Warm outbound from zero

How each step works

1 · Ingest — Parse + dedup

Vector fires a webhook for every de-anonymised visitor. The workflow normalises the payload (33 fields: identity, company, page behaviour) and runs a 5-key dedup check against seen_visitors before spending a single credit.

2 · Enrich — LinkedIn lookup

Visitors who arrive without a company are rescued: Serper finds their LinkedIn profile, Apify scrapes it, and name-matching validates the result before backfilling company and title.

3 · Signal — Page → bucket

Page URLs are mapped to intent buckets (pricing, docs, dashboard, compare, homepage…). Each bucket carries a pre-written opener line and subject line used later in drafting.

4 · Qualify — ICP filter + AI

A deterministic pre-filter blocks non-ICP titles and industries cheaply. Survivors get an AI fit-check (Claude Haiku) that scores B2B fit 0–10 and returns a company summary.

5 · Email — 7-provider waterfall

Dropleads → Hunter → LeadMagic → Prospeo → Deepline-native → Crustdata → