Winland Foods Ties AI Agents to Cash Metrics to Escape Pilot Trap

Winland Foods将AI智能体与现金指标挂钩，摆脱“试点陷阱” 企业AI部署之所以陷入停滞，是因为团队在缺乏长期支持基础设施的情况下自动化了不稳定的流程；这要求企业将工具锚定在应收账款周转天数等现有指标上，并将归属权直接移交给一线员工。 Winland Foods与EY通过优先考虑上线后支持、一线业务归属权以及应收账款周转天数（DSO）等硬性指标，而非流于表面的技术实验，建立了持久的企业级AI应用。通过将重复的数据录入和常规扣款核对自动化，该公司消除了低价值的繁琐杂务，并使员工能够积极追回存在争议的收入。Winland借助严格的“流程、人员、数据和系统”四支柱评估框架，结合分阶段的人在回路推广步骤，化解了员工的焦虑并稳步扩大了自主权。每周持续的异常情况审查与一线员工的直接共同设计，将最初对操作的疑虑转化为了企业自发推动的广泛需求。 企业采用AI的最大阻碍并非软件本身，而是确保业务信任、归属权以及上线后的运营支持。 (0:04) 除非企业建立专门的支持机制，在底层系统和界面发生变化时对工作流进行修补和调整，否则AI试点将面临崩溃的风险。 (0:04) 为防止AI沦为投机性实验，相关项目应锚定在应收账款周转天数（DSO）等既定运营指标上，而非新奇的AI指标。 (2:16) 每周针对智能体使用情况和采用趋势的运营报告，能确保预期的财务回报切实落地，并促使自动化工具得到积极利用。 (2:16) 在订单到收款流程中实现常规数据提取和表单录入的自动化，消除了50%的“杂务负担”，让员工精力转向追回无效的客户扣款。 (4:24)

Drawing on a three-year enterprise deployment at Winland Foods, EY's Oscar Marin and Winland Foods' Joe Hutchings argue that successful AI scaling requires anchoring initiatives to existing operational metrics, establishing ongoing support systems, and securing frontline business ownership rather than chasing software capabilities alone. Anchoring Automation to Established Financial Metrics (2:16) Winland Foods bypassed open-ended experimentation by tying Finance Order-to-Cash (O2C) automations directly to existing baseline metrics, primarily Days Sales Outstanding (DSO). By measuring automations against DSO rather than speculative AI performance indicators, leadership established verifiable baselines to quantify working capital impacts, accelerate cash recovery, and test business assumptions. Beyond DSO, the team tracks operational dividends such as hour-for-hour time savings and the ability of leaner teams to absorb transaction volume growth without additional headcount. To prevent value erosion after launch, support teams supply weekly operational reports tracking system consumption and adoption trends, ensuring the tools remain utilized on schedule rather than relying on lagging quarterly reviews. Eliminating Clerical Friction to Recover Disputed Deductions (4:24) An operational analysis of frontline activities within Winland Foods' Order-to-Cash division revealed that roughly 50% of employee time was consumed by routine administrative tasks, such as copying data from emails, entering forms, and approving low-risk decisions. In a consumer packaged goods environment, resolving frequent customer deductions for damaged goods or short shipments had deteriorated into mechanical data entry and rubber-stamping credit memos. By delegating extraction and standard reconciliation to intelligent agents, the company shifted frontline staff from clerical processing toward investigating and clawing back invalid deductions. This redirection recovered disputed funds to the bottom line while noticeably improving day-to-day employee fulfillment by replacing repetitive data entry with investigative problem-solving. Alleviating Agentic Anxiety Through Operational Feedback Loops (6:55) To counter workplace unease regarding automation and job displacement, IT teams avoided designing theoretical workflows in isolation and instead paired technical staff directly with frontline operators. Leadership framed the initiative not as headcount reduction, but as superpowering existing personnel to handle higher transaction throughput with reduced administrative friction. This trust was reinforced by a structured exception feedback loop. When an agent encounters an edge case it cannot resolve, the task is handed off to a human operator. IT and operational personnel review these handoffs during weekly operational audits, studying how the human resolved the exception to iteratively refine the agent's underlying logic or correct systemic data flaws. Disciplined Use-Case Qualification Across Four Pillars (9:12) Winland Foods evaluates automation candidates against four core pillars: Process, People, Data, and Systems. Hutchings warns that attempting to automate ten poor processes merely generates ten poor automations plagued by continuous exceptions; workflows must be clean and standardized before development begins. Additionally, business units must actively champion ownership, and underlying data and legacy systems must possess the hygiene and integration maturity necessary for automated execution. The team intentionally operates under the title Intelligent Automation rather than AI to avoid deploying artificial intelligence for its own sake. Hutchings stresses that deploying expensive, probabilistic agentic architectures to tasks solvable by deterministic rule engines or basic scripts wastes capital. Furthermore, Marin notes that organizations must maintain strict consumption guardrails to monitor cloud and AI operating costs. Progressive Trust Building from Oversight to Autonomy (12:04) The organization implemented a phased rollout that began with deterministic robotic process automation before layering generative and agentic systems under strict Human-in-the-Loop review. Frontline users initially inspected and approved all proposed agent inferences before data was written back to enterprise platforms, removing concerns over opaque system decisions. As operators observed consistent accuracy, user sentiment shifted from manual verification to requesting full autonomous execution. To maintain transparency, weekly audit reports detailing every automated transaction, decision path, and escalation are shared directly with finance teams. Enterprise launch announcements also deliberately highlight photographs of frontline business co-designers rather than backend technical developers, fostering widespread internal ownership. Marin and Hutchings conclude that moving beyond stalled pilots requires treating enterprise AI as a hybrid operational model sustained by dedicated post-deployment maintenance teams. They caution that without upfront process standardization, clear consumption guardrails, and persistent frontline change management, enterprise automation initiatives risk devolving into costly, unanchored science experiments.

