# How to Transform a Company With AI
**作者**: Varick Agents
**日期**: 2026-05-26T22:14:37.000Z
**来源**: [https://x.com/varickagents/status/2059397823674958265](https://x.com/varickagents/status/2059397823674958265)
---
Varick leads AI transformations helping companies unlock nine-figure efficiency gains by redesigning how work gets done. By the end of this article, you should understand how a business should be rebuilt from the ground up around AI to unlock value at this scale.
You'll know how to identify which workflows in your business are worth automating, and how to redesign them in a way that doesn't disrupt operations.
Follow this account for more break downs and case studies of how AI is being adopted at the largest companies across America. If this work excites you, we're hiring across the board - join us at varickagents.com/careers.
## Intro to Transformations
You will not transform your company without rebuilding operations from the ground up.
The industrial revolution taught us that productivity gains don't show up if you don't do this. For 30 years, factories swapped out the steam engine with electric motors and saw little financial benefit. Old factories were built around one central steam engine in the basement that powered every machine in the building.
When electricity came along, factories just replaced that engine with a motor and changed nothing else. They kept the same building, the same layout, and continued working the same way.
What moved the needle was a complete operational redesign from the ground up around electricity. The real unlock was that electric motors could be small and cheap, so every machine was able to have its own. That meant factories no longer had to be built around a single power source, and they could spread out and put machines in the order that work actually flowed. This led to the assembly line, which created major productivity gains.
Henry Ford figured out that you need to rebuild around the technology to create value in the early 1900s, and we're using that same operational redesign playbook today.
## Buying software doesn't get you there
Which brings us back to today. Most companies trying to transform with AI are hoping to buy their way there by swapping their SaaS stack for AI tools. Agentic software seats, Copilot licenses, and no-code workflow builders rarely move the needle on their own, because transformation is not a piece of software you can purchase. It's a structural change in how the business operates, and it starts with the people and the processes that run it.
If the AI does not understand the underlying process, it will not create meaningful value. And if the people who own that process are not brought along, adoption will be weak even if the technology works.
That's why you need to spend a few weeks with teams across the business — from accounts payable, to procurement, to operations — and understand how their work actually gets done from end to end.
You should map every workflow, figure out what the ROI of an agent would be in each particular workflow and how to approach it from an engineering perspective, then choose where to deploy the agents where they'd be a good fit (which we'll go over later).
From there, capture company context (tribal knowledge) and convert it into rules, instructions, and decision logic that the agents can follow.
Doing this with each team is the only way to get the context required to redesign the business around AI, and the buy-in required for the transformation to actually stick.
## The Operational Redesign
Once every process is mapped end to end, the next step is deciding which workflows should actually be redesigned around AI. This is the operational redesign.
Take it from an agent company — please do not put agents in every workflow. There is a point where agents create more problems than they solve.
In this section, we'll cover how to create value from an AI transformation and how to not disrupt the business while doing so.
How to create value
A transformation is about redesigning each workflow so deterministic work is automated, judgment work is handled by AI where appropriate, and high-risk, high-judgment decisions remain with humans.
Done correctly, this does more than cut cost. The agents should work to give people better context, and better context helps people make better decisions, faster. Consistently better decisions unlock revenue growth. This means that a proper transformation should yield both topline growth and efficiency gains.
We saw this clearly in a sales transformation for a multibillion-dollar revenue enterprise software company. Too much of the sales process was trapped in busy work, and large deals touched six teams across eleven handoff points. So we ran through the entire process outlined in this article, found the right workflows to automate (either with agents or with scripts), and delivered $25m in value in the first year through margin expansion (revenue growth + savings). Proper transformation isn't just a cost-cutting initiative.
How to pick the right workflow
One of the most important parts of an AI transformation is choosing the right workflow to redesign first. Not every process is worth automating, and not every process is a good fit for agents.
The best workflows usually have a few things in common: high volume, lots of manual effort, fragmented systems, repeated handoffs, tribal knowledge, and clear financial impact.
You're looking for places where work is already happening over and over again, but the process is messy enough that traditional automation hasn't solved it. Think data moving through email, Slack, spreadsheets, portals, and ERP systems.
A good workflow to redesign usually has four traits:
1. It happens often enough to matter. The process should run hundreds or thousands of times a month, or touch enough revenue or cost that improving it creates real value.
