psquared-development/find-leads

Find New Leads for InboxMate

Parameters

/find-leads [N] [track] — track is optional:

• inbox — EMAIL-ONLY mode: search email-ops-heavy verticals (see "Email-volume signals" below — rotate 2-3 verticals per wave, never fixate on one industry). Every qualified lead gets Track: INBOX in its note. A chatbot on their site is a positive signal here, never a skip. Demos for these run via /inboxmate-inbox-demo.
• chatbot — chatbot-only mode: classic behavior, skip companies that already have a chatbot, every lead gets Track: CHATBOT.
• omitted — mixed: decide the track per lead (see Step 2b).

Announce:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Lead Finder started.
Target: [N] new leads · Track: [INBOX / CHATBOT / mixed]
Checking environment...
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

---

Autonomy — Run Without Asking

This skill should run as autonomously as possible. The user may not be available to answer questions. Do NOT ask for confirmation between leads — just process them one by one:

• Search → validate → skip or add → next. No pauses.
• If a check is ambiguous (e.g., unclear if email is personal), err on the side of skipping and add to skip-list. Don't ask.
• Only stop if a critical error occurs (e.g., CRM token invalid, network down).
• Report results at the end, not during.

---

Legal Context — READ THIS FIRST

B2B cold email in Germany and Austria is a legal grey area. The strict reading of UWG §7 (Germany) and TKG §174 (Austria) says prior consent is required. However, the prevailing practical interpretation (used by thousands of companies daily) allows individual B2B outreach IF certain conditions are met.

Our position: We send individually crafted, genuinely valuable emails (with a working demo built for their company). This is not spam. But we MUST document why each lead qualifies for outreach.

Every lead MUST pass ALL 8 of these checks:

The 7 Qualification Criteria

| # | Criterion | What to check | Why it matters | |---|-----------|---------------|----------------| | 1 | German company | Domain is .de, Impressum shows German address (DE), company is registered in Germany | Austrian law (TKG §174) is stricter — no B2B cold email exception. Germany (UWG) has more practical tolerance for relevant individual B2B outreach. | | 2 | B2B only | Company, not a private person | Consumer protection is stricter | | 3 | Objectively relevant | Company would genuinely benefit from a chatbot OR from email automation (customer-facing business with visible inquiry/booking volume) | UWG requires the offer to be relevant to the recipient's business | | 4 | Publicly available email | Email found on their public website (contact page, imprint, team page) — NOT scraped from LinkedIn, purchased lists, or leaked databases | Data source must be GDPR-compliant (publicly made available by the company itself) | | 5 | Specific contact person | Email goes to a named person with decision authority (founder, CEO, marketing lead, head of sales) — NOT info@, office@, or generic addresses | Shows individual outreach, not bulk | | 6 | Genuine value offer | We built something specifically for them (a demo) — not a generic pitch | Differentiates from spam | | 7 | Not already in CRM | Company doesn't already exist in our CRM | No duplicate outreach | | 8 | Website is active | Website loads, has real content, is maintained (copyright not 2+ years old) | No dead businesses |

If ANY of the 8 criteria fails → SKIP the lead. Add to skip-list.json. Do not add to CRM.

---

Skip List — Persistent Rejected Leads

The file skip-list.json in the current working directory tracks companies that were already checked and rejected. Before researching any company, check this list first. If the domain is in the list, skip immediately — don't waste time fetching the website again.

On first run: If skip-list.json doesn't exist, create it with [].

Format:

[
  { "domain": "example.de", "reason": "has chatbot", "date": "2026-03-19" },
  { "domain": "other.at", "reason": "no personal email", "date": "2026-03-19" }
]

When a company fails any check: Append to skip-list.json with the domain and a short reason (max 5 words).

Before checking any website: Read skip-list.json, check if domain is already there. If yes, skip without fetching.

Short reason examples:

• "has chatbot, no email volume" — chat widget present AND no email-automation angle (a chatbot alone no longer disqualifies — it routes to the INBOX track)
• "no personal email" — only info@/office@ found
• "dead website" — site down or parked
• "outdated site" — copyright 2+ years old
• "too small" — sole proprietor, no real web presence
• "not B2B-facing" — wholesale/internal only
• "already in CRM" — duplicate
• "no email on site" — no contact info found
• "generic only" — only kontakt@/info@ available

---

STEP 0 — Check Environment

Read .env in the current working directory using the Read tool. Extract these tokens:

| Variable | Purpose | Used for | |----------|---------|----------| | PSQUARED_CRM_TOKEN | Twenty CRM GraphQL API | Checking duplicates, creating companies/persons/notes |

The .env file is at the root of the working directory (claude-overlord-folder). If it doesn't exist, stop and tell the user.

Read skip-list.json (or create it as [] if it doesn't exist).

Once verified: Environment OK. Skip list: [N] entries. Starting lead search...

---

STEP 1 — Search for Leads

Use WebSearch to find German and Austrian businesses that would benefit from an AI chatbot on their website.

Search strategies (rotate between these — Germany only, .de domains):

• "[industry] Unternehmen" site:.de
• "[industry] GmbH" Kontakt Email site:.de
• "Geschäftsführer" "[industry]" site:.de
• "Immobilienmakler" OR "Versicherungsmakler" OR "Steuerberater" OR "Rechtsanwalt" site:.de
• Industry-specific: real estate agencies, insurance brokers, IT consultancies, marketing agencies, SaaS companies, e-commerce shops, dental practices, law firms, accounting firms

Good target industries for InboxMate:

• Real estate (Immobilien) — high inquiry volume, FAQ-heavy
• Insurance/finance — customer questions, lead qualification
• IT services / MSPs — support requests, product questions
• Marketing/digital agencies — client communication, onboarding
• SaaS/software companies — customer support, feature questions
• Professional services (law, tax, consulting) — initial consultations
• E-commerce — product questions, order support
• Healthcare (dental, physio) — appointment scheduling, FAQ

Email-volume signals (qualify for the INBOX track — the paid email product): InboxMate's email product is NOT just answering inquiries — it sorts, routes, answers and triggers actions on operational mail. Target ANY German B2B whose operations run through the inbox. Pick 2-3 verticals per wave and ROTATE — do not fixate on one industry (e.g. not only Thermen):

• Order/booking-heavy: hotels, venues, tour operators, leisure (Thermen/parks), restaurants with group business, Werkstätten/Autohäuser (Terminanfragen)
• Ops-mail-heavy: Großhandel/Distributoren (Bestellungen per Mail), Hausverwaltungen (Mieteranfragen, Handwerker-Koordination), Logistik/Spedition (Transportanfragen), Handwerksbetriebe (Angebotsanfragen), Personaldienstleister (Bewerbungseingang)
• Document-heavy: businesses with visible Rechnungseingang/Lieferanten flows — InboxMate erkennt Rechnungen und routet sie an die Buchhaltung
• Universal signals: visible info@/office@ as THE contact channel, forms that just send email, no ticket system, Gutscheine/Stornos/Angebotsanfragen on the site
• The Demo-Postfach (/inboxmate-inbox-demo) proves the fix — seeded threads should mirror THEIR ops mail, incl. a Rechnung or Bewerbung being auto-routed where plausible

Bad targets (skip):

• Very small sole proprietors with no website
• Government / public institutions
• Non-profits (unless large)
• Companies with neither chatbot relevance nor email volume (e.g., heavy industry, agriculture)

---

STEP 2 — Validate Each Lead

For each potential lead found:

2a — Check skip list first

Extract the domain from the company URL. Check if it's in skip-list.json. If found, skip immediately and move to the next lead — no need to fetch the website.

2b — Check the website (WebFetch)

Visit their website. Verify:

• Website loads and is active
• Impressum shows a German address (city in Germany, not Austria/Switzerland) — this confirms criterion #1
• Content is current (check copyright year, last blog post, news)
• Company does customer-facing business (not just B2B wholesale with no public presence)
• Check whether they already have a chatbot on their site (chat widgets like Intercom, Drift, Zendesk, HubSpot, Tidio, Crisp, LiveChat in the HTML source). Having a chatbot is NOT a skip anymore — it decides the track:
  • No chatbot + customer-facing site → CHATBOT track (/inboxmate-demo)
  • Has a chatbot (or weak chatbot fit) + clear email-volume signals → INBOX track (/inboxmate-inbox-demo) — an existing chatbot even confirms they invest in automation
  • Has a chatbot + NO email-volume signals → skip with reason "has chatbot, no email volume"

Skip if: website is down, parked, outdated, or NOT a German company. Add to skip-list.json with reason.

2c — Find a contact person with email

Look for contact information on:

1. /kontakt or /contact page
2. /impressum or /imprint page
3. /team or /ueber-uns page
4. Footer of the homepage

What we need:

• The best contact person who will understand our offer — someone technical or product-oriented
• Ideal roles (in order of preference): CTO, Head of IT, Head of Product, Head of Digital, Technical Lead, Marketing-Leiter with tech focus
• Fallback roles: Geschäftsführer, CEO, Head of Sales — but only if no technical contact is available
• Their email address — must be visible on the company's PUBLIC website
• Their role/title

Email rules:

• MUST be a personal email ([email protected] or [email protected])
• MUST be found on the company's own website (screenshot/note the exact page)
• Must NOT be from a purchased list, LinkedIn scrape, or any third-party source
• AVOID generic addresses (info@, office@, hello@, kontakt@) — these rarely reach the right person
• Generic emails are acceptable ONLY as last resort if no personal email is available AND you also found a specific contact name on the site
• If no usable email found → add to skip-list.json with reason and move on

2d — Determine relevance

Write a 1-2 sentence justification for why InboxMate would be valuable for this specific company. Reference something concrete:

• "High-traffic real estate listing site with FAQ section that could be automated"
• "IT consultancy with complex service portfolio — chatbot could pre-qualify leads"
• "Marketing agency with multiple service pages — visitors need guidance to the right offering"

2e — Check CRM for duplicates

Check by domain first (most reliable), then fall back to company name:

Step 1 — Search by domain:

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"{ companies(filter: { domainName: { primaryLinkUrl: { like: \\\"%[domain]%\\\" } } }, first: 1) { totalCount } }\"}"

Step 2 — If not found by domain, also search by name (catches companies added without a domain):

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"{ companies(filter: { name: { like: \\\"%[Company Name]%\\\" } }, first: 1) { totalCount } }\"}"

If either query returns totalCount > 0 → SKIP. Add to skip-list.json with reason "already in CRM".

---

STEP 3 — Add Qualified Lead to CRM

For each lead that passes ALL 7 criteria:

3a — Create Company

Always set outreachFor: [INBOX_MATE] so downstream InboxMate skills pick this company up (demos, campaigns, email drafts filter on this field). Never set PERSONAL_ONLY or INBOX_MATE_MUSEUM here — PERSONAL_ONLY is reserved for companies Martin adds manually and must never be touched by automation; INBOX_MATE_MUSEUM is reserved for the curated museum list (imported 2026-05-08) and is handled outside this skill.

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"mutation { createCompany(data: { name: \\\"[Company Name]\\\", domainName: { primaryLinkUrl: \\\"https://[domain]\\\" }, outreachFor: [INBOX_MATE] }) { id name outreachFor } }\"}"

3b — Create Person (contact)

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"mutation { createPerson(data: { name: { firstName: \\\"[First]\\\", lastName: \\\"[Last]\\\" }, emails: { primaryEmail: \\\"[email]\\\" }, companyId: \\\"[companyId]\\\", jobTitle: \\\"[Role/Title]\\\" }) { id } }\"}"

3c — Create Note with justification

Two-step process: Create the note first (with just a title), then update it with the full body. This avoids issues with long content in a single mutation.

Step 1 — Create note:

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"mutation { createNote(data: { title: \\\"Lead Qualification: [Company Name] — [date]\\\" }) { id } }\"}"

Step 2 — Update note with full content:

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"mutation { updateNote(id: \\\"[noteId]\\\", data: { body: \\\"Email source: [URL where email was found]\\\\nContact: [Name], [Role]\\\\nRelevance: [1-2 sentence justification]\\\\n\\\\nTrack: [CHATBOT | INBOX]\\\\nDemo approach: [suggest best angle — CHATBOT e.g. 'FAQ automation for property listings'; INBOX e.g. 'Demo-Postfach: Gutschein-Anfragen, Gruppenbuchungen, Storno-Mails vorsortiert + beantwortet']\\\\n\\\\nChecks: German company ✓, B2B ✓, Relevant ✓, Public email ✓, Named contact ✓, Value offer ✓, Not in CRM ✓, Active website ✓\\\" }) { id } }\"}"

Step 3 — Link note to company:

curl -s -X POST https://crm.psquared.dev/graphql \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PSQUARED_CRM_TOKEN" \
  -d "{\"query\":\"mutation { createNoteTarget(data: { noteId: \\\"[noteId]\\\", companyId: \\\"[companyId]\\\" }) { id } }\"}"

The "Demo approach" field should suggest the best angle for the InboxMate demo based on what you found on their website. Examples:

• Real estate: "FAQ bot for property listings — automate viewing requests and price inquiries"
• IT services: "Lead qualification chatbot — help visitors find the right service package"
• Agency: "Portfolio navigator — guide prospects to relevant case studies and services"
• E-commerce: "Product advisor — help customers choose the right product, reduce support load"

Announce after each: Added: [Company Name] — [contact email] — [1-line reason]

---

STEP 4 — Report

Announce:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Lead Finder complete.

Added: [N]
  - [Company A] — [email] — [reason]
  - [Company B] — [email] — [reason]

Skipped: [M]
  - [Company C] — already in CRM
  - [Company D] — no personal email found
  - [Company E] — already has chatbot

Next step: Run /create-demo-for-crm-lead to build demos
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

---

Important Reminders

• Quality over quantity. 5 well-qualified leads are worth more than 50 questionable ones.
• Document everything. The note on each company is your legal paper trail.
• Only public emails. If you can't find an email on their website, skip the lead.
• Rotate industries. Don't search for 10 real estate companies in a row — diversify.
• Check for chatbots. If they already have Intercom/Drift/Zendesk/HubSpot chat, skip — they're already served.
• Germany only. Only .de domains. Austrian law (TKG) is stricter — no B2B exception. Verify Impressum shows German address.