AI News
官方更新 Claude Blog 2026-08-25T00:00 正文 ok 1593 words

Anthropic 一名区域营销人员如何使用 Claude Code 向每位销售代表发送每周个性化更新

英文标题：How an Anthropic field marketer uses Claude Code to send weekly personalized updates to every sales rep

Anthropic 的一名区域营销人员借助 Claude Code，将原本每周耗时数小时的销售更新流程自动化，为每位销售代表生成基于其账户的个性化 Slack 周报。该系统通过 MCP 连接 BigQuery，并结合 CRM 中的销售区域、账户信息及 Slack 更新，匹配活动、网络研讨会、内容资源和后续行动建议。团队通过小范围试点和持续反馈，将错误修正固化为明确规则，例如绝不编造 URL、核验活动受众与联系人职位、动态读取数据表头。该机制已推广至销售、BDR、客户成功和联盟团队，并曾在一周内将一场高管晚宴的报名人数提升一倍。
打开原文相关 6 · 机会 1
media11.8 KB media1.2 MB media403 B media17.7 KB media9.6 KB media49.2 KB media63.8 KB

中文

作为一名营销人员，我面临的最大挑战之一，是让销售团队及时了解区域市场中正在发生的一切。大多数营销人员都熟悉这样的走廊对话：一名销售代表说：“哦，我从没听说过那个活动。”（或者是那份新白皮书、那场网络研讨会。）此时你会意识到，自己错过了一次向销售代表、进而向客户分享最新工作的机会。

我最初的解决方案，是许多营销人员都会采用的做法：每周一早晨与销售团队进行 15 分钟站会。周日晚上，我会汇总公司各处的更新，将其制作成适合展示的幻灯片；然后在会议中现场讲解，并在 Slack 中分享演示文稿。任务完成了，对吗？并不完全是：有了 Claude 后，这一切显得过于手工化；随着团队规模扩大、我开始支持多个销售团队，我的幻灯片流程已经难以跟上。更新内容也变得不那么有用，因为我不再有时间挑选真正适合每个团队的机会。

我希望 Claude 来完成这项工作，并为每位销售代表打造更好的“产品”：一份根据他们负责的账户量身定制、并匹配我们所有营销活动的每周摘要。

幸运的是，我们组织了一场营销黑客松：专门腾出时间，使用 Claude Code 重建可重复的流程和工作流。我和团队聚在一起，为这个问题投入了一个小时，而这带来了巨大改变。黑客松这类轻松、通常由同伴主导的学习机会，能让人们进行实验和探索；在日常工作中，你原本可能不会专门留出这些时间。我们的团队也不例外。

你不需要会编程，你需要会解释

同行营销人员最常问我的问题之一是：“我该如何开始使用 AI？”我的方法，尤其是在使用 Claude Code 时，是先通过 prompt 向 Claude 说明：尽管我没有技术背景，但我面临一个具体挑战；Claude 应将我视为一名深刻理解业务问题的产品经理，并与我逐步协作。我习惯边想边说，因此经常录下自己解释问题的音频，再把转录文本交给 Claude；这样一来，Claude 就能获得全部业务背景。

对于我们团队的每周 AE 摘要，我先向 Claude 概述目标：向每位销售代表发送一条每周 Slack 消息，说明营销团队正在开展什么工作，以及这些工作将如何帮助他们的客户。接着，我编写了一份虚构的每周更新，作为 Claude 要实现的模板。我知道销售代表以行动为导向，因此我用“本周最重要的三件事”列表开头，其中包含三项行动事项，例如即将举办的活动或近期内容，他们可以将其分享给客户。我还为经理汇总版本单独编写了模板，因为经理通常需要团队的整体视图，而不只是各个账户的情况。

接下来，我通过 MCP 将 Claude 连接到 BigQuery；BigQuery 是我们营销团队的单一事实来源，能够提供来自 HubSpot、Clay 和 Salesforce 的细粒度数据洞察。我希望从简单的部分开始，因此先接入了活动和网络研讨会的单一事实来源。为了让每项更新实现个性化，我让 Claude 从 CRM 中提取销售代表的负责区域，以及 Slack 中传达的所有相关账户更新。这样，Claude 就能综合解析这两类信息，创建个性化的每周更新。

