1. Create your first ICP

An ICP (Ideal Customer Profile) tells LeadSleuth who to look for. You don't fill in a form of dropdowns — you describe your product in plain language.

1. Go to ICP Profiles → New ICP (or press n anywhere in the app).
2. Describe what you sell and who buys it, e.g. "CI/CD observability for platform teams at 50–500 person startups". One or two sentences is enough.
3. LeadSleuth parses it into a structured profile: keywords, problem phrases, competitor names and target roles. Review and edit anything that looks off.
4. Confirm the profile. Your first scan starts within about a minute, with a widened 7-day lookback so you see results on day one. After that, scans run daily.

You can keep several active ICPs — each gets its own daily scan and a fair share of the dossier quota.

2. Tune an ICP that finds too much (or too little)

Too much noise

• Remove broad keywords ("software", "automation") and keep specific problem phrases people actually write ("flaky pipelines", "SOC 2 evidence collection").
• Add competitor names — complaints about competitors are among the highest-intent signals.
• Rate bad leads 👎. Feedback recalibrates scoring for that ICP (see guide 5).

Too few results

• Add synonyms and adjacent phrasings of the problem; scans also expand your keywords automatically with related queries.
• Check the watch tier on the Leads page — borderline leads (score 40–59) may already be there.
• Some niches genuinely post rarely. A day with zero new leads is reported honestly as "no matches", not padded with filler.

Edits apply from the next scan. To try a variation without losing the original, duplicate the ICP first.

3. Understand intent scores and the watch tier

Every matching post gets an intent score (0–100): how strongly this person signals a need you can solve, judged from the text of their public post.

• 60–100 — high intent. The lead enters your list and a dossier is built (quota permitting). Typical signals: asking for recommendations, complaining about a competitor, describing the exact problem you solve.
• 40–59 — watch tier. Plausible but not clear-cut. These are listed so you can judge them yourself; no dossier is generated and they never trigger emails.
• Below 40 — discarded.

Separately, the fit score (shown after the dossier is built) measures how well the person and company match your ICP — a high-intent post from a poor-fit company scores low on fit.

4. Read a dossier: evidence, confidence and warm paths

A dossier is a research report on one lead. Reading order that works well:

1. The signal. The original post, quoted with a link. This is why the lead exists.
2. Identity & company. Who they appear to be professionally. Built only from public data.
3. ICP match. The fit score with reasons for and against.
4. Warm paths. Concrete ways in: shared communities, recent threads they're active in. Each carries a confidence level.
5. Draft outreach. A starting point you edit and send yourself — LeadSleuth never sends anything on your behalf.

Confidence levels

High means directly supported by a cited source; medium is inferred from strong indirect evidence; low is a plausible guess — verify before relying on it. Before delivery, key claims are re-checked against their cited sources; anything that fails verification is flagged rather than silently asserted.

5. Make scoring smarter with your feedback

Two kinds of feedback flow back into scoring, per ICP:

• Explicit: the 👍/👎 on each lead ("was this signal accurate?").
• Implicit: lifecycle outcomes — marking leads contacted, replied, won, or not-a-fit.

With enough datapoints, scoring calibrates toward what you consider a good lead: sources and phrasings that earned 👎 get discounted, patterns behind your wins get boosted. Track the effect in Analytics → Quality, which shows signal precision over time. A dozen honest ratings in the first week is the single most effective thing you can do to improve your results.

6. The daily dossier quota, explained

During the beta every account includes 10 dossiers per day. Dossiers are the expensive part of the pipeline (research, verification, drafting), so the quota keeps the service free while capacity is limited.

• The quota resets at midnight UTC. The usage meter in the app shows today's count and the reset time.
• With multiple active ICPs, the 10 is split fairly across them so one noisy ICP can't starve the others.
• High-intent leads found beyond the quota are not lost: they're marked "quota full · builds tomorrow" and their dossiers are generated automatically after the reset.
• Watch-tier leads never consume quota.

7. Ask Sherla, the AI copilot

Sherla answers questions using your own ICPs, leads and dossier evidence — not generic web knowledge.

• Open it with ⌘J / Ctrl+J, the sidebar button, or the Ask tab on mobile.
• Good questions: "What's worth following up today?", "Why is this lead a good fit?", "Help me find customers building DevOps tools."
• Actions ask first. When Sherla proposes creating an ICP or opening a lead, it shows a confirmation card — nothing changes until you approve.
• History syncs. Your last 20 conversations are stored with your account and available on any device.

Sherla can be wrong. Answers cite the leads and dossiers they're based on, so you can check the underlying evidence in one click.

8. Export leads to CSV

1. Open the Leads page and apply any filters (status, source, score) you want reflected in the file.
2. Choose which columns are visible — the export includes exactly the visible columns.
3. Click Export CSV in the toolbar. The file downloads immediately; nothing is sent anywhere.

