class at [13:08] 2:00 p.m. if you're interested. Uh but backing  away from my shameless plug there, the trap of [13:14] efficiency is sitting there looking for savings  by optimizing something that is likely messy. [13:20] So, for example, you go in and look at someone's  job, you can easily optimize two or three of their [13:24] core tasks. The other 17 and 20, maybe not  because that's not just their job is moving [13:31] things from left to right. They also answer  the phone. They're a customer service rep. [13:35] They're all these other things, right? So, that  efficiency trap is the easy one to fall into. [13:41] Think about the growth opportunity. Think about  the other opportunities that go along with that. [13:45] And that's how we kind of turned into I hate it.  I'm going to say value maxing, but I'm going to [13:49] say going with an ROI driven approach on a go  forward basis. I mean, I think you get into [13:55] a dilemma that I keep seeing is everybody's  focused on automation. Mhm. They're not focused on [14:04] redesign and augmentation. And I And that's  like a critical concept for me because if [14:10] especially if you're focusing on I'm going to  give everyone licenses and then everybody's going [14:14] to build their own workflow to automate a task  that they're all doing. Did you Did you really [14:21] create ROI and solve a problem? You didn't really,  right? You're You're You're basically kind of like [14:27] I don't know, I'm putting a little icing on  the pig to make it look like it's a cake. [14:31] You really should look at the problem and think in  this new way, how do we redesign for this future? [14:39] Agreed. And there's a side bullet to that that  I don't want anyone to walk out and forget. [14:44] Your people who work in your various  industries, whether it's insurance, real estate, [14:47] anything else, financial services, doesn't matter  what, they have a domain expertise. They know [14:52] something about that space. So, augmenting that  I don't And thinking about that, even beyond And [14:58] I'm sure someone here is going to tell me that's  just ontology and context, but there's more to [15:02] it than that. There's a human interaction and  more importantly a brand awareness to that. [15:07] So, how does that they showing up? And how it  does replacing Again, I had to change my flight [15:13] and doing that lovely interaction with the chatbot  to get it to change. That took forever because I [15:18] couldn't get a representative. How do you want  that opportunity to be brand representative as [15:23] well? So, that's when you if you're redesigning  the work, you can kind of ring-fence. This is what [15:27] we're good at. This is what I want to be known  for. And then more importantly, this is what [15:31] we can do different to create a different level  of expertise to go along with it cuz we already [15:36] have that in-house. What am I really solving for?  And And And I think that changes the workload. [15:43] Oh, completely. Right? Because as we all know,  the cost of code and the speed of code is like [15:48] unbelievably fast. It's like near getting  near zero. So, it's not can you build it? It's [15:54] what should you build? Which means you  have to kind of We have to kind of go back [15:58] almost like back in time to old-school business  analyst work to figure out That's fast forward, [16:03] right? Yeah, so this is the thing that I think is  important is one, we did a study just internally [16:09] benchmarking like what do you want to  spend time on? Like what do you want to do? [16:13] And figuring out what to do is actually  now takes roughly 50% of development time. [16:19] So, if you would think like old-school, I mean,  I've been doing this for a little while now. [16:23] You think old-school development cycles, you  had kind of maybe simple math you guys 25% kind [16:28] of design up front. 50% build, then 25, maybe 30%  depending on where you kind of fluctuated on kind [16:34] of QA testing and kind of push out. Now, when most  of that's automated, what do you do differently? [16:41] Well, we've actually kind of discovered for  our most successful projects, 50% of the time [16:46] is actually spent in that design phase. How do I  determine that ROI? How do I determine who's going [16:51] to use this? What other dependencies are out  there that have to change to go along with it? [16:55] And that was a massive shift for us as well. As  you kind of think about your legacy mindset and [17:01] everyone's got their Jira workflows and epics,  I'm sure that they're thinking about right now. [17:05] How did that mindset and process have to change  to then align to that new model that says, well, [17:10] building's actually the really fast part now.  So. I think too, you talked about adoption. [17:17] And we've had this discussion before, but like  when you get into the adoption piece, it's [17:21] it's how good does it have to be for a user to use  it, right? Like the the whole consistency might [17:28] I argue consistency beats accuracy. This is like  the great debate right now, at least amongst [17:36] several people I've spoke to even at this  conference, is how accurate does accurate [17:40] need to be and what are you willing to pay to go  from 80% to 90%, 90% to 100%. And that's kind of [17:47] the interesting difference now with AI that  I think we don't really acknowledge as much. [17:51] So, for example, you're looking for a house, you  open Zillow, right? And you look at the number, [17:55] I did this the other day, be like, "Okay, what's  my house worth?" But you look at that number, [17:59] how accurate does that number have to  be for you to really kind of say, "Okay, [18:03] it's a good time to go look in that neighborhood,  it's a good time to go look over there." That's [18:06] an interesting parable to kind of where you're  going is how much do you really need to know? [18:10] If it's payroll, yeah, it's 100% accurate. The  CNN guys on the first day where they're talking, [18:15] yeah, if I'm going to put someone's name on the  board, it need to be 100% accurate and validated. [18:20] If I'm going in and making a directional choice  on forecasting for where I want to allocate my [18:25] resources, I don't know if I need to be 100% or  80% because there's going to be enough variability [18:30] in the ecosystem. It's the same thing going back  to that Zillow example of I just needed to know [18:35] that that house was worth that much money and now  I can kind of start looking in that neighborhood. [18:40] So. And that changes if we think  about that, if we if if 80% [18:45] consistently is enough to make decisions to move  the business forward, it changes how we think [18:51] about data. We I mean, we don't have to do these  massive centralization projects just to get the [18:57] data there to double check and model it and make  it perfect. Like we can leave it where it lies [19:03] and even if we know it's going to be  a little off, right? I mean, that's a [19:06] that's a completely different mind shift where we  want Yeah, and I will say this for us. Like one [19:10] of the things we want to speak is like if you are  a three-year-old company that can be AI-native, [19:16] great. I'm I have some jealousy of that. I'm  a 125-year-old company that has gone through [19:22] at least 37 different acquisitions through  buying different firms and companies. So, [19:26] I can't do that. So, when you look at how you want  to allocate your capital and control your capital, [19:31] is that giant MDM project worth it? Do I need to  make sure that I have the {quote} golden record [19:36] or can I match it to a to a statistical  degree of certainty and then I can just go. [19:42] That level of effort, capital, resource engagement  allows you to pick up so much speed and there's [19:47] a hundred other examples like that as well.  But, it's that idea of is consistency above [19:52] 100% accuracy the value to get to market to  drive much faster change. I think that that's a [20:00] a design First you got loud. Um a design factor  that you have to sort of work in to not just [20:07] your development cycles, but actually in  the conversation with your end users and the [20:12] stakeholders. You have to tell them, "Hey, look,  we're not trying to go for perfect because we [20:17] don't need perfect to actually move the business  forward and drive through business outcome." [20:22] And like that I think that's a mental shift that  many of us, and me included being a data person, [20:27] really really sort of struggle with um in terms  of changing. Um last question for you. Sure. Um [20:34] it's been 18 months or so now. Like if you had  one piece of advice that you would give yourself [20:40] as when you started all of this, what would it be?  All right. I will do the shameless plug bit again, [20:45] so apologies everyone. Uh one, there are some  basic guardrails that you think are just naturally [20:51] implied. It's kind of like when someone tells you  something, oh, I wish I should have known that. [20:55] I always use the example of your backup camera on  your car. You never knew you wanted one until you [20:59] had it and now you're like, "Man, I can't back  this back up without it." Like it's like that. [21:04] Uh so, there's some basic guardrails around that.  The other one is I would have activated my talent [21:08] strategy way faster. I was kind of on the I  went to hybrid and then went into kind of build [21:14] a little bit not completely simultaneously. I  should have been activating my talent strategy [21:18] much faster to then kind of re-evaluate because  again, the the thing that's happening is the [21:25] for whether it's technology leaders  in particular, the the job changed [21:28] and therefore the job of the people who work for  me all changed. So therefore acknowledging that [21:33] and then moving that along would have been  one of the biggest changes. And secondly, um [21:39] everyone offered a pilot for everything right  away for so long. There you we had so many [21:44] pilots going but trying to prove value or  even prove attention, I would have just [21:48] shut all those down right away and said we're  going to do two things well and then move on. [21:54] Awesome. Well, thank you Colin for your time.  Thanks, Chris. Always a pleasure, man. Yeah, [21:57] thank you. Let's hear it for Colin Campbell  and Chris Slovak. Thanks, guys. Great job.