Anthropic 商务拓展团队如何利用 Claude 规模化开展入站与出站业务
英文标题：How Anthropic's business development team uses Claude to run inbound and outbound at scale
Anthropic 的商务拓展代表（BDR）团队将 Claude Cowork 用于销售收件箱回复、线索跟进、Salesforce 更新、账户研究和外呼准备等重复性工作。团队通过销售知识库、个人写作风格档案及定时任务，让 Claude 每小时扫描邮件、起草回复，并在人工审核后发送。对于同时负责逾百个账户的 BDR，Claude 可整合 Salesforce、Apollo、Common Room、Gong 和数据仓库中的信息，生成账户简报、评分与外呼策略。Anthropic 强调，AI 的作用是减少行政与重复劳动、提升客户沟通和战略分析质量，而不是取代人工审核与决策。
8.5 KB
1.0 MB
677.3 KB
876.1 KB
1.2 MB
1.5 MB
1.6 MB
733.7 KB
403 B
13.0 KB
14.0 KB
8.7 KB
10.5 KB
53.9 KB
中文
职业生涯早期，我从事商务拓展工作时，客户经理会交给我包含数百个账户的名单，而我必须逐一研究每家公司、寻找合适联系人、搜集邮箱地址并起草外联内容。入站业务同样存在高度依赖人工且耗时的工作流程。
去年夏天加入 Anthropic 后，我开始负责管理销售收件箱。除管理自己负责的客户组合外，我每天还要花大约 5 小时，手动回复潜在客户的入站咨询；这些咨询往往是在询问有关我们产品的相同或相似问题。
如今，其中许多工作都已被配置为 Claude Cowork 中的技能和定时任务。个性化客户邮件会先准备成草稿，供我在发送前审核和定制。我的出站工作则从一份详细研究开始，而这些研究不再需要我耗费数小时进行整理。
因此，我和团队成员花在手动、重复性工作上的时间更少，能够把更多精力投入真正重要的事情：帮助客户。
以下介绍 Claude Cowork 如何提升 Anthropic 的商务拓展职能，使我们能将更多时间投入战略性工作、理解客户问题，以及帮助客户了解 Claude 能如何解决这些问题。
自动化行政和可重复性任务
BDR 位于销售流程的起点，负责筛选入站需求并为业务建立出站销售管道。在 Anthropic，这些工作动作如今都会先通过 Claude Cowork 运行。
我们入站体系的一个基础组成部分，是一份我汇集的文档，其中包含销售收件箱最常收到的问题，以及针对这些问题的最佳答案。这份文档充当我们的销售知识库，Claude 会在起草任何待发送回复之前阅读它。Claude 帮我创建了这份文档——我只需将其指向相关信息源——如今，它还会持续验证内容是否为最新状态，标记可能已经过时的信息，供用户确认。
基于该文档构建的最重要工作流，是一个每小时运行一次的收件箱技能：它会扫描销售代表的收件箱，找出所有该代表需要回复的邮件线程，并起草一封回复，供代表阅读、编辑和发送。该技能由精简的系统提示词、作为上下文的知识库，以及代表个人写作风格档案组成；我们每个人也会使用一个语气技能来创建自己的档案，该技能会阅读我们撰写过的文档、消息和邮件。
收件箱技能以精简的系统提示词、团队销售知识库（作为产品事实来源）以及每名销售代表定制的语气档案为基础运行。收件箱技能会扫描销售代表的收件箱，并留下待审核的回复草稿。此预览中的所有信息均已匿名化后发布。
我还依赖两项更轻量的技能来减轻行政工作负担。每位 BDR 都知道，会议爽约和潜在客户突然失联令人头疼。为此，我构建了一项会监控 Gmail 和 Google Calendar 的技能，在这些情况发生时通知我，以便我能迅速跟进。
另一项技能使用我们的 CRM 连接器扫描所有新线索，并起草个性化的首次触达内容。它会在一天中按计划运行，以确保我们不会让线索长时间等待。
我们还有一项技能可通过阅读内部关于商机阶段的指引，并将其与 Gmail 和 Gong 中实际发生的情况进行核对，来保持 Salesforce 信息更新。如果我们已经与客户会面并进入定价问题阶段，商机可能就应当推进一个阶段。Claude 会提出每一项 Salesforce 更新建议，并附上支持该建议的证据，然后等待批准。当我编辑或拒绝一项建议时，它会记录原因，从而避免再次犯同样的错误。
销售管道扫描技能会提出 Salesforce 更新建议，供销售代表批准。此处展示的是演示数据，所有信息均已匿名化后发布。
优化出站与营收工作
平均而言，我在任意时点都要负责超过一百个账户。之所以能够覆盖所有这些账户，要归功于一项在夜间作为定时任务运行的技能。它会对我负责的整个账户组合进行客户开发，观察每个账户的当前状态；例如，我们正在与谁联系、他们目前如何使用 Claude，以及有哪些相关信号。为实现这一点，Claude 会连接 Salesforce、Apollo 和 Common Room 等销售工具、Gong 以及我们的数据仓库，进行深度研究，并根据团队整理的出站指引和 ICP 标准进行验证。
这些技能组件增加了上下文，使 Claude 能更像 Anthropic 的一名 BDR 那样工作。每天早上，我打开 Claude Cowork 时，都能看到每个账户的简报、评分和出站策略。
随着时间推移，这一工作流会越来越有用，因为每位 BDR 都可以对 Claude 的结果提供反馈，而这些反馈会回流到技能中。该技能会保留一个小型记忆文件和分类账，从而避免重复或冗余的工作。
我们会在后续对话中使用这些研究成果，使外联内容更具针对性；当我们与客户交谈时，也能充分了解其业务，并贴近其问题，从而开展更深入的战略性讨论。
探索式电话会议是我们正借助 Claude 改进的另一部分出站流程。我们使用一项技能，按照探索式电话会议手册评估 Gong 转录文本，并为每通电话生成评分卡，依据具体对话提供反馈。反馈包括表现最好的三项内容、最需要改进的三项领域、依据我们标准给出的明确通过或不通过评分，以及下一步最值得练习的一项关键能力。
