AI News
官方更新 Claude Blog 2026-10-01T00:00 正文 ok 1451 words

Anthropic 销售团队如何用 Claude Managed Agents 重塑入站销售

英文标题：How Anthropic's sales team rebuilt inbound with Claude Managed Agents

Anthropic 构建了基于 Claude Managed Agents（测试版）的购买 Agent，帮助企业客户从咨询 Claude 方案到完成购买，并每天处理数千场对话。该 Agent 能回答定价、安全、数据和席位等问题，复杂交易则连同完整上下文转交销售代表。与旧版联系销售表单相比，经 Agent 转交的线索转化为销售机会的概率超过两倍，成交速度约快 5 天。Anthropic 表示，这一实践让销售人员能减少重复答疑，将更多时间投入高价值客户沟通，同时为企业部署面向客户关系全流程的 Agent 提供了参考。
打开原文相关 6 · 机会 2
media8.7 KB media93.4 KB media244.7 KB media190.7 KB media403 B media8.8 KB media9.9 KB media17.7 KB media6.9 KB media53.6 KB

中文

作为一名销售负责人，我不得不承认，不久前，那些希望为公司购买 Claude 的用户还无法足够快地获得所需答案。他们填写了我们的“联系销售”表单，却要等待太久才能收到回复，有时甚至长达数天。他们的大多数问题其实都很简单：某个方案的价格是多少、是否设有最低席位数，或者我们能否满足 HIPAA 的合同要求。答案都已在我们的文档和支持文章中，但客户希望有人能快速为他们讲解。

因此，我们在 Claude Managed Agents（测试版）上构建了一个购买 Agent，帮助潜在客户从“我想为公司购买 Claude”走到完成购买。该购买 Agent 现在每天进行数千场对话，回答客户最迫切的问题，并引导他们完成结账流程。被它升级转交给我们团队的客户，已经获得更多信息，也更准备好作出购买决定：与旧表单带来的线索相比，他们转化为销售机会的比例超过两倍，成交速度也快约五天。我们特意将这一体验设计为自愿选择模式，也就是说，客户可以在一开始根据自己的偏好，选择与 Agent 或销售代表沟通。

为什么旧有的入站流程失效了

在构建购买 Agent 之前，我们的入站流程相当直接：买家填写表单，线索被分配给业务发展代表（BDR），由其进行资格筛选后再转交给客户经理（AE）。当团队能够与每一位对产品感兴趣的客户交流时，这套流程依然运行良好。

但我们每月收到数以万计的入站请求，入站 BDR 团队无法跟上。销售代表每天都在回答文档中已有说明的问题，而我们也没有有效的方法触达队列中的所有客户。这并不是我们希望为客户提供的体验。

我们希望每位客户都能在任何时间、使用任何语言获得专业解答，并能以他们偏好的方式购买——这通常意味着无需与销售代表交流。我们希望销售代表将时间投入到那些能够改变交易结果走向的对话中。

弥合缺口的购买 Agent

我们将 Agent 部署在客户本来就在提问的地方：我们的联系销售和定价页面、产品内部以及电子邮件中。任何客户都能在我们的 Claude.ai 产品内找到该购买 Agent；这是一个简单的聊天体验，可帮助客户理解哪种方案最适合自己。客户描述团队需求后，Agent 会提出几个后续问题，然后回答有关价格、安全性和数据的问题，再推荐方案与席位数量。每场对话会以以下三种方式之一结束：

  • 购买。 买家直接进入结账流程。
  • 转交。 对于规模更大或更复杂的交易，Agent 会将买家连同完整对话详情转交给销售代表。
  • 快速答复。 有时，客户所需的只是一个答案。

我们构建该 Agent 时，让它遵循销售代表相同的工作逻辑：理解客户希望解决什么问题、他们目前如何使用 Claude，以及我们能够如何提供最佳帮助。

为什么使用 Claude Managed Agents？

从底层看，这个购买 Agent 很简单：一个提示词、少量工具，以及运行在 Claude Managed Agents 上的 Claude。由于该平台负责托管、会话管理和工具编排，我们的工程师得以与销售团队一起专注于设计购买体验，而非构建基础设施。

我们最终选择 Managed Agents，原因如下：

  • 快速投入生产。 一名工程师仅用几周时间，就在 Managed Agents 上构建了初版购买 Agent。
  • 技术与非技术贡献者共同参与。 工程团队负责代码，但销售和内容负责人可直接在 Console 中审阅和编辑系统提示词。变更会先进入我们的预发布 Agent，因此任何人都可以在其触达客户前测试调整效果。
  • 专注领域，而非支撑框架。 Managed Agents 负责运行 Agent 循环、会话和托管。我们将时间花在提示词、工具和知识库上——这三项才是销售场景特有的内容。
  • 版本管理降低迭代成本。 对 Agent 的每项变更都会作为独立版本保存。内部测试开始约一周时，我们已迭代至 v7；上线后，我们仍持续每周发布提示词改动。必要时，我们可以轻松让新会话回退使用旧版本。
  • 灵活支持更多用例。 目前，客户通过我们的网站发现该 Agent。Managed Agents 支持定时运行，这为我们探索其他类型的客户互动方式提供了路径。