随着时间推移，我与营销团队中的其他团队协作来丰富数据，因此这份简报如今还涵盖博客文章、电子书、客户案例、网络研讨会，甚至合作伙伴生态系统中的活动等新内容。

用户反馈才是真正的 prompt engineering

为了将这项机制推广给一线团队，我先从一个同意充当测试组的销售团队开始。如果出错，先发给 10 人的小组不会那么令人不安，而且这个小组承诺会提供反馈。首次发送后，我做了几项调整。

有些问题只是错误。例如，当某个活动在源数据表中没有 URL 时，Claude 会生成一个看起来合理、但实际上无法访问的链接。我们立即将其写入 prompt，作为一条硬性规则：绝不编造 URL。现在，只有当链接地址逐字符来自源数据表时，链接才会显示。后续版本则彻底从简报中删除了没有链接的活动，因为我们意识到，销售人员无法为任何人报名的活动只会带来噪音。

第一周结束时，prompt 中已包含九条内容规则，每一条都可以追溯到销售人员或经理的一项反馈。一名销售人员指出，系统为一位工程副总裁推荐了一场面向知识工作者的研讨会，因此现在会将联系人职位与活动的目标受众进行核验，并且不作说明地剔除不匹配项。行业规则会让零售账户不再收到金融晚宴邀请；而尚未拥有账户的新销售人员，则会收到简短的欢迎信息，而不是一条空白消息。

另一些问题则是数据问题。任何做营销的人都知道，维护单一事实来源有多难。例如，区域活动数据表的列顺序在六周内已被调整过三次。为此，我们修改了 prompt：每次运行开始时，先读取数据表的表头行，并在生成任何内容前验证列映射关系。我们不再硬编码“查看 C 列”，而是使用类似“查看包含活动 URL 的列”这样的指令。

将摘要推广至整个业务部门

在完成这些初步运行后，我将摘要扩展到我支持的每一个团队；如今，区域营销团队为所有销售人员运行这一流程。每周一早晨，多个 Anthropic 销售细分团队的客户经理都会在 Slack 中收到一条私信，其中列出本周三项优先行动、与其账户相关的区域活动、已报名即将举行网络研讨会的联系人、可分享的相关营销内容，以及其他后续建议。

每条消息都根据收件人自己的账户列表生成，因此没有两条消息是相同的。这份摘要正在发挥作用：最近，我们仅仅因为在周一早晨让正确的销售代表看到了正确的活动，就在一周内将一场高管晚宴的报名人数提升了一倍。

当 Anthropic 的业务发展代表（BDR）希望拥有自己的摘要版本时，我们为他们复制了 prompt，只修改了一个字段，因为在我们的 CRM 中，BDR 与账户的映射关系不同于客户经理。prompt 的结构和内容规则完全沿用，BDR 版本在两天内就上线了。此后，我也为客户成功团队和联盟团队完成了同样的工作，并为销售以外的其他跨职能合作伙伴提供所有营销活动的概览。

无论业务节奏有多快，在 Claude 的帮助下，我和团队都会确保销售代表在周一开始工作时，清楚地知道本周将发生什么，以及应优先关注哪些账户和活动。每周一发送的内容都会被完整归档，因此我可以调出任意销售人员在任意日期收到的确切内容；经理也能通过单一汇总视图查看整个团队的推荐事项。我仍然会阅读发送出去的内容，不过系统已不再等待我的审批。几周前我休假时，周一的发送任务自行顺利完成，没有出现任何问题。