Winland Foods 与安永（EY）指出，企业人工智能试点失败的核心原因往往不是底层技术不足，而是缺乏业务部门的信任、运维支持与流程稳定性；成功的落地依赖于将自动化与既有财务指标直接挂钩，并通过分阶段推进来建立组织认同。 超越炒作：运维支持与既有指标锚定 (0:04) 在企业实际运营中，构建人工智能面临持续且繁杂的运维摩擦。Winland Foods 发现，底层技术往往已经完备，但业务流程并非一成不变：系统更新、界面微调以及因客户反馈而改变的规则，都会导致缺乏维护的自动化工具迅速失效。因此，建立专门的运维支持团队以持续修复和演进工具，其重要性远超过初期的原型搭建。 为避免人工智能项目沦为空泛的实验，Winland Foods 与安永选择将项目直接锚定在财务订单到收款（Order-to-Cash，O2C）流程中既有的运营指标上，例如应收账款周转天数（DSO）。该团队并未发明新奇的 AI 专属 KPI，而是依靠已有的基准来量化加速资金回笼的效果，并通过每周一次的实际使用率与采纳趋势追踪报告，替代滞后的季度回顾，确保自动化方案真正产生持续的财务与工作量消化效益。 消除事务性损耗与缓解员工焦虑 (4:24) 在日常业务分析中，Winland Foods 发现 O2C 部门一线员工约 50% 的时间被“繁琐事务”（busy work）消耗，包括从邮件中复制信息、填表、重复输入数据以及批准低风险决定。在消费包装品（CPG）行业中，客户常因货物损坏或短缺而直接从发票扣款。以往员工仅充当机械的纸面处理者，将数据提取与核对交给智能智能体后，员工精力得以转向调查无效扣款，追回争议资金，从而在提升现金流的同时改善日常工作满意度。 针对自动化引发的岗位替代焦虑（即“智能体焦虑”），IT 团队并未局限在管理层会议室勾勒理论流程，而是直接与一线操作人员并肩工作。团队将项目定调为“赋予员工超能力”而非裁员，并建立了围绕“业务异常”的反馈循环：当智能体遇到无法处理的边缘情况移交人工后，IT 会在每周复盘中分析人工的处理路径，持续更新逻辑规则，使异常移交转化为持续优化的学习契机。 四支柱准入框架与审慎的技术选型 (9:12) 为防止盲目引入技术，Winland Foods 制定了包含“流程、人员、数据、系统”的四支柱评估框架。企业明确警示不能将不完善的流程自动化，否则只会导致持续不断的异常处理。因此，必须在编写代码前确保流程稳定规范；同时要求业务部门具备长期所有权，底层系统和数据资产具备自动化决策所需的一致性、清洁度与集成接口。 出于审慎考量，团队被命名为“智能自动化团队”而非“AI 团队”，以避免单纯为了追求 AI 概念而增加成本。如果确定性规则引擎或简单脚本即可解决特定问题，使用昂贵且具概率性的智能体架构被视为资金浪费。针对每个实际问题，团队都会在确定性规则与生成式智能体之间权衡，以选用最具成本效益的方案。 分阶段推进与全透明审计建立信任 (12:04) Winland Foods 摒弃了风险较高的一步到位（big bang）上线模式，转而采取循序渐进的推进策略。实施路径首先从基础的机器人流程自动化（RPA）与确定性工作流切入，随后叠加生成式与智能体 AI 能力，并在初期严格执行“人机协同”（Human-in-the-Loop）审查机制。在此期间，业务人员必须先核准智能体的推论与决策建议，系统才会将数据写回企业核心平台。 由于消除了黑箱操作的担忧，员工在观察到近乎无误的执行表现后，态度迅速从审慎防备转变为主动要求系统自主执行。此外，团队与安永联合开发了全面的审计追踪报表，每周直接与财务团队共享，详细展示触发的每笔交易、决策路径与升级异常，透明的审计机制为自主化运行提供了可验证的安全保障。 以价值为先的落地执行与共创文化 (15:49) 对于希望摆脱试点停滞状态的企业，首要任务是在评估具体软件之前明确量化价值目标，选定受影响的运营 KPI（如营运资本或周转周期）并建立基线。在架构层面，除了技术选型外，还必须为基于资源消耗的云服务与 AI 成本设立严格的护栏，避免运营支出失控。 在组织协同上，Winland Foods 强调业务人员的全程参与，由一线团队向技术部门演示日常真实操作。系统上线时，内部宣传会刻意突出提供设计贡献的一线业务人员，而非后端的 IT 工程师。这一举措增强了一线员工的认同感与主人翁意识，推动自动化方案在全公司范围内以滚雪球的方式持续扩展。 该方案表明，企业 AI 的成功取决于前期严格的流程准入、持续的异常运维以及对一线员工的心理疏导。然而，对话也明确指出，以消耗量为计费基础的模型对企业成本控制提出了新挑战，如何在广泛自主化与严格财务护栏之间取得长期平衡，仍是企业扩展智能体规模时需持续应对的课题。

## Background 本视频由 Ai4 频道于 2026 年 8 月 19 日发布在 YouTube 平台。内容记录了安永（EY）执行总监 Oscar Marin 与 Winland Foods IT 业务应用高级总监 Joe Hutchings 的炉边对谈，深入探讨了企业如何打破 90% AI 项目困于概念验证（PoC）的僵局，并在订单到收款（Order-to-Cash）业务中实现 AI 智能体规模化落地的实战经验与操作手册。 ## TL;DR Winland Foods 与安永系统性分享了企业 AI 跨越试点陷阱、实现全业务线落地的成功路径。项目拒绝脱离业务的‘科学实验’，而是将 AI 部署直接锚定在应收账款周转天数（DSO）等核心业务 KPI 上，成功消除了员工日常 50% 的机械繁琐工作并转向高价值的无效扣款追回。为了化解基层的‘智能体焦虑’，团队通过敏捷分步推进、人机协同（Human-in-the-Loop）与透明化审计日志逐步建立信任，让一线业务人员从旁观者转变为方案的主导者与共创者。这套实践表明，企业 AI 规模化的决胜点在于流程标准化、持续的运维支撑团队、严格的按量计费成本防护栏以及将聚光灯留给业务用户的组织文化。 ## Atmosphere & style 整场对谈务实、坦诚且充满了一线实战的质感，彻底褪去了技术炒作的浮华。两位讲者以‘在业务战壕中摸爬滚打’的平实口吻互相补充，既有技术架构与成本控制的严谨思考，更充满了对组织阵痛与员工心理的温情体察。观众能清晰体会到从最初面对变革阻力时的谨慎摸索，演进至业务部门主动寻求创新、全员赋能共享胜利的振奋情绪，极具启发性与实操指导意义。 ## Key takeaways - [01:13](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=73s) 跨越 PoC 试点陷阱的关键在于建立长期运维支持体系，应对动态变化的业务流程与系统升级，其重要性远超最初构建技术方案本身。 - [03:00](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=180s) 价值衡量必须直接锚定现存的核心业务 KPI（如订单到收款中的 DSO），并通过每周报告持续追踪系统采纳率与增量价值，而非依赖低频复盘。 - [05:15](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=315s) AI 落地可消除员工日常约 50% 的机械繁琐工作，将精力转移至调查与追回无效扣款（Invalid Deductions）等高价值环节，大幅提升员工工作质量。 - [07:38](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=458s) 消除基层员工的“智能体焦虑”需坚持赋能而非替代的定位，通过深入一线共同设计工作流，明确技术是为员工提供“超能力”而非削减人员编制。 - [09:51](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=591s) 场景准入需严格评估流程、人员、数据和系统四大支柱，坚持务实的自动化技术哲学，坚决反对脱离实际需求的“为了 AI 而 AI”。 - [12:04](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=724s) 将智能体转交人工处理的“业务异常（Exceptions）”视作推动系统迭代的重要学习资源，以此驱动企业内部形成持续创新的文化。 - [14:07](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=847s) 建立对 AI 的信任需采取分阶段渐进路线，从基础规则自动化起步，结合人机协同（Human-in-the-Loop）把关与全透明审计日志逐步放开权限。 - [16:23](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=983s) 企业推进 AI 落地应拒绝脱离业务指标的自娱自乐式“科学实验”，必须始终围绕具体业务指标与实际价值创造展开布局。 - [17:28](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=1048s) 自动化实施前需先行评估流程标准化程度，采用敏捷小步快跑的迭代策略快速验证价值，从而引发滚雪球式的全业务线推广。 - [18:02](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=1082s) 在技术架构上针对按量计费设立精细成本与安全防护栏，在组织宣传中把荣誉和聚光灯留给一线业务共创者，极大激发全员拥抱 AI 的积极性。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=4s) 从技术炒作到落地实战：三年规模化 AI 的经验开篇 安永（EY）工业产品领域数据与 AI 业务执行总监 Oscar Marin 与 Winland Foods IT 业务应用高级总监 Joe Hutchings 共同回顾了双方过去三年在企业内部规模化部署人工智能的历程。Oscar 指出，当下围绕 AI 的讨论往往充斥着兴奋与宏大愿景，但当真正深入一线日常业务的“战壕”中去推进实施时，会遭遇各种截然不同的现实挑战。两人希望通过坦诚分享这些在实战中淬炼出的经验，为渴望跨越试点陷阱的企业提供借鉴。 --- ### [01:13](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=73s) 走出概念验证（PoC）：为什么支撑体系比构建方案更重要 当 Winland Foods 开始从最初的实验探索迈向全企业范围的实际应用时，最大的障碍并非技术本身。Joe Hutchings 强调，技术能力足以完成大部分预想的自动化任务，但真正的难点在于确保业务部门的声音被充分倾听，并让业务方对这些解决方案拥有归属感（Business Ownership）与信任感，从而创造真正的业务价值。 在最初的试点阶段，团队很快就让模型和程序跑了起来，但现实中的业务流程和底层系统是动态变化的。例如，日常系统的版本升级可能会导致界面按钮位置发生偏移，或者客户反馈会促使业务流程发生调整。如果方案无法灵活应对这些变动，自动化就会瞬间失效。Winland Foods 由此意识到，建立一套能够长期维护、迭代并响应变化的运维支持团队（Support Team），其重要性甚至超过了最初构建解决方案本身。 --- ### [03:00](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=180s) 业务价值的定义与度量：以订单到收款（Order-to-Cash）与 DSO 为锚点 ```mermaid flowchart LR accTitle: 业务价值衡量与监控闭环流程 subgraph S1["基线设立"] A["业务核心 KPI<br/>(如 DSO / 应收账款周转天数)"] end subgraph S2["智能体部署与协同"] B["自动化流程与<br/>AI 智能体上线运行"] end subgraph S3["持续追踪与报告"] C["每周运营报告<br/>(采纳率 / 效率 / 业务异常)"] D["增量价值产出<br/>(时间节省 / 团队扩容)"] end A --> B B --> C C --> D D --> A ``` 为了防止 AI 项目沦为初期热情高涨、随后价值不断衰减的摆设，Winland Foods 在财务领域的“订单到收款”（Order-to-Cash, O2C）流程中确立了严格的价值评估体系。团队并未凭空创造新指标，而是直接锚定业务部门现存且每日监控的核心业务 KPI，尤其是“应收账款周转天数”（DSO, Day Sales Outstanding）。 通过为现有指标设立基线，团队能够清晰追踪自动化流程和部署的 AI 智能体（Agents）对该指标的直接改善。除了 DSO 等硬性财务收益外，团队还将监控维度拓展至协同效益，包括时间节省、流程效率提升，以及赋能规模较小的团队承担更大业务体量的能力。支持团队每周都会发布运营报告，持续跟踪采纳率（Adoption）与系统使用情况，确保预期价值能够切实兑现。 - **核心量化指标**：以应收账款周转天数（DSO）为基准度量业务改善。 - **监控机制**：支持团队每周出具报告，而非依赖季度或年度复盘，实时掌握系统使用率与采纳动态。 --- ### [05:15](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=315s) 超越财务指标：消除 50% 机械繁琐工作与提升员工生活质量 在资金周转速度加快和工时节省等显性财务回报之外，AI 落地带来了意料之外的重要无形收益——员工工作生活质量（Quality of Life）的显著改善。在深入调研订单到收款流程时，团队发现员工日常约 50% 的工作都是纯粹的“磨人杂务”（Busy Work），例如填写表单、从客户处收集信息录入系统，以及处理低价值的微小日常决策，这些工作枯燥且缺乏成就感。 作为一家快消品（CPG）制造企业，Winland Foods 经常会收到客户提出的扣款索赔（Deductions），如客户声称“货物少发”或“产品破损”并要求抵扣货款。过去，业务人员需要耗费大量精力去核对每一份单据并手动处理常规退款，工作价值极低。通过将这些机械的核对与录入工作交由 AI 智能体自动化处理，员工得以将精力集中于真正高价值的环节——调查并追回“无效扣款”（Invalid Deductions）。这不仅为企业挽回了真金白银，更让员工从繁琐的“打勾盖章”转变为从事具有探索性和成就感的核心业务。 - **日常杂务占比**：自动化前员工约 50% 的时间消耗在机械性的数据录入和表单核对上。 - **高价值转移**：从被动的常规扣款处理，转向主动调查与追回不合理的无效扣款（Invalid Deductions）。 --- ### [07:38](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=458s) 化解“智能体焦虑”：与一线深度共建而非替代人力 企业引入新技术时，管理层往往欢欣鼓舞，但基层员工中普遍会产生所谓“智能体焦虑”（Agentic Anxiety），担忧自己的岗位会被 AI 取代。Joe Hutchings 指出，化解这种焦虑的核心在于建立业务互信。Winland Foods 从一开始就深入一线，不仅与业务主管开会，更直接坐到具体经办人员身边，向他们请教日常操作中的“小窍门”（Tips and Tricks）以及流程图上从未体现过的突发状况与现实挑战。 管理层向员工明确传达的核心信息是：技术的目的是为员工“赋予超能力”（Superpowering them），让大家能够以更少的资源完成更多工作，而不是削减人员编制。同时，支持团队通过每周报告实时捕捉 AI 无法处理而转交人工的“业务异常”（Business Exceptions），并与业务人员共同分析解决路径，进一步优化自动化逻辑。这种把员工视作共同设计者与主导者的沟通方式，有效消除了对立与恐慌情绪。 > “这不是为了裁减人员，而是为了让我们能以更少的资源做成更多的事情……让员工的工作变得更加充实有趣，而不是整天陷入单纯的数据录入。” --- ### [09:51](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=591s) 评估 AI 适用性：流程、人员、数据与系统四要素 并不是所有场景都适合立即套用 AI。Winland Foods 建立了一套严密的准入评估框架，涵盖四大关键支柱： | 评估维度 | 核心考量标准 | |:--- |:--- | | **流程（Process）** | 流程是否清晰稳定？糟糕的流程盲目自动化只会产生“10 个糟糕的自动化”，并导致海量的业务异常。 | | **人员（People）** | 最终承担责任的业务用户是否对方案感到兴奋、具备主人翁意识并迫切渴望引入该技术？ | | **数据（Data）** | 现有数据质量与丰富度是否足以支撑 AI 智能体做出精准的业务推断与决策？ | | **系统（Systems）** | 底层软硬件系统是否具备良好的集成接口？对于老旧遗留系统（Legacy Systems），叠加智能体可能并不划算。 | Joe 强调，他们的团队被命名为“智能自动化团队”（Intelligent Automation Team）而非“AI 团队”，这代表了一种务实的技术哲学：坚决反对“为了 AI 而 AI”。在很多业务场景中，简单的确定性工作流或传统自动化（如标准 RPA）成本更低、见效更快，团队会根据实际问题精准匹配最适宜的技术手段。 --- ### [12:04](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=724s) 从抵触到渴望：业务异常（Exceptions）驱动的创新文化 随着早期项目的成功与员工工作体验的切实提升，企业内部经历了深刻的文化转变。最初的疑虑与观望逐渐消散，业务部门开始主动找上门来，源源不断地提出新的自动化需求：“我们能否把另一个流程也自动化？”“这个特定业务场景能否用 AI 来解决？” 在日常运维中，智能体遇到未预设情况转交人工的“业务异常”不再被视作系统故障，而是被当作推动持续创新的重要学习资源。团队通过分析异常背后的根因（是流程缺陷、数据缺失，还是需要引入更具自主性的智能体方案），持续迭代系统能力。公开透明的沟通与看得见的生活质量改善，促使全员形成了拥抱 AI 与人机协同的创新文化。 --- ### [14:07](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=847s) 建立对 AI 的信任：分阶段落地、人机协同与透明化审计 为了打破员工对“AI 黑盒”的不信任，Winland Foods 与安永采用了一种类似敏捷开发（Agile）的分波次推进模式，拒绝一次性上线的“大爆炸”（Big Bang）策略。团队先从规则明确的基础 RPA 自动化切入，随后在适宜节点分层引入 AI 智能体，并严格设置“人机协同”（Human-in-the-Loop）机制——让业务人员在操作真正执行前，全面复核智能体的推理判断与决策建议。 随着准确率在实测中不断得到验证，业务人员很快建立了信心，甚至主动提出：“别再让我们每周人工核对了，它大多数时候都是对的，直接让它执行吧。”此外，安永协助设计了全透明的审计追踪（Audit Trail）机制，详细记录智能体执行了哪些操作、由谁触发以及决策路径为何。这套报告直接向财务业务团队呈现，彻底消除了技术黑盒带来的不安全感。 - **推进路径**：基础规则自动化（RPA） $\rightarrow$ 叠加 AI 决策推断 $\rightarrow$ 人工把关复核（Human-in-the-Loop） $\rightarrow$ 业务信任建立后放开全自动执行。 - **透明审计**：提供端到端的可追溯日志，将业务执行细节直接向财务部门公开，而非仅停留在 IT 内部。 --- ### [16:23](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=983s) 周一早晨即可执行的落地法则：拒绝“科学实验”，聚焦 KPI Oscar 指出，Winland Foods 的实践本质上构建了一种新型工作模式（New Way of Working）——人类员工与软件智能体高效协作的混合工作流。当被问及其他企业领导者若想在“周一早晨”立即采取行动推动 AI 跨越试点阶段时，Joe 给出了首要建议： 企业必须首先将目光聚焦于能够创造真实价值的地方，从既有的业务核心 KPI 出发，明确想要影响和提升的具体业务指标。脱离了可度量的业务指标，任何 AI 部署都只会沦为自娱自乐的“科学实验”（Science Experiment）。在明确了价值锚点之后，再深入剖析流程与人员的就绪度，稳扎稳打地推进规模化落地。 ### [17:28](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=1048s) 小步快跑与快速迭代：从流程标准化到雪球效应 在推进自动化之前，首要任务是评估业务流程本身是否已经准备就绪：团队是否需要先开展改进活动来对特定环节进行标准化，以及负责推动这些改进的人员是否已做好准备。Winland Foods 取得成功的关键因素之一在于采用了敏捷迭代的方法——先以较小的切入点启动，一旦验证了实际业务价值并赢得业务部门的认同与采纳（buy-in and adoption），便立即开展快速迭代。这种透明度与可见性让项目的成功自然而然地形成了“滚雪球效应”，推动自动化在更广泛的范围内迅速扩展。 - **流程就绪度评估**：在引入 AI 或自动化前，先诊断现有业务流程是否需要先期优化与标准化。 - **敏捷小步切入**：不盲目追求一步到位的大型工程，而是从小规模试点起步，快速验证成效。 - **价值驱动的飞轮**：随着业务端看到实际收益与价值，主动采纳意愿显著提升，项目推进势头呈滚雪球式加速。 ```mermaid flowchart LR accTitle: 流程标准化与敏捷迭代路径 subgraph S1["前期准备"] A["流程标准化评估"] --> B["业务团队就绪度确认"] end subgraph S2["敏捷迭代"] B --> C["小规模试点验证"] C --> D["快速验证业务价值"] end subgraph S3["规模化推广"] D --> E["业务认同与主动采纳"] E --> F["滚雪球式全业务线推广"] end ``` --- ### [18:02](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=1082s) 技术防护栏与全员赋能：把聚光灯留给一线业务用户 EY 与 Winland Foods 强调，这趟转型之旅是全方位的。技术与系统架构固然是基石——必须选对技术路线，并在当今普遍采用按量计费（consumption-based）的模式下，建立严格的技术防护栏（guardrails）来实时监控并追踪成本。然而，最深刻的教训依然在于“人”与“组织”的准备：必须确保组织上下全力支持该举措，让每个人都感受到赋能与支持，并在设计阶段及早开展全透明的沟通，而不是等 AI Agent 或自动化已经在生产环境中上线后才告知一线员工。 在让业务人员参与设计的过程中，一线员工向 IT 团队全面展示了他们日常的实际操作细节。项目团队最重视的一点是“庆祝阶段性胜利”并向全公司广泛宣传：每次有新的 Agent 或自动化流程成功上线时，公司发布的全员公告中放的并不是构建系统的 IT 团队照片，而是深度参与设计与落地的业务用户照片。这种将荣誉归于业务一线的做法，极大地激发了全公司的参与热情与兴奋感。 - **按量计费成本监控**：现代云原生与 AI 技术均为按需消耗模式，必须在架构层面配置精细的成本追踪与安全防护栏。 - **设计阶段深度共创**：业务人员不仅是使用者，更在设计初期就深入展示日常工作流，扮演共创者角色。 - **让业务唱主角的变革宣导**：上线宣传重点突出业务用户的贡献与面孔，有效打破 IT 与业务的壁垒，建立起全员拥抱 AI 的文化氛围。 ## Notable quotes > [07:38](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=458s) 这不是为了裁减人员，而是为了让我们能以更少的资源做成更多的事情……让员工的工作变得更加充实有趣，而不是整天陷入单纯的数据录入。 — Joe Hutchings > [01:13](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=73s) 技术能力足以完成大部分预想的自动化任务，但真正的难点在于确保业务部门的声音被充分倾听，并让业务方对这些解决方案拥有归属感与信任感，从而创造真正的业务价值。 — Joe Hutchings > [01:13](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=73s) 建立一套能够长期维护、迭代并响应变化的运维支持团队，其重要性甚至超过了最初构建解决方案本身。 — Joe Hutchings > [09:51](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=591s) 糟糕的流程盲目自动化只会产生‘10 个糟糕的自动化’，并导致海量的业务异常。 — Joe Hutchings > [14:07](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=847s) 别再让我们每周人工核对了，它大多数时候都是对的，直接让它执行吧。 — 业务人员 > [16:23](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=983s) 脱离了可度量的业务指标，任何 AI 部署都只会沦为自娱自乐的‘科学实验’。 — Joe Hutchings > [12:04](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=724s) 我们能否把另一个流程也自动化？这个特定业务场景能否用 AI 来解决？ — 业务部门 > [18:02](https://www.youtube.com/watch?v=hpqIzDdZsG4&t=1082s) 每次有新的 Agent 或自动化流程成功上线时，公司发布的全员公告中放的并不是构建系统的 IT 团队照片，而是深度参与设计与落地的业务用户照片。 — Joe Hutchings ## Mentioned EY（安永）, Winland Foods, Joe Hutchings, Oscar Marin, Ai4, Order-to-Cash (O2C), DSO (Day Sales Outstanding), RPA, AI Agents, Human-in-the-Loop, PoC ## Video description Joe Hutchings - Senior Director - IT Business Applications, Winland Foods Oscar Marin - Executive Director, EY 90% of AI pilots never make it past proof-of-concept. EY and Winland Foods broke that pattern, and they'll tell you exactly how. In this candid fireside chat, the business and technology leaders behind a real AI Factory deployment share what it actually took to move from promising experiment to enterprise-wide adoption across Order-to-Cash operations: the governance fights, the change management failures, and the operating model that finally stuck. Walk away with a repeatable playbook — not a polished slide deck, but hard-won lessons you can use Monday morning. [Source](https://www.youtube.com/watch?v=hpqIzDdZsG4)

[7:38] Joe Hutchings: 这不是为了裁减人员，而是为了让我们能以更少的资源做成更多的事情……让员工的工作变得更加充实有趣，而不是整天陷入单纯的数据录入。 [1:13] Joe Hutchings: 技术能力足以完成大部分预想的自动化任务，但真正的难点在于确保业务部门的声音被充分倾听，并让业务方对这些解决方案拥有归属感与信任感，从而创造真正的业务价值。 [1:13] Joe Hutchings: 建立一套能够长期维护、迭代并响应变化的运维支持团队，其重要性甚至超过了最初构建解决方案本身。 [9:51] Joe Hutchings: 糟糕的流程盲目自动化只会产生‘10 个糟糕的自动化’，并导致海量的业务异常。 [14:07] 业务人员: 别再让我们每周人工核对了，它大多数时候都是对的，直接让它执行吧。 [16:23] Joe Hutchings: 脱离了可度量的业务指标，任何 AI 部署都只会沦为自娱自乐的‘科学实验’。 [12:04] 业务部门: 我们能否把另一个流程也自动化？这个特定业务场景能否用 AI 来解决？ [18:02] Joe Hutchings: 每次有新的 Agent 或自动化流程成功上线时，公司发布的全员公告中放的并不是构建系统的 IT 团队照片，而是深度参与设计与落地的业务用户照片。

[9:12] Joe Hutchings: if you have 10 poor processes and you just try to automate those, you're going to get 10 poor automations [6:55] Joe Hutchings: superpowering them [12:04] Joe Hutchings: just let it do it [4:24] Joe Hutchings: busy work [6:55] Oscar Marin: agentic anxiety [9:12] Joe Hutchings: Intelligent Automation team [15:49] Joe Hutchings: science experiment [12:04] Joe Hutchings: black box

[0:04] Thank you so much and thank you everybody for  making time and joining us in this discussion. Uh [0:09] I'm here with my friend uh Joe and we're going  to talk a little bit about the experience that [0:15] we have had for the last three years working  together building artificial intelligence at [0:20] scale and hopefully share some of those learnings  with all of you uh today. Again uh nice to meet [0:27] you all Oscar Rad. part of Ersen Young. I'm  an executive director in our um data and AI [0:33] practice for industrial products. And with me  is Joe. Hello everybody. Nice to meet see you. [0:39] And uh maybe we can start getting into it and  and I know when uh everybody's talking about AI, [0:44] it's always about the excitement and all of the  great things that are coming with artificial [0:48] intelligence. reality is when you're doing this  in the trenches and working on a day-to-day basis, [0:54] there are all kinds of different challenges  and we have experienced that firsthand. So [1:00] Joe, maybe we can start there. Um, and as Will  started moving from experimentation towards [1:06] adoption, uh, what was one of the hardest moments  and the biggest obstacles that you have to face? [1:13] Um I think when we started getting into it, we  realized the technology was there to do a lot of [1:18] the automation that we were exploring. Uh but what  we wanted to make sure we were doing is hearing [1:24] from the business, listening to the business and  making sure that they were supportive uh and going [1:29] to own the solutions that we developed so we could  create true value. uh when we st first started [1:34] piloting, one of the things that we realized was  you get something running very quickly. Uh but [1:40] making sure that that solution could adapt uh to  changes in the business, the business process, you [1:45] know, one day your system's working some way and  you do an upgrade and something changes, a buttons [1:50] in a different place or a business process has  adjusted based on customer feedback and we needed [1:55] to make sure we had a mechanism to support those  type of changes and the solution could evolve uh [2:00] as we went forward. So making sure that we had  the support team behind the solutions we were [2:06] building we started to realize was almost more  important than building the solutions themselves [2:10] in addition to having the business ownership  and alignment and trust in those solutions. [2:16] Yeah. So it needs to be a holistic approach right  it's not a technology first type of to of approach [2:23] and before we get into the whole technology  discussion with no is very important [2:30] one of the things that I think is  critical in this type of capabilities is [2:36] how did the team define what value actually sure  because everybody have different priorities so [2:42] how do you guys define value and then how If you  make sure the value not degrade after initial t [2:50] it's very easy to start with the expectation  of hey we're going to be doing all these great [2:53] things right but the test of value is going to  be overtime. So how do you guys deal with that? [3:00] Sure. So um in this particular example we  started with our finance order to cash processes [3:07] and you know most teams are already monitoring  specific KPIs that they're tracking to and [3:12] reporting on. So that's what we looked at first is  what are the KPIs the teams are monitoring today [3:18] uh that impact the outcome of the business and  so we started there and in particular we looked [3:23] at DSO day sales outstanding and looked to see  where we could make a difference there and where [3:28] we could actually measure uh the impact from the  automations and and the agents we were deploying. [3:34] Um and so the important thing was you have a  baseline metric that you're already monitoring [3:38] today. uh we would have a way to monitor the  improvement of that metric going forward. [3:42] But also from a business case standpoint, were we  seeing the value not only to the metric but from [3:48] some of the other you know tangental benefits  that we expected to see both in efficiency and [3:53] time savings um and the ability to take on more  work for a smaller team. And so having those [4:00] capabilities to monitor is really what we  built into our process. Um as I mentioned the [4:05] support team not only provides the support  for the agents but they also report out [4:10] on a weekly basis uh the adoption the  usage uh to make sure that we were seeing [4:16] you know the true benefits that we expected  to out of these solutions. So I know when [4:24] thank you hopefully this is better. So, so I  know that uh when you're talking about value, [4:29] there is a lot of index on um for example what the  financial benefit is going to be doing. Of course, [4:35] I have an AI agent doing a particular task that  means that a person is not doing that task. [4:41] Can you speak a little bit about what are the  other intangibles that you also saw you talk about [4:46] adoption right like that that's very important to  make sure that the technology is being used uh in [4:51] and consumed right because there is an investment  being made from that perspective but maybe you [4:56] can talk a little bit more about outside of  the typical financial metrics of hours saved [5:03] revenue generated especially since you guys did  a lot of stuff with order to cash right like so [5:08] getting access to capital faster is always a good  opportunity. What other metrics or or value levers [5:15] you experience with uh the work? Yeah, I mean I  think there's the obvious, you know, financial [5:20] benefits that you can measure, but um one thing  that maybe we didn't foresee, but we're seeing now [5:26] is um you the quality of life for our employees,  the ability to automate some of those boring and [5:32] routine tasks that they had to do as part of their  work, especially in order to cast processes that [5:37] we were automating. When when we dug into it, we  found about 50% of the work they were doing was [5:42] really just busy work um because they had to  fill out a form, right? Gather information from [5:48] a customer, enter it into the system, uh what have  you, make very minor decisions about activities [5:54] they had to perform. Not very fulfilling work  in some cases. But what we were able to do with [6:00] our automations and our agents were to automate  some of those routine tasks so that they could [6:06] focus on the value targets. So, one example  that we automated a lot on, we're in a um a [6:11] CPG company and so we get deductions from our  customers. For example, we ship a product and they [6:17] say, "Hey, you short shipped us or it's a damaged  product, so we want to take a credit on that." Um, [6:22] the value there is recovering uh those quote  unquote invalid deductions. And the ones where the [6:28] team was strictly going through and checking the  boxes to make sure they had all the paperwork and [6:32] were able to do the, you know, process the credit,  not very valuable. when you give them the focus on [6:38] the opportunity to where can they actually  recover some of those funds um and do some of that [6:43] investigation, it's a little bit more exciting,  a little more interesting to their work. So, [6:47] not only do you get the financial benefits,  but you get the quality of life benefits for [6:50] employees to feel like they're working on  value added activities. And and and Joe, a [6:55] little bit of a followup from that perspective,  right? Because introducing the technology, it's [7:00] going to be great, right? and and and I believe  the the executives are going to be supportive of [7:06] oh more technology to do our work better that can  create a little bit of an anxiety on on the user [7:11] group right uh can you speak a little bit about  your experience dealing with let's call it agentic [7:17] anxiety as you're deploying the technology some  folks are going to be adopters and they're going [7:22] to feel great about hey I have this capability  others are going to be uh oh right like we hear [7:27] the news and AI is coming and all of that good  stuff right like maybe speak a little bit about [7:33] how you guys dealt with the agentic anxiety  that comes with deploying technology. Sure. [7:38] Yeah. So, it's really with these things, it's  all about trust um with your business users and [7:42] your business partners. And part of the way we  achieved that is, you know, we worked with the [7:47] people doing the work from the very beginning. We  we sat down not only with the leaders to identify [7:52] the processes that we could automate, but we  wanted to hear from the people that did the work, [7:56] right? What were the uh tips and tricks that they  could teach us? some of the challenges that they [8:01] would encounter um things that might not be in a  flowchart that they have to deal with every day [8:06] and really about communicating. This is  about giving them we're superpowering them, [8:10] right? We're giving them the ability to do more  with a lot less. And so the communication was [8:15] not looking to eliminate resources. We're looking  to be able to do more with less resources and kind [8:22] of to my last point, make their lives a little  bit more exciting and interesting with the work [8:25] they're doing rather than just data entry in some  cases. So um I think that was a big part of it and [8:32] part of the support model that we've put in place  is exactly that right are we seeing the adoption. [8:38] So we do week weekly reports we're not waiting for  the you know a quarterly report or the end of the [8:42] year where we're trying to see how the solutions  are working. We're doing weekly reporting [8:47] looking to understand is the adoption there  that we expected. Are we encountering business [8:53] exceptions that we didn't plan for, you know,  in the solution that it's having to hand off [8:57] to a human to handle those? And how can we work  with those business users to understand how did [9:02] you overcome that particular business exception  and can we further add automation to handle that [9:08] um and so that was a lot of what we did to to  work with them. Well, that's great. I mean, it's [9:12] always important to look after all the different  components. Of course, technology is going to [9:16] be one part of that. uh the process is going to  be another part and you know make sure that the [9:21] people which is the the heart and soul of of the  organization is is behind it and and going back [9:27] into the topic of value as as well, right? Like  so once the value was clear and I think you guys [9:34] did an amazing job on articulating what the value  case was going to be for all the different agents [9:39] that were deployed. How did you guys determine  which use cases were actually ready for AI [9:45] and and based on the maturity of the process, the  people, the data and the underlying systems? Sure. [9:51] Yeah. I think you just touched on it, right? Is  that's the very first thing is uh your process. [9:56] You know, do you have a clean stable  process that you can execute and automate? [10:00] um if you have 10 poor processes and you just  try to automate those, you're going to get 10 [10:05] poor automations and you're going to have a lot  of business exceptions trying to handle, you know, [10:10] the one-off occurrences that you run into. So,  that was one of the first things we looked at is [10:13] do we have a clean process and if we, you know,  are there ones we can make minor adjustments to [10:19] uh to start with those? The next one would be the  people u making sure the people are excited about [10:24] it. uh the people feel ownership uh the business  users who are ultimately going to be accountable [10:29] to take advantage of these solutions um are  looking to are clamoring for it in a way. Um the [10:36] next thing would be do we have the data for the  systems to make the type of decisions that would [10:41] need to be made. um and do we have the system to  actually perform some of the activities and so [10:47] and can we integrate with those systems right  in some cases you might have a legacy approach [10:51] uh maybe we're not to a point where laying  an agent on top of that would make sense yet [10:56] um I think another thing that we we spent time  focusing on was not to just implement AI for [11:02] AI's sake but to treat our team is is called  intelligent automation so we're not called the [11:08] AI team we're called the intelligent automation  team because there's some cases where a simple [11:13] workflow or deterministic automation makes  more sense and would provide a lot more value [11:18] than trying to put an agent on top of something  that's going to cost you a lot more. And so that's [11:22] one of the things going through these is making  sure are we applying the right automation approach [11:27] to the problem and do we have those four areas  covered from process people data and systems. [11:33] No, that that's very interesting and the other  thing that you know we we experience together is [11:37] the fact that as you are creating the right uh  boundaries and the right communication picking [11:42] the correct technology it also creates like a  culture of innovation uh in in the company and [11:48] as you provide more visibility how things are  progressing then folks are starting to get more [11:52] comfortable. Can you speak a little bit about the  experience? So have you seen that cultural shift [11:58] of going from a little bit of anxiety for the  automation towards embracing the technology and [12:04] building that culture of innovation? Sure. Yeah.  Yeah. I think there there was some trepidation at [12:08] first from um the business whether this was going  to work and once we started to show the value [12:14] and they started to experience it. We're getting  requests now all the time. How can we automate a [12:18] different process? How can we apply AI to this  particular business situation? Um, so they're [12:23] really coming to us uh to help innovate and it's  a partnership between them and us. Um, I talked [12:31] about the reporting we do every week, a business  exceptions or what we call when an agent needs to [12:36] hand off to uh a person to help with a task and  we can learn a lot from those business exceptions [12:42] as well. Uh why are those things happening? Um  are those process issues? Are those data issues? [12:49] um how can we refine or can we add additional  automation on top of it or in some cases maybe [12:55] we need to apply an a more agentic solution to  those type of business exceptions to handle them. [13:01] Um, but what we've seen just through  communicating and that quality of life [13:05] improvement for the employees, they're coming to  us every day saying, "We want more automation, [13:09] right? We want more automation. We need to be able  to do more with less, even more with less." Um, [13:15] and so I think there's been a good adoption  curve of people being excited about these [13:19] solutions since then. Yeah. And and then the  the other piece that I believe is is important [13:24] to to cover in in this discussion is of course  as you have technology doing more of the work [13:31] and and the different processes there is going to  be an issue with trust right like typically you [13:37] assign some work to Jenny or Joe or or Mary  and you can trust that they're doing the work [13:43] and you can ask for follow-ups and just make  sure that the activity is being uh executed. [13:49] So especially in the in the age of artificial  intelligence and any type of technology deployment [13:55] there is always the the concerns around hey is  what the AI telling me accurate right like so [14:02] how do you guys deal with the concept of  building the trust in in the technology sure [14:07] so the approach we took was um I would say  similar to an agile methodology where we worked [14:13] in waves uh so we didn't do a big bang solution  all up front and and just apply AI immediately [14:18] in some cases We started with more simplistic, you  know, RPA automation as something to start. Um, [14:25] and then we would layer on uh AI where it made  sense and allow an opportunity for the human [14:30] in the loop to review what sort of decisions the  agent would have made. Uh, what the agent would [14:36] have inferred from the data that it was given.  Um, so they started to feel comfort from that [14:41] before we just let it loose and start working  within the system. Um we found very quickly [14:47] that uh the trust was there from that that initial  approach working in those waves. Uh because [14:54] when we turned on the agentic approach and we said  okay you can review what the agent will do before [15:00] it performs that activity. Uh it wasn't long  after that they said just just let it do it right. [15:04] We're tired of having to monitor and check on this  every week. It's right most of the time. Um and so [15:11] between demonstrating that to build the trust and  then with that again those weekly reports that [15:16] you know with our partnership at EY they develop  very clear reporting what what did the