2. It has repeatable decisions. The work does not need to be identical every time, but it should follow patterns. Agents are most useful when they can learn from past decisions, apply business rules, and route exceptions.
3. It depends on context spread across systems. The more humans are searching between tools to gather information, the more valuable an agent can be. AI is especially useful when the work requires pulling context from contracts, emails, CRM records, ERPs, documents, and internal rules.
4. It has measurable pain. You should be able to measure the current cost of the workflow (cycle time, error rate, manual hours, delayed revenue, duplicate payments, approval delays, etc.) before and after deployment.
The goal is to separate the work into three buckets: what can be handled with deterministic automation, what should be handled by agents, and what has to stay with humans.
Agents need to self-improve
Make sure to build human-in-the-loop feedback into the system from the start. During training and shadow mode, humans can approve, reject, or correct an agent's actions. Always log the agent's output, the human's response, and the surrounding context so the system can improve over time.
This makes agents meaningfully more accurate after deployment. In the same sales case, accuracy increased by 10% within a few weeks, and the result was an increase in the amount of work that agents can handle autonomously, which led to higher net dollar value created.
How to not disrupt the business when transforming
You need to ensure that the transformation doesn't create complications and is as cost-effective as possible. A couple of steps we live by:
Don't force massive migrations. Most companies have already spent years moving onto systems like Salesforce and NetSuite. Forcing companies to rip out and replace their software just to adopt AI slows the transformation down and forces teams to relearn the software they depend on. At Varick, we strongly encourage building on top of the systems already in place, whether through APIs or computer-use agents. This avoids expensive data migrations, keeps the workflows the business already runs on, and allows the operational redesign to survive even if the underlying software changes later.
Know the data and keep it segmented. In most workflows, the data that powers the transformation falls into four categories: the system of record, the business rules, the raw intake data, and the feedback or memory the agent accumulates over time. Keeping those layers separate is very important. It means an operations person can update a rule without calling an engineer, and it makes the system easier to maintain and scale after deployment. Your goal is to design a transformation so it can keep running smoothly with minimal intervention after deployment.
## Over time, the organization starts to reshape around AI
In the first weeks after deployment, agents run in sandboxes, then in shadow mode alongside humans, and only later in supervised production use cases. As confidence grows, workflows are not just automated but often redesigned and improved.
The goal is to redesign operations where it helps the business move faster and create more value.
When we start with a new company, the first workflows we scope are typically:
- Accounts Payable — Invoice automation, GL coding, purchase order matching, etc.
- Procurement — Vendor onboarding, supplier scorecards, contract compliance, etc.
- Sales — Deal desk routing, CRM enrichment, forecast intelligence, commission calculation, etc.
- Operations — Exception detection and routing, allocation optimization, returns disposition, etc.
The company gets used to handing off busy work to AI and owning real work, and efficiency gains start showing up on the P&L in weeks.
This is what the outcome of your transformation should look like if done correctly. The process takes its time because it's the foundation to becoming an AI-native company, but when that happens, financial uplift shows up fairly quickly. Make sure you're building your way there workflow by workflow, function by function. It'll take its time, but good transformations are well worth the investment.