通话教练技能会为我们的探索式电话会议提供评分卡，并给出具体改进建议。此处展示的是演示数据，所有信息均已匿名化后发布。
简化一次性请求的处理
BDR 团队经常会收到临时提出的请求，而 Claude 使我们能够以更具战略性的方式与 AEs 协作。如果某位 AE 想了解某个重点账户的使用趋势，我们只需一个提示词，就能提供一份清晰、描述性强的仪表盘，突出显示相关趋势。
为目标账户生成的支出分析报告。
与 Claude 协作进行数据分析和报告，也同样适用于出站工作。我最喜欢的工作流之一，是运行“未发现使用情况”提示词。它会考察一名 AE 负责的全部账户，并在账户层面寻找我们尚未建立销售商机的使用信号。这往往是一个很好的信号，表明我们应开始联系客户，并共同帮助他们优化 Claude 的使用方式与体验。
对一名 AE 账户组合进行以产品为重点的扫描后，会返回所有已在使用该产品、但尚未匹配任何商机的账户。此处展示的是演示数据，所有信息均已匿名化或修改后发布。
我们还将 Claude 用于活动外联。最近，我的一位 AE 提醒我即将举办 Claude Code for Data Engineering 网络研讨会，并询问我是否能从他负责的账户中找出可能有兴趣参加的客户。我并没有为此准备专门技能，但对于这类请求，一个提示词就足够了。Claude 会检查整个账户组合中的使用数据和 CRM 历史记录，按照我们的 ICP 对每个账户评分，并标记出最匹配的账户及值得邀请的联系人。
在被要求找出适合接收网络研讨会邀请的账户后，Claude 会扫描账户组合，并根据团队的理想客户画像对每个账户进行评分。
这些技能、定时任务以及我们整理的上下文共同将 Claude 变成了一位始终在线的商务拓展伙伴。
给商务拓展团队的 Claude Cowork 入门建议
以下是商务拓展团队开始使用 Claude Cowork 的一些建议：
- 先构建知识库，再构建工作流。 将团队反复回答的问题及最佳答案收集到一份面向外部的文档中。你不必手动撰写：只需让 Claude 访问相关产品文档和团队频道，让它构建第一个版本。
- 向 Claude 提供团队工作方式的示例。 Claude 会根据你提供的上下文起草内容。对于出站业务，这些上下文可以包括成功的消息示例和理想客户画像。每位销售代表还可以让 Claude 学习自己的写作风格，让草稿的语气听起来更像发件人本人。
- 每次发送都保留人工参与。 Claude 可以生成草稿，但我们仍会阅读、编辑并亲自发送。
- 在团队内共享技能。 我们团队会将最常用的技能保存在一个共享插件中；当确认销售代表会在日常工作中持续使用某项技能后，我们就将其纳入该插件。
- 让技能具备足够的通用性，以适应整个团队的工作方式。 不同销售代表的细分市场、账户组合和工作流各不相同，因此我们会保持共享技能足够通用，使其能够适配不同情况，而不是只针对某一个人的日常流程。
- 将反馈写回技能中。 当你否决一个外联切入点或修正一份草稿时，让 Claude 将原因记录到技能中，以避免重复犯同样的错误。
我最重要的建议是什么？直接开始实验。你提供给它的上下文和工具越多，它能帮你完成的工作也就越多。
视频/嵌入内容：观看 John 在 Claude Cowork for Business Development Representatives 网络研讨会中演示这些技能。
立即开始使用 Claude Cowork。
本文中的所有 UI 模拟图均使用合成数据制作，并不代表真实公司或个人。
English / 原文
Early in my career in business development, account executives would hand me lists with hundreds of accounts and I’d have to investigate each company, find the right contacts, hunt down emails, and draft outreach. The inbound side had similarly manual and time-consuming workflows.
When I joined Anthropic last summer, I took over the responsibility of managing our sales inbox. I would spend around 5 hours per day manually responding to inbound interest from prospects, often answering the same or similar questions about our products, on top of managing my own book of business.
A lot of that work is now set up as skills and scheduled tasks in Claude Cowork. Personalized customer emails are prepared as drafts that I need to review and customize before sending. My outbound work begins with detailed research that I didn’t need to spend hours compiling.
As a result, my teammates and I spend less time on manual, repetitive work and more time on what matters: helping our customers.
Here’s how Claude Cowork is upleveling the business development function at Anthropic, allowing us to dedicate more time to strategic work, understanding customer problems, and helping educate them on how Claude can solve them.
Automating administrative and repeatable tasks
BDRs sit at the beginning of the sales process, qualifying inbound demand and building outbound pipeline for the business. At Anthropic, those motions now run through Claude Cowork first.
A foundational piece of our inbound setup is a document where I’ve collected the questions we most commonly receive in our sales inbox, along with our best answers to those questions. This document functions as our sales knowledge base, which Claude reads before drafting any replies we send. Claude helped me create that document (I simply pointed it to the relevant sources of information), and now continuously verifies that it is up to date by flagging information that might potentially be stale, which users can validate.
The heaviest workflow built on that document is an inbox skill that runs every hour: it scans a rep's inbox, finds every thread that the rep needs to answer, and drafts a reply for the rep to read, edit, and send. This skill is made of a thin system prompt, the knowledge base as context, and a profile of the rep’s writing style (which each of us also creates using a voice skill that reads through documents, messages, and emails we have written).
_The inbox skill runs on a thin system prompt, the team's sales knowledge base as the source for product facts, and a customized voice profile for each rep._The inbox skill scans a rep’s inbox and leaves drafted replies for review. All information in this preview has been anonymized for publication.
I also lean on two lighter skills that help with my administrative workload. Every BDR knows the pain of meeting no-shows and prospects going dark. To address this, I built a skill that watches Gmail and Google Calendar to notify me when that happens, so I can follow up quickly.
The other skill uses our CRM connector to scan for all new leads and draft a personalized first touch. It runs on a schedule throughout the day to ensure we don’t leave leads waiting.
We also have a skill that keeps Salesforce current by reading our internal guidance on opportunity stages and checking it against what's actually happening in Gmail and Gong. If we've met with a customer and moved on to pricing questions, the opportunity should probably progress a stage. Claude proposes each Salesforce update with the evidence behind it and waits for approval. When I edit or reject a proposal, it records the reason why so it doesn’t repeat the mistake.