The CSV includes the signal quote and original post URL, so your CRM keeps the evidence trail. Company-level export is available on the Companies page.

9. Signal sources: what we cover — and what we don't

LeadSleuth scans public posts only — no private messages, no logged-in scraping, no purchased contact lists.

Active sources

• Hacker News — stories and comments.
• Bluesky — public posts.
• dev.to — articles and discussions.
• GitHub — public issues and discussions.
• Lobsters — stories and comments.
• Job postings (Greenhouse/Lever) — hiring signals for roles your product serves.
• Competitor pricing pages — change monitoring you can point at competitors.

Honest limitations

• X/Twitter and LinkedIn are not scanned — their terms prohibit this kind of access, and we don't do grey-area scraping.
• Coverage is strongest for developer-tool and B2B SaaS audiences, which dominate these communities.
• Sparse niches can have zero-signal days; we report them as such.

10. Delete your data

Two paths, no email ping-pong:

• In the app: Settings → Data & privacy → delete account. Account data is erased within 30 days.
• If you were researched by a LeadSleuth user and want your public-data profile removed: use the removal page — no account needed.

Retention while active: leads and dossiers are kept at most 12 months from creation, then deleted or anonymized; operational logs up to 90 days. Full details on the Security page and in the Privacy Policy.

11. Custom research questions: ask your own due-diligence questions

Every ICP can carry up to 5 custom research questions — things you always want answered about a lead, like "What tech stack do they use?" or "Any signs of budget?". The research agent answers them in every new dossier, in a dedicated Custom research section.

How to set them up

1. Open ICP profiles and edit an ICP (or add questions while creating one).
2. In the Research questions box, type a question and press Enter. Add, edit or remove up to 5.
3. Save. Questions apply to all dossiers built for that ICP from then on — existing dossiers are not rebuilt.

How answers work

• Every answer cites its sources. Answers are grounded strictly in the public evidence gathered for the dossier, and each carries clickable source links plus a confidence badge.
• Honest "not found". If the public evidence doesn't answer a question, the dossier says so — it never guesses or fabricates.
• Budget-capped. The research stage has its own slice of the per-dossier budget. If it runs out, remaining questions are marked "skipped" rather than eating into identity or outreach drafts.

Tips: specific, verifiable questions work best ("Do they mention Kubernetes?") — broad or subjective ones ("Are they nice?") usually come back as not found.

12. Monitor target accounts with the Watchlist

The Watchlist is ABM-style targeted monitoring: give LeadSleuth a list of companies and people you care about, and every scan checks public signals against it.

Adding entries

1. Open Watchlist in the sidebar.
2. Add companies by name and/or domain (e.g. acme.io), and people by name plus a public platform handle (e.g. a GitHub, Hacker News or Bluesky username).
3. Add entries one by one, or import many at once — upload or paste a CSV with columns kind,label,domain,handle,platform. The list holds up to 200 entries per account.

What happens on a hit

• When a scanned public post's author handle, author name, mentioned company domain or company name matches your list, the signal becomes a lead immediately — bypassing the usual intent-score threshold. It still gets an intent score for reference.
• The lead carries a WATCHLIST badge, and you can filter Leads (and export CSV) by watchlist hits. The API and MCP list_leads accept a watchlist parameter too.
• Hits are pushed to your webhook and Slack integrations with a watchlist field in the payload.

Privacy: only public signals are matched — LeadSleuth never scrapes private accounts or private data. Your watchlist is visible only to you.

13. Work as a team: invite members, roles and shared data

Teams share one workspace: ICPs, leads, dossiers, outreach drafts, notes and integrations are visible to every member. Free during Beta, up to 5 seats.

1. Invite someone: Settings → Team → enter their email and pick a role. They get an email with a link that's valid for 7 days; you can also copy the link and send it yourself.
2. Accepting: the invitee signs in with the invited email (magic link — new users are registered on the spot) and clicks Accept invitation. Their existing data merges into the team workspace.
3. Roles: Admins manage members, invitations, the team name and integrations. Members work with all shared data — scan, review leads, build dossiers, edit statuses — but can't manage the team.
4. Removing a member: Settings → Team → Remove. Data they contributed stays with the team; they get a fresh personal workspace on their next sign-in.
5. Quota: the daily dossier quota is pooled per team: 10 per seat, up to 30 per day.

Unread markers and email digests stay personal — marking a lead as read for you doesn't mark it for teammates.