agents do [15:23] who triggered these actions what's the audit  trail of what it performed you know very easy [15:27] for them to see uh what's taking place and with  that sort of transparency in the process not a [15:33] black box so the agent's just working and it  comes out and it's done you know we have full [15:38] traceability to everything that's happening and  sharing that directly with the business. It's not [15:42] an IT report that comes to IT. It's a report that  goes to the finance team and they can actually see [15:47] what's happening. And that's fascinating because  the other thing that that we saw each other [15:51] together is that it's effectively creating like  a new way of working, right? Like today you are [15:57] working with Joe, Jenny, and Mary and now you're  going to be working with a technology agent like [16:02] like an agent. So building that culture of a new  way of working is very important and and I think [16:09] like your organization did a fantastic job of  embracing hey we have a new way of working now is [16:15] a hybrid approach of humans plus machines and AI  driving a lot of of the workflow and I know we're [16:23] getting close to the end of our conversation and I  want to close with this point um and I want to ask [16:29] you this question there is a lot of people that  join this conversation because they want to learn [16:34] a little bit about how you guys are scaling with  AI and what you guys did to be successful because [16:39] we started this program a few years ago and it  has been evolving as a new technology is becoming [16:43] available we're introducing new capabilities and  making it smarter and smarter more autonomous [16:49] but for everybody in the room for the leaders  in the room what do you think are like the three [16:53] things that you do on a Monday morning if you want  AI initiatives to move beyond pilots and deliver [17:00] value sure I think the first thing is to look  at your look at where you can add value, right? [17:06] Look at your KPI KPIs, determine what metrics you  want to influence um and start there. You know, [17:12] don't just apply an AI solution otherwise  it's just a science experiment at that point [17:17] if you can't really measure the value from what  you're attempting to automate. Um I think it [17:23] from there you then start to dig into your process  and your people and determine you know are our [17:28] processes ready for the for automation? are do we  need to do a improvement activity to standardize [17:34] certain activities and are the people looking to  make these type of improvements are they ready? [17:39] Um I think and then one of the things that we  did is we iterated right I think that was what [17:45] made it so successful is that we started um in  a in a somewhat small approach but we iterated [17:51] rapidly when we started to see that value and we  started to really get that buy in and adoption. [17:56] Um so once we had um the visibility to those  things it just started to naturally snowball [18:02] you know from there. Yeah. No and I mean it it  has been a very comprehensive journey right and [18:08] uh of course the technology is important the  architecture is critical making sure that you have [18:13] the right technology selection that you have the  guardrails to keep track of your cost right like [18:18] because the moment you're consuming the technology  everything that we do by now is by consumption [18:23] right so making sure that you have all of those  guardrails is going to be very important but I [18:28] think like Joe one of the big lessons learned is  about the importance of actually preparing the [18:33] organization and and the people right like make  sure that the people is behind the initiative [18:38] and that um those individuals feel supported.  I think your organization did a terrific job [18:44] with disability making sure that they understand  what is happening early communication not just [18:50] after the automation or the agent is working in  production. Maybe you can speak a little bit about [18:55] how the folks got involved during the process of  the design as well. Yeah, it was really about, [19:02] you know, showing us what they did, the day-to-day  work. And I think one of the biggest things [19:05] talking about that is, you know, we celebrated  the wins. You know, we communicated that out [19:10] to the business. It it wasn't an agent launch  in isolation and just performing an activity. [19:14] We advertised these things, right? Every  time we launched a new automation or a new [19:17] agent uh we were proud to announce that and  in the announcement, it was the picture of [19:23] the business users, right? It wasn't the picture  of the IT team that built it. It was a picture [19:26] of the business users that helped us design it  and implement that solution um to really foster [19:31] that excitement across the business. So um that  was a huge part of it and I think continues to [19:36] Amazing. Joe, uh we're at time. Thank you so much  for for joining and uh thank you Sher. Thank you.