构建购买 Agent 时学到的经验

给 Claude 一个目标，而不是一堆规则。 向 Agent 解释目标的效果优于给出详细的指令清单。例如，像“你的目标是理解客户需求、评估潜在客户资格，并推荐最佳方案”这样简单的表述，比在流程图中逐项列出所有资格判定要求更有效。

少即是多。 我们试验了长度、复杂性和结构各异的提示词，发现“少即是多”这一准则最有效。如今的模型能够理解复杂且细致的目标，并从目标反向推导实现路径。在实践中，这意味着不要过度干预 Claude：只要为它提供成功所需的知识和上下文，例如按席位计价涵盖哪些内容、计费周期在何时运行即可。

让领域专家参与开发循环。 在 Agent 编程不断加速开发周期的时代，这一点正变得越来越有价值。借助 Managed Agents 和 Claude Code 节省的时间，我们的工程师能够投入更多时间与销售团队共同打造以客户为中心的体验。我们得以频繁、尽早地测试 Agent，从而实现快速迭代。

聚焦真正适合客户的选择。 该 Agent 经常向小型团队推荐我们的 Team 方案而非 Enterprise 方案，因为在某些情况下，这才是对他们正确的选择。我们认为这是一项功能。买家如果选择了正确方案，而不是被向上销售，更可能持续使用产品并获得更好的体验。

将每一次 Agent 向销售代表的升级转交视为反馈。 每当 Agent 将买家转交给销售代表时，它都会说明原因。在早期，这些原因大多是客户尚无法通过我们的自助产品自行完成的事项。这些原因塑造了我们改进体验的方向，而需要人工协助完成交易的对话占比已下降约一半。

客户体验发生了什么变化

该 Agent 现在全天候每天进行数千场对话。客户可以立即获得问题答案，不必等待预约电话，或在数天内反复往来电子邮件。实际上，许多客户更愿意与 Agent 沟通而非真人，因为 Agent 可以在问题出现时迅速提供有依据的回答。许多新的自助式 Enterprise 客户会在购买前与购买 Agent 交流；我们发现，事先与 Agent 沟通过的客户，对自己购买的内容有更好的理解。

Agent 如何改变我们的销售组织

该 Agent 改变了销售代表每天的工作内容。如今，我们能将团队的大部分时间和精力投入到更早阶段教育客户、进行更多实时客户沟通，以及在线下活动中面对面交流。

最具影响力的变化是，销售代表现在获得了信息更充分的线索；与旧联系表单相比，这些线索更常转化为销售机会，且成交速度约快五天。销售代表开始对话时，已知道客户需要什么、客户此前获得了哪些信息，以及我们能如何提供最佳帮助。

例如，我们的一位内勤销售代表 Ojas 过去通常要往来约 10 封电子邮件才能成交；如今只需约 6 封。自购买 Agent 上线以来，他已将赢单交易的产出提高至原来的 2.5 倍，因为他能够专注于那些信息充分、已准备就绪、只需获得一点帮助即可完成购买的客户。

不久前，入站线索还是一个我们团队无法跟上的队列。现在，它变成了任何买家都能在任何时间发起的对话，而我们的销售代表也能更高效地利用时间。购买 Agent 是我们迈向覆盖整个客户关系的 Agent 体系的第一步。

准备好构建你自己的 Agent 了吗？了解更多 Claude Managed Agents，或联系我们的团队，了解如何将 Agent 引入你的销售流程。

本文由 Carl Johnson 撰写，Izzy Lee、Bobby P.、Lina Ochman、Yana Gevorgyan、Jerico Johns 和 Taylre Duarte 共同贡献。

English / 原文

As a sales leader, it pains me to admit that not long ago, people who wanted to buy Claude for their company weren’t getting the answers they needed quickly enough. They had filled out our Contact Sales form but would wait too long to hear back, sometimes for multiple days. Most of their questions were simple: what a plan costs, whether there's a seat minimum, or whether we can meet HIPAA’s contract requirements. The answers were in our documentation and support articles, but customers wanted someone to walk them through, quickly.

So we built a buying agent on Claude Managed Agents (beta) that takes a prospect from "I want Claude for my company" to a completed purchase. The buying agent now holds thousands of conversations a day, answers customers’ most pressing questions, and gets them through checkout. The customers it escalates to our team arrive educated and more ready to make purchase decisions: they turn into opportunities more than twice as often as leads from the old form, and close about five days faster. We intentionally left this experience as opt-in, meaning our customers choose at the start if they’d like to talk to an agent or sales rep based on their preference.

Why the old inbound process broke

Before building the buying agent, our inbound process was fairly straightforward. A buyer filled out a form. The lead was routed to a business development rep, who qualified it and passed it to an account executive. That still works well when a team can talk to everyone who’s interested in the product.

We had tens of thousands of inbound requests every month, and our inbound BDR team couldn’t keep up. Reps spent their days answering questions the docs already covered, and we didn’t have an effective way to reach all of the customers in the queue. This wasn't the experience we wanted for our customers.