1. 创建你的第一个 ICP

ICP（理想客户画像）告诉 LeadSleuth 该找什么样的人。你不需要填一堆下拉框——用自然语言描述你的产品即可。

1. 进入 ICP Profiles → New ICP（或在应用内任意页面按 n）。
2. 描述你卖什么、谁在买，例如：「面向 50–500 人创业公司平台团队的 CI/CD 可观测性工具」。一两句话就够。
3. LeadSleuth 会解析为结构化画像：关键词、问题短语、竞品名称与目标角色。检查并修正不准确的部分。
4. 确认画像。首次扫描约一分钟内启动，并放宽到 7 天回溯窗口，让你第一天就能看到结果。之后每天自动扫描。

可同时保持多个活跃 ICP——每个都有独立的每日扫描，并公平分摊 dossier 配额。

2. 调优 ICP：结果太多（或太少）怎么办

噪音太多

• 删除宽泛关键词（「软件」「自动化」），保留用户真实会写出的具体问题短语（「流水线不稳定」「SOC 2 证据收集」）。
• 加入竞品名称——抱怨竞品是意图最强的信号之一。
• 给差线索点 👎。反馈会重新校准该 ICP 的打分（见指南 5）。

结果太少

• 补充问题的同义与相邻表述；扫描也会自动做关键词扩展。
• 看看 Leads 页的待观察层——临界线索（40–59 分）可能已经在那里。
• 有些细分领域确实发帖少。零新线索的一天会诚实显示「无匹配」，不会用凑数内容填充。

修改从下一次扫描起生效。想试新变体又不想丢原版，可先复制 ICP。

3. 读懂意图分与待观察层

每条匹配的帖子都有一个意图分（0–100）：仅根据公开帖子的文字判断此人对你能解决的问题的需求强度。

• 60–100——高意图。进入线索列表并生成 dossier（配额允许时）。典型信号：求推荐、抱怨竞品、描述你正好能解决的问题。
• 40–59——待观察层。有可能但不确定。列出供你自行判断；不生成 dossier，也不会触发邮件。
• 40 分以下——直接丢弃。

另外还有匹配分（fit score）（dossier 生成后显示）：衡量此人及其公司与你 ICP 的匹配程度——高意图但公司不匹配的线索 fit 分会很低。

4. 读懂 dossier：证据、置信度与切入路径

Dossier 是针对单条线索的研究报告。推荐阅读顺序：

1. 信号。原帖引文+链接。这是这条线索存在的原因。
2. 身份与公司。此人的职业身份画像，仅基于公开数据。
3. ICP 匹配。fit 分及正反两方面理由。
4. 切入路径。具体的接触方式：共同社区、其近期活跃的讨论串，均标注置信度。
5. 外联草稿。供你编辑后自行发送的起点——LeadSleuth 绝不代你发送任何消息。

置信度等级

高=有引用来源直接支持；中=由较强的间接证据推断；低=合理猜测——使用前请自行核实。报告送达前，关键论断会与所引来源逐条核对，核实不了的会明确标注，绝不悄悄断言。

5. 用反馈让打分越来越准

两类反馈会按 ICP 回流进打分：

• 显式：每条线索上的 👍/👎（「这个信号准吗？」）。
• 隐式：生命周期结果——标记已联系、已回复、赢单或不匹配。

数据点足够后，打分会向你认可的好线索校准：被 👎 的来源与表述会被降权，你赢单背后的模式会被加权。在 Analytics → Quality 面板可以看到信号精度随时间的变化。第一周认真评十几条，是提升结果质量最有效的一件事。

6. 每日 dossier 配额说明

Beta 期间每个账号包含每日 10 份 dossier。Dossier 是管线中成本最高的部分（研究、核验、起草），配额让服务在容量有限时保持免费。

• 配额在 UTC 午夜重置。应用内的用量仪表显示当日已用与重置时间。
• 多个活跃 ICP 时，10 份按 ICP 公平分摊，避免一个高噪音 ICP 挤占其他。
• 超出配额的高意图线索不会丢失：会标注「今日配额已满 · 明日生成」，重置后自动补建 dossier。
• 待观察层线索不消耗配额。

7. AI 助手 Sherla 的用法

Sherla 基于你自己的 ICP、线索和 dossier 证据回答问题——不是泛泛的网络知识。

• 唤起方式：⌘J / Ctrl+J、侧栏按钮，或移动端的 Ask 标签。
• 适合问：「今天有什么值得跟进的？」「这条线索为什么匹配？」「帮我找做 DevOps 工具的客户。」
• 操作先确认。Sherla 建议创建 ICP 或打开线索时会显示确认卡片——你批准前不会有任何变更。
• 历史跨设备同步。最近 20 段对话随账号保存，任何设备都能继续。

Sherla 可能出错。回答会引用其依据的线索与 dossier，一键即可核对底层证据。

8. 导出线索为 CSV

1. 打开 Leads 页，先应用你希望体现在文件中的筛选（状态、来源、分数）。
2. 选择可见列——导出内容与可见列完全一致。
3. 点击工具栏的 Export CSV。文件立即下载，数据不会发送到任何地方。

CSV 包含信号引文与原帖链接，让你的 CRM 保留证据链。公司维度的导出在 Companies 页。

9. 信号源覆盖范围——以及我们不做什么

LeadSleuth 只扫描公开帖子——不碰私信、不做登录态抓取、不买联系人名单。

已启用的信号源

• Hacker News——帖子与评论。
• Bluesky——公开帖子。
• dev.to——文章与讨论。
• GitHub——公开 Issues 与 Discussions。
• Lobsters——帖子与评论。
• 招聘 JD（Greenhouse/Lever）——与你产品相关岗位的招聘信号。
• 竞品定价页——可指定竞品做变更监测。

诚实的局限

• X/Twitter 与 LinkedIn 不在扫描范围——其条款禁止此类访问，我们不做灰色抓取。
• 覆盖度对开发者工具与 B2B SaaS 受众最强，这些社区正是其聚集地。
• 小众领域可能出现零信号日，我们会如实呈现。

10. 删除你的数据

两条路径，无需邮件来回：

• 应用内：Settings → Data & privacy → 删除账号。账号数据 30 天内清除。
• 如果你被 LeadSleuth 用户研究过，希望移除基于公开数据的画像：使用移除页面——无需账号。

活跃期间的保留策略：线索与 dossier 自创建起最多保留 12 个月，之后删除或匿名化；运行日志最长 90 天。完整细节见安全页与隐私政策。

11. 自定义研究问题：让尽调回答你关心的问题

每个 ICP 最多可配置 5 个自定义研究问题——那些你希望每条线索都被回答的问题，比如「他们使用什么技术栈？」或「是否有预算迹象？」。调查 Agent 会在每份新 dossier 中逐题作答，展示在独立的「自定义研究」段。

如何配置

1. 打开 ICP profiles 并编辑某个 ICP（创建时也可添加）。
2. 在「研究问题」区输入问题并回车。最多 5 个，可增删改。
3. 保存后即对该 ICP 之后生成的所有 dossier 生效——已有 dossier 不会重建。

答案如何产生

• 每题答案必须带来源引用。答案严格基于尽调过程中收集的公开证据，每题附可点击的来源链接与置信度徽章。
• 诚实的「未找到」。如果公开证据无法回答某个问题，dossier 会如实说明——绝不猜测或编造。
• 预算受控。研究问题阶段有独立的子预算，超出时剩余问题标记为「已跳过」，不会挤占身份研究或触达草稿的预算。

小贴士：具体、可验证的问题效果最好（「他们是否提到 Kubernetes？」）——宽泛或主观的问题（「他们人好吗？」）通常会返回未找到。

12. 用监控名单盯紧目标客户（ABM）

监控名单是 ABM 式的定向监控：把你关心的公司和人交给 LeadSleuth，每轮扫描都会用公开信号比对名单。

添加条目

1. 在侧边栏打开 Watchlist（监控名单）。
2. 公司按名称和/或域名添加（如 acme.io）；人按姓名加公开平台账号添加（如 GitHub、Hacker News 或 Bluesky 用户名）。
3. 可逐条手动添加，也可批量导入——上传或粘贴 CSV，列为 kind,label,domain,handle,platform。每个账号上限 200 条。

命中后会发生什么

• 当扫描到的公开帖子作者账号、作者姓名、提及的公司域名或公司名命中名单时，该信号立即进入线索——绕过常规意图分门槛，同时仍会打意图分供参考。
• 线索会带上「来自名单」徽章；Leads 页可按名单命中过滤，CSV 导出含该列；API 与 MCP 的 list_leads 也支持 watchlist 参数。
• 命中会推送到你的 webhook 与 Slack 集成，payload 中包含 watchlist 字段。

隐私：只匹配公开信号——LeadSleuth 绝不抓取私有账号或私有数据。名单仅你本人可见。

13. 团队协作：邀请成员、角色与共享数据

团队共享同一个工作区：ICP、线索、dossier、触达草稿、备注与集成对所有成员可见。Beta 期间免费，最多 5 个席位。

1. 邀请成员：Settings → Team → 输入对方邮箱并选择角色。对方会收到一封含链接的邮件，链接 7 天内有效；你也可以复制链接自行发送。
2. 接受邀请：受邀者用受邀邮箱登录（魔法链接——新用户即时注册），点击「接受邀请」。其已有数据会合并进团队工作区。
3. 角色：管理员可管理成员、邀请、团队名称与集成；成员可使用全部共享数据——扫描、审阅线索、生成 dossier、修改状态——但不能管理团队。
4. 移除成员：Settings → Team → 移除。其贡献的数据保留在团队中；下次登录时会获得全新的个人工作区。
5. 配额：每日 dossier 配额按团队合并计算：每席位 10 份，每天最高 30 份。

未读标记与邮件摘要仍是个人的——你标记已读不会影响队友。