The pipeline scanner skill proposes Salesforce updates for approval by the rep. Shown here with demo data and all information anonymized for publication.
Optimizing outbound and revenue work
On average, I work upwards of a hundred accounts at any given time. I’m able to cover all these accounts thanks to a skill that runs as a scheduled task overnight. It prospects across my entire book, observing the current state of each account; for example, who are we in touch with, how do they use Claude today, and what signals are relevant. To accomplish this, Claude connects to Salesforce, sales tools like Apollo and Common Room, Gong, and our data warehouse, performs deep research, and validates it against outbound guidance and ICP criteria that our team has curated.
These pieces of the skill add context to help Claude work more like a BDR at Anthropic. In the morning, I open up Claude Cowork to a brief, a score, and an outbound play for each account.
This workflow becomes increasingly useful over time as each BDR can provide feedback on Claude’s results, which then feeds back into the skill. The skill keeps a small memory file and ledger, preventing repetitive or duplicative work.
We use this research in follow-up conversations, so our outreach is tailored and when we talk to customers we're informed on their business and close enough to their problems to have a deeper strategic discussion.
Discovery calls are another part of our outbound motion we are working to improve with Claude. We use a skill that evaluates Gong transcripts against our discovery call playbook and builds a scorecard for each call, with specific feedback based on the conversation. The feedback includes top three things done well, top three areas to improve, an explicit pass or fail score on our criteria, and a single highest-leverage thing to practice next.
The call coach skill provides a scorecard for our discovery calls, along with specific recommendations on improving them. Shown here with demo data and all information anonymized for publication.
Streamlining one-off requests
Often, requests come to the BDR team in an ad-hoc manner and Claude makes it possible for us to partner with our AEs in a more strategic way. If an AE is curious about usage trends for a top account, we are a prompt away from providing a legible and descriptive dashboard that highlights the relevant trends.
A spend analysis report generated for target accounts.
Working with Claude on data analysis and reporting comes into play in outbound work, too. One of my favorite workflows is running an undiscovered usage prompt. It considers an AE’s full book and finds usage signals on the account level where we do not yet have a sales opportunity. Often, this is a great signal for us to begin reaching out and working together with a customer to optimize their usage and experience with Claude.
A product-focused sweep of one AE's book returns every account already using the product with no matching opportunity. Shown with demo data and all information anonymized or modified for publication.
We also use Claude for event outreach. One of my AEs recently flagged that we have an upcoming Claude Code for Data Engineering webinar and asked if I could find accounts in his book that would be interested in attending. I don’t have a skill for that, but for this type of request a prompt was enough. Claude checked usage data and CRM history across the book, scored each account against our ICP, and flagged the best fits with contacts worth inviting.
Asked to find the right accounts for a webinar invite, Claude sweeps the books and scores each account against the team's ideal customer profile.
Together, these skills, scheduled tasks, and the context we've curated turn Claude into an always-on business development partner.
Advice for business development teams on getting started with Claude Cowork
Below, are some tips for business development teams on getting started with Claude Cowork:
- Build the knowledge base before the workflows. Collect the questions your team answers repeatedly, and your best answers, into a single external-facing document. You don't have to write it by hand: point Claude at your relevant product docs and team channels and have it build the first version.
- Give Claude examples of how your team works. Claude drafts against the context you give it. For outbound, this can include examples of messages that worked and your ideal customer profile. Each rep can also have Claude learn their writing style, so drafts arrive sounding like the sender.
- Keep a person on every send. Claude can generate drafts, but we still read, edit, and send them.
- Share skills across the team. Our team keeps its most-used skills in a shared plugin, promoting a skill there once we establish that reps use it consistently in their daily work.
- Make skills general enough to adapt to the whole team’s way of work. Segments, books, and workflows differ across reps, so we keep shared skills general enough to adapt rather than scoped to one person's routine.
- Write feedback back into the skills. When you dismiss a hook or correct a draft, have Claude record the reason in the skill so it doesn’t make the same mistake again.
My best advice? Just start experimenting. The more context and tools you give it, the more you can get done.
Watch John demo these skills during our Claude Cowork for Business Development Representatives webinar.
Get started with Claude Cowork today.
All UI mockups in this article are depicted with synthetic data and do not represent real companies or individuals.