周一简报示例，采用 UI 模拟图展示；其中使用的合成数据不代表真实公司或个人。

开始使用 Claude 的最佳实践

下面分享一些受我使用 Claude Code 经验启发的技巧和建议：

  1. 从小处着手，选择你已经在手动完成的事情。 当外界充斥着人们使用 AI 的各种说法时，起步可能很困难。我的建议是：选择你投入最多实际操作时间的重复性任务，并让 Claude 重建这个流程。这样，你能够判断输出结果，因为你本来就知道什么是好的结果。如果问题仍然显得过大，可以将 Claude 作为思考伙伴，帮助你将其拆解为多个步骤。如果任务涉及与其他人共享，先将早期运行结果发给自己，以便在其他人看到之前发现错误。
  2. 用自然语言编写指令，并为每份文档建立版本。 像向新同事交代工作一样向 Claude 说明任务，剩下的交给 Claude。指示 Claude 将每次更新保存为带编号的版本，并附上一行说明变更内容的注释，这样你就能保留生成每次历史运行结果所使用的 prompt 记录。我们的版本是一个由同事为各自业务细分团队运行的 Markdown 文件；最初我们使用共享 Google Doc，后来随着需要编辑的人增多，迁移到了 GitHub。
  3. 与一个规模小且投入度高的群体进行试点。 我们最初与少数客户经理进行了测试；我们知道他们愿意投入时间提供反馈，并持续改进报告，帮助我们发现错误，或提出如何扩展覆盖范围和实现更多个性化的建议。
  4. 利用反馈改进 prompt，将每一项修正都纳入明确规则。 当收件人开始向我们分享反馈、且每项修正都变成 Claude 的明确规则时，这份营销简报才真正变得有用。

Claude 自动化了一个过去每周日都要花费我数小时的手工流程。但通过这个项目，我和团队获得的不只是时间：我们的产出如今更个性化、更有用，也更可衡量。你能用 Claude 改进哪一项营销流程？

立即开始使用 Claude Code。

English / 原文

One of the biggest challenges I’ve faced as a marketer is keeping the sales team up to date with everything that’s going on in the field. Most marketers know the hallway conversation where a sales rep says, “Oh, I never heard about that event” (or that new whitepaper, that webinar) and you realize you’ve missed a chance to share the latest work with sales reps, and in turn, your customers.

My initial solution was one many marketers will recognize: the 15-minute Monday morning stand-up with the sales team. I spent Sunday evenings collating updates from across the business and turning them into presentable slides, and then delivered the info live in the meeting and shared the deck in Slack. Job done, right? Not quite: with access to Claude, this all felt overly manual and as our team grew and I started supporting multiple sales teams, my slide routine couldn’t keep up. The updates were also becoming less useful, because I no longer had time to pick out the opportunities that were right for each team.

I wanted Claude to do the work and create a better “product” for each sales rep: a weekly digest tailored to their accounts and matched to everything we had going on in marketing.

Thankfully, we had organized a marketing hackathon: dedicated time to rebuild repeatable processes and workflows with Claude Code. I huddled with my team and we dedicated an hour to this problem, which made all the difference. Casual, often peer-led learning opportunities like hackathons allow for experimentation and exploration you wouldn’t otherwise carve out time for in your day to day, and our team was no exception.

You don't need to code, you need to explain

One of the biggest questions I get from fellow marketers is, “How do I get started with AI?” My approach, especially with Claude Code, is to open with a prompt explaining to Claude that although I’m not technical, I have this specific challenge, and Claude should treat me as a product manager who deeply understands the business problem, and work with me step by step. I think out loud, so I'll often record myself explaining the problem and give Claude the transcript; that way, Claude has all the business context.

In the case of our team’s weekly AE digest, I started by outlining the goal to Claude: a weekly Slack message to each rep on what’s happening in marketing and how it would help their customers. I then wrote a fake weekly update to give Claude a template to work towards. I know sales reps are action-oriented, so I started with a “top three things for the week” list, featuring three action items, such as upcoming events or recent content, they can share with their customers. I also wrote a separate template for manager roll-ups, since managers typically want a holistic view of their team rather than just individual accounts.

Next, I connected Claude to BigQuery via MCP; BigQuery is our marketing team’s source of truth, offering granular insights into data from HubSpot, Clay, and Salesforce. I wanted to start simple, so I began with our single source of truth for events and webinars. To personalize each update, I had Claude pull the rep’s territory from our CRM and any relevant account updates communicated in Slack.That way, Claude can parse the two together to create a personalized weekly update.

Over time, I’ve worked with other teams across marketing to enrich the data, so the briefing now includes new content like blog articles, and ebooks, customer stories, webinars, and even events from our partner ecosystem.

