Cold Email Outreach

End-to-end cold outreach: research, draft, send, follow up, route replies. Strategy is grounded in proven hook frameworks (number-led / question / pain-point / benefit-first); the execution runs on Apollo, Firecrawl, the LinkedIn scraper, and Gmail through the Hyper MCP.

Out of scope — defer to other skills

Requirements

• Hyper MCP installed and connected. https://app.hyperfx.ai/mcp
• Gmail integration connected at https://app.hyperfx.ai/integrations — supplies the sending account.
• Apollo integration connected — supplies prospect search and email enrichment.
• Firecrawl (bundled) — for company-page signals.
• Optional: LinkedIn scraper (bundled, runs through Apify) — for richer per-prospect personalization.

If gmail_send_message and apollo_mixed_people_search are not in the agent's tool list, stop and tell the user to enable the Hyper MCP and connect Gmail + Apollo.

Tool surface

Critical rules

1. Never loop apollo_people_match for multiple prospects. For 2+ records always batch into apollo_people_bulk_match. Apollo's tool description warns about this explicitly — looping single-match calls burns credits and is much slower.
2. Default send mode = drafts-first for review. For any campaign with 4+ prospects, draft the first 1–3 with gmail_create_draft, show them to the user, get explicit approval, then batch-send the rest with gmail_send_message. Never send a full campaign without showing samples first.
3. One label per campaign. Create a cold/<campaign-name> label with gmail_create_label at the start, apply it to every send, then track replies by searching that label. This is what makes Phase 6 reply routing actually work.
4. Stay under Gmail's send limits. ~500 messages/day per consumer Gmail account, ~2,000/day per Workspace user. Space sends out — see references/deliverability.md for warming and per-day pacing.
5. Personalization must connect to the problem. If the personalized opener could be deleted and the email still makes sense, it isn't doing any work. The opener should naturally bridge into why you're emailing.
6. One ask per email, one CTA. Interest-based (Worth exploring?) beats meeting requests on cold touch 1.
7. Honor unsubscribes immediately. Apply an unsubscribed label on any "remove me / not interested" reply and never re-target that address from the same Hyper workspace.

Workflow

Phase 1 — Define the campaign (always do this first)

Get the user to commit to:

1. ICP — Role(s), industry, company size, tech stack, geography. Concrete: "Heads of Growth at US-based pre-seed-to-Series-A B2B SaaS, 10–50 employees, using HubSpot."
2. The ask — What does a "yes" look like? (15-min call, async reply, demo, intro to someone else.)
3. Value prop in one sentence — "We help X do Y so they can Z."
4. Proof point — One specific result: "We helped Notion cut their CAC by 31% in 90 days." (Made up examples are worse than no example — get a real one.)
5. Trigger / signal (optional but powerful) — Funding round, hiring, pricing-page change, recent blog post, product launch, leadership change.
6. Sender + reply-to — Which Gmail account is sending. (Confirm with gmail_list_labels to verify the integration is live.)
7. Volume + cadence — Total prospects, max sends/day, follow-up gap pattern.

If they're stuck on any of these, push back. A campaign without proof or a clear ask will not perform regardless of how clever the writing is.

Phase 2 — Build & enrich the prospect list

# Search by ICP
apollo_mixed_people_search(
  person_titles=["Head of Growth", "VP Growth", "Director of Growth"],
  organization_num_employees_ranges=["11,50"],
  person_locations=["United States"],
  per_page=50,
)

Then for the prospects you actually want to contact, batch-enrich for emails:

# CORRECT — one bulk call for many prospects
apollo_people_bulk_match(
  details=[
    {"first_name": "...", "last_name": "...", "domain": "..."},
    ...up to 10 per call...
  ],
  reveal_personal_emails=True,
)

Only fall back to apollo_people_match for single-prospect lookups (e.g., the user pastes one LinkedIn URL).

For deeper company-level context (industry, revenue range, tech stack), call apollo_mixed_companies_search by organization name on the companies you want to enrich. Note that person search results already include core company fields (headcount, industry, location) — only reach for apollo_mixed_companies_search when you need data beyond what the person search returns.

Phase 3 — Per-prospect signals (the personalization layer)

For each prospect, gather one specific observation that connects to the problem you solve. Use the cheapest signal that works:

Personalization tiers (use the highest tier you can afford for this campaign):

• Tier 1 (mass / low-effort) — first name + role + company + industry. Acceptable only when the value prop is sharp enough to carry the email on its own. Reply rates: low.
• Tier 2 (signal-based) — the prospect is in a role/stage where the problem you solve is acute (e.g., a new Head of Growth in their first 60 days). Reply rates: meaningfully better.
• Tier 3 (observation-based) — references something from the company site, pricing page, careers page, or a recent product launch. This is the sweet spot.
• Tier 4 (deep) — references a recent LinkedIn post, blog post they wrote, or talk they gave. Reserve for high-value targets.

Anything below Tier 2 should be treated with suspicion — {{FirstName}} swaps don't count as personalization.

Phase 4 — Draft emails (drafts-first by default)

Pick a framework that matches the situation. The four shapes that consistently work:

• Observation → Problem → Proof → Ask — "You're hiring SDRs. That usually means meetings/SDR ratio is the bottleneck. We help