We wanted every customer to get a knowledgeable answer at any hour and in any language, and to be able to buy however they preferred, which often means without talking to a sales rep. We wanted our reps to spend their time on the conversations where they can change the trajectory of the outcome.

A buying agent that closes the gap

We deployed the agent where customers were already asking questions: on our Contact Sales and Pricing pages, inside the product, and in our emails. Any customer can find the buying agent within our Claude.ai product, it’s a simple chat experience to help customers understand what plan is best for them. A customer describes what their team needs, the agent asks a few follow-up questions, and it then answers the ones about pricing, security and data before recommending a plan and seat count. Each conversation ends one of three ways:

  • Purchase. The buyer goes straight to checkout.
  • Hand-off. For larger or more complex deals, the agent passes the buyer to a rep along with the full conversation details.
  • A quick answer. Sometimes that's all a customer needs.

We built the agent to follow the same work of our sales reps, which means understanding what the customer is trying to solve, what their current Claude usage looks like, and how we can best help them.

Why did we use Claude Managed Agents?

Under the hood, the buying agent is simple: a prompt, a handful of tools, and Claude, running on Claude Managed Agents. Because the platform takes care of hosting, session management and tool orchestration, our engineers spent their time with sales designing the buying experience instead of building the infrastructure.

We ultimately landed on Managed Agents for the following reasons:

  • Fast to production. One engineer built the initial buying agent on Managed Agents in just a few weeks.
  • **Tech & non-tech contributors. **Engineering owned the code, but sales and content leads reviewed and edited the system prompt directly in the Console. Changes went to our staging agent first, so anyone could try a tweak before it reached customers.
  • Focus on the domain, not the harness. Managed Agents runs the agent loop, sessions and hosting. We spent our time on the prompt, the tools, and the knowledge base—the three things that are specific to selling.
  • Versioning makes iteration less expensive. Every change to the agent is saved as its own version. We were on v7 about a week into internal testing, and we kept shipping prompt changes weekly after launch. When needed, we could easily point new sessions back to a previous version.
  • Flexibility to support more use cases. Today, customers discover the agent on our website. Managed Agents supports scheduled runs, which gives us a path to explore other types of customer engagement.

What we learned building the buying agent

Give Claude a goal, not rules. Explaining the goal of the agent worked better than a detailed list of instructions. For example, something as simple as "Your goal is to understand customer requirements, qualify prospects, and recommend the best plan" was more effective than listing out every qualification requirement in a flowchart.

Less is more. We experimented with prompts of varying lengths, complexity, and structure, and found the maxim of 'less is more' works best. Models today can understand complex, nuanced goals and work backwards from them. In practice, that has meant getting out of Claude's way by giving it the knowledge and context it needs to be successful, such as what seat-based pricing covers or when billing cycles run.

Use SMEs in the development loop. This is something that is becoming increasingly valuable in a world where agent coding is speeding up development cycles. By virtue of saving time using Managed Agents and Claude Code, our engineers were able to spend more time with the sales team to shape a customer-centric experience . We were able to test the agent early and often, leading to quick iterations.

Focus on what’s right for the customer. The agent often points small teams to our Team plan instead of Enterprise, because that's the right answer for them in some cases. We decided that's a feature. A buyer who lands on the right plan instead of being upsold is more likely to stay and have a better experience.

Treat every agent to rep escalation as feedback. Each time the agent passes a buyer to a rep, it explains why. Early on, most of those reasons were things customers couldn't yet do on their own with our self-service products. Those reasons shaped how we improve the experience, and the share of conversations that needed a person to help close the deal has fallen by about half.

What changed for customers

The agent now holds thousands of conversations a day, around the clock. Customers get their questions answered instantly instead of waiting for a scheduled call or exchanging emails over multiple days. A lot of our customers actually prefer to talk to an agent rather than a human, because it can give them quick, informed answers as questions arise. Many of our new self-serve Enterprise customers talk to the buying agent before they buy, and we've found that those who talked to the agent beforehand had a better understanding of what they were buying.

How the agent changed our sales organization

The agent changed what our reps spend their days. We've now been able to focus the majority of the team's time and energy on educating customers earlier in their buying journey, having more live customer conversations and spending time in person at events.

The most impactful change has been that our reps now get more informed leads, convert leads to sales opportunities more often than our old contact form, and they close about five days faster. A rep starts the conversation already knowing what the customer needs, what they've been told, and how we can best help them.

For example, one of our inside sales reps, Ojas, used to exchange about 10 emails to close a deal; now it takes about six. He’s been able to 2.5x his output on closed won deals since launching the Buying Agent as it allowed him to focus on highly educated and ready customers who just needed help getting across the line.

Not long ago, inbound was a queue our team couldn't keep up with. Now it's a conversation any buyer can start at any hour, and our reps spend their time more efficiently. The buying agent is our first step toward agents across the whole customer relationship.

Ready to build your own? Learn more about Claude Managed Agents, or talk to our team about bringing agents to your sales motion.

This article was written by Carl Johnson, with contributions from Izzy Lee, Bobby P., Lina Ochman, Yana Gevorgyan, Jerico Johns, and Taylre Duarte.