User feedback is the real prompt engineering

To roll this out to the field, I started with one sales team that agreed to be the test group. Sending to a group of 10 people felt less daunting in case errors came up, and the group was committed to providing feedback. After the initial send, I made a few tweaks.

Some issues were just errors. For example, where an event had no URL in the source sheet, Claude composed a plausible-looking one that led nowhere. We immediately wrote it into the prompt as a hard rule: never invent a URL. A link now renders only if the address comes character for character from the source sheet. A later version dropped linkless events from the briefing entirely, because we realized that events for which our sellers can't register anyone are just noise.

By the end of the first week, the prompt held nine content rules, each traced to a piece of feedback from a seller or a manager. A seller flagged an engineering VP recommended for a workshop aimed at knowledge workers, so contact titles are now checked against an event's intended audience, and mismatches are dropped without comment. An industry gate keeps retail accounts off finance dinner invitations, and brand-new sellers who don’t have accounts yet get a short welcome note instead of a blank message.

Other issues were data problems. Anyone in marketing knows how hard it is to maintain a single source of truth. The field events sheet, for example, has had its columns rearranged three times in six weeks. To plan for that, we changed the prompt to open every run by reading the sheet's header row and verifying the column map before composing anything. Instead of hard-coding “look at Column C,” the instruction is now something like, “Look at the column with the event URL.”

Rolling the digest out across the business

After these initial runs, I expanded the digest to every team I support, and field marketing now runs it for all of sales. Every Monday morning, account executives across several Anthropic sales segments open Slack to a direct message that lists three priority actions for the week, field events for their accounts, contacts who have already registered for upcoming webinars, relevant marketing content to share, and other follow-up suggestions.

Each message is composed from the recipient's own account list, so no two messages are alike. The digest is working; we recently doubled registrations for an executive dinner in a week, purely because the right reps had the right event in front of them on Monday morning.

When Anthropic's business development representatives (BDRs) wanted their own version of the digest, we duplicated the prompt for them with a change in one field, since BDRs map to accounts through a different relationship in our CRM than account reps do. The prompt structure and content rules carried over unchanged, and the BDRs were live within two days. I’ve since done this for the customer success and alliance teams too, and I also provide an overview of all marketing activities for other cross-functional partners outside sales.

No matter how fast the business moves, my team and I, with Claude’s help, make sure that sales reps start their Monday knowing exactly what’s happening that week and what accounts and events to prioritize. Each Monday's send is archived in full, so I can pull up exactly what any seller received on any date, and managers see their whole team's recommendations in a single roll-up. I still read what goes out, though the system no longer waits for my approval. When I went on holiday a few weeks ago, the Monday send went off on its own, without a hitch.

An example of what a Monday brief looks like, shown with a UI mockup depicted with synthetic data that does not represent real companies or individuals.

Best practices for getting started with Claude

Below, I share tips and tricks inspired by my own experience working with Claude Code:

  1. Start small, with something you already do manually. It can be hard to get started when there’s so much noise about what people are doing with AI. My advice: pick the repetitive task you spend the most hands-on time on and ask Claude to rebuild it. That way, you’ll be able to judge the output because you already know what good looks like. If the problem still feels too big, use Claude as a thought partner to break it into steps. And if it’s something you share with other people, route the early runs to yourself first so you catch the errors before anyone else does.
  2. Write instructions in plain language and version each document. Brief Claude the way you’d brief a new colleague and Claude will do the rest. Instruct Claude to save each update as a numbered version with a one-line note of what’s changed, so you have a record of the prompts that produced each past run. Ours is a markdown file my colleagues run for their own segments; we started from a shared Google Doc and moved to GitHub once more people needed to edit it.
  3. Pilot with a small, committed group. We ran our first tests with a handful of account executives who we knew would be willing to spend the time on providing us feedback and improving the report over time, helping us detect errors or offer suggestions on how to expand or personalize coverage.
  4. Use feedback to improve your prompt, fold in each correction as an explicit rule. The marketing briefing became useful when the recipients started sharing feedback with us and each correction became an explicit rule for Claude.

Claude automated a manual process that used to take me hours each Sunday, but with this project, my team and I have gained something much better than time: our output is now more personal, more useful, and more measurable. What marketing process can you improve with Claude?

Get started with Claude Code today.