Every lesson in this guide has been learned over dozens of large-scale transformations we've done for clients. If your business is doing over a billion dollars in revenue and wants to see the results our clients are seeing, we're now accepting engagements for July. Find us at varickagents.com
## 相关链接
- [Varick Agents](https://x.com/varickagents)
- [@varickagents](https://x.com/varickagents)
- [275K](https://x.com/varickagents/status/2059397823674958265/analytics)
- [varickagents.com/careers](https://varickagents.com/careers)
- [varickagents.com](https://varickagents.com/)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [6:14 AM · May 27, 2026](https://x.com/varickagents/status/2059397823674958265)
- [275.5K Views](https://x.com/varickagents/status/2059397823674958265/analytics)
- [View quotes](https://x.com/varickagents/status/2059397823674958265/quotes)
---
*导出时间: 2026/5/27 19:01:18*
---
## 中文翻译
# 如何利用 AI 转型公司
**作者**: Varick Agents
**日期**: 2026-05-26T22:14:37.000Z
**来源**: [https://x.com/varickagents/status/2059397823674958265](https://x.com/varickagents/status/2059397823674958265)
---
Varick 主导 AI 转型，通过重新设计工作完成方式，帮助企业实现数亿美元的效率提升。阅读完本文后，您应该了解如何围绕 AI 从零开始重建一家企业，以释放这一量级的价值。
您将知道如何识别企业中哪些工作流值得自动化，以及如何在不中断运营的前提下重新设计它们。
请关注本账号，获取更多关于美国最大企业如何采用 AI 的解析和案例研究。如果您对这项工作充满热情，我们正在全方位招聘——请访问 varickagents.com/careers 加入我们。
## 转型入门
如果不从零开始重建运营体系，就无法实现公司的转型。
工业革命告诉我们要想获得生产力提升，必须这样做。在那 30 年里，工厂用电机替换了蒸汽机，却几乎没有看到任何经济效益。旧工厂的设计围绕着地下室里的一台中央蒸汽机，由它为大楼内的所有机器提供动力。
当电力出现时，工厂只是简单地用电机替换了那台引擎，而没有任何其他改变。他们保留了相同的建筑、相同的布局，并继续以相同的方式工作。
真正带来改变的是围绕电力进行的从零开始的全面运营重新设计。真正的突破在于电机可以变得小而便宜，因此每台机器都可以配备自己的电机。这意味着工厂不再需要围绕单一电源建造，它们可以分散布局，按照工作实际流动的顺序排列机器。这催生了流水线，从而带来了巨大的生产力提升。
亨利·福特（Henry Ford）在 20 世纪初就意识到，你需要围绕这项技术进行重建才能创造价值，而我们今天正在使用同样的运营重新设计策略。
## 购买软件无法达成目标
这让我们回到了今天。大多数试图利用 AI 进行转型的公司都希望能通过将 SaaS（软件即服务）栈替换为 AI 工具，用钱买到这种转型。智能软件席位、Copilot 许可证和无代码工作流构建器本身很难带来实质性改变，因为转型不是你可以购买的一件软件产品。它是企业运营方式的结构性变革，而且这种变革始于管理它的人员和流程。
如果 AI 不理解底层的业务流程，它就无法创造有实质意义的价值。而且，如果负责该流程的人员没有被带动起来，即使技术再有效，落地应用也会非常薄弱。
这就是为什么您需要花几周时间与跨部门的团队——从应付账款、采购到运营——在一起，深入了解他们的工作在端到端的过程中究竟是如何完成的。
您应该绘制每一个工作流的图谱，弄清楚在特定的工作流中智能体的投资回报率（ROI）是多少，以及如何从工程角度来处理它，然后选择在那些合适的地方部署智能体（这一点我们稍后会详细讨论）。
然后，捕捉公司的上下文（隐性知识），并将其转化为智能体可以遵循的规则、指令和决策逻辑。
与每个团队一起完成这项工作，是获得围绕 AI 重新设计业务所需的上下文，以及确保转型真正落地所需的员工认同感的唯一途径。
## 运营重新设计
一旦每个流程都完成了端到端的梳理，下一步就是决定哪些工作流实际上应该围绕 AI 进行重新设计。这就是运营重新设计。
作为一家智能体公司，我们要说——请不要在每一个工作流中都安放智能体。到了某个节点，智能体制造的问题会多于其解决的问题。
在本节中，我们将介绍如何从 AI 转型中创造价值，以及在此过程中如何不干扰业务。
如何创造价值
转型意味着重新设计每个工作流，使确定性工作被自动化，判断性工作在适当的情况下由 AI 处理，而高风险、高判断力的决策仍保留在人类手中。
如果做得正确，这不仅仅是削减成本。智能体应该致力于为人们提供更好的上下文，而更好的上下文能帮助人们更快地做出更好的决策。持续更好的决策能解锁收入增长。这意味着，一次成功的转型应该能同时带来营收增长和效率提升。
我们在一家营收数十亿美元的企业软件公司的销售转型中清楚地看到了这一点。销售过程中有太多时间被琐事占据，大额交易涉及 6 个团队，经过 11 个交接点。因此，我们按照本文概述的整个流程运行了一遍，找到了合适的自动化工作流（通过智能体或脚本），并在第一年通过利润率扩张（收入增长 + 节省）创造了 2500 万美元的价值。恰当的转型不仅仅是一项削减成本的举措。
如何选择正确的工作流
AI 转型最重要的部分之一是选择正确的工作流首先进行重新设计。并非每个流程都值得自动化，也并非每个流程都适合智能体。
最好的工作流通常有几个共同点：量大、大量人工投入、系统碎片化、反复交接、依赖隐性知识以及清晰的财务影响。
您要寻找的是那些已经在反复发生，但流程又足够混乱，以至于传统自动化无法解决的问题。想一想在邮件、Slack、电子表格、门户和 ERP 系统之间流动的数据。
一个值得重新设计的好工作流通常具有四个特征：
1. 它发生的频率足以产生价值。该流程应该每月运行数百或数千次，或者涉及足够的收入或成本，以至于改进它能创造真正的价值。
2. 它具有可重复的决策。工作不需要每次都完全相同，但应该遵循某种模式。当智能体能够从过去的决策中学习，应用业务规则并处理异常情况时，它们是最有用的。
3. 它依赖于分散在各系统中的上下文。人类越是需要在工具之间来回搜索以收集信息，智能体的价值就越高。当工作需要从合同、电子邮件、CRM 记录、ERP、文档和内部规则中提取上下文时，AI 特别有用。
4. 它有可衡量的痛点。您应该能够衡量部署前后的工作流当前成本（周期时间、错误率、人工工时、收入延迟、重复付款、审批延迟等）。
目标是将工作分为三类：可以通过确定性自动化处理的工作，应该由智能体处理的工作，以及必须由人类保留的工作。
智能体需要自我改进
务必从一开始就在系统中构建人机协同的反馈机制。在训练和影子模式期间，人类可以批准、拒绝或纠正智能体的行动。始终记录智能体的输出、人类的反应以及周围的上下文，以便系统能够随着时间的推移而改进。
这使得智能体在部署后的准确性显著提高。在同样的销售案例中，几周内准确率提高了 10%，结果是智能体可以自主处理的工作量增加了，从而创造了更高的净美元价值。
如何在转型时不干扰业务
您需要确保转型不会造成并发症，并尽可能具有成本效益。我们要遵循几个步骤：
不要强制进行大规模迁移。大多数公司已经花费数年时间迁移到了 Salesforce 和 NetSuite 等系统上。强迫公司为了采用 AI 而拆除并替换其软件，只会拖慢转型速度，并迫使团队重新学习他们依赖的软件。在 Varick，我们强烈建议在现有系统之上构建，无论是通过 API 还是计算机使用智能体。这避免了昂贵的数据迁移，保留了业务已经运行的工作流，并允许即使底层软件未来发生变化，运营重新设计依然可行。
了解数据并保持分层。在大多数工作流中，驱动转型的数据分为四类：记录系统、业务规则、原始摄入数据，以及智能体随时间积累的反馈或记忆。将这些层分开是非常重要的。这意味着运营人员可以在无需呼叫工程师的情况下更新规则，并使系统在部署后更易于维护和扩展。您的目标是设计一种转型，使其在部署后能在最少干预的情况下平稳运行。
## 随着时间推移，组织开始围绕 AI 重塑
在部署后的最初几周，智能体在沙盒中运行，然后与人类并行进入影子模式，最后才进入受监督的生产用例。随着信心的增长，工作流不仅被自动化，而且通常被重新设计和改进。
目标是重新设计运营，从而帮助公司更快地发展并创造更多价值。
当我们与新公司合作时，我们首先规划的典型工作流包括：
- 应付账款 —— 发票自动化、总账编码、采购订单匹配等。
- 采购 —— 供应商入职、供应商记分卡、合同合规等。
- 销售 —— 交易台路由、CRM 丰富、预测智能、佣金计算等。
- 运营 —— 异常检测和路由、分配优化、退货处置等。
公司开始习惯于将琐事交给 AI，自己掌握真正的工作，效率提升在几周内就开始体现在损益表（P&L）上。
如果操作正确，这就是您的转型结果应有的样子。这个过程需要时间，因为它是成为 AI 原生公司的基础，但当这一点实现时，财务收益会相当快地显现。请确保您是通过逐个工作流、逐个职能的方式构建这一路径。这需要时间，但好的转型是非常值得投入的。
本指南中的每一个教训都是我们从为客户进行的数十次大规模转型中总结出来的。如果您的企业营收超过 10 亿美元，并希望看到我们客户所看到的成果，我们现在正在接受 7 月份的项目委托。请访问 varickagents.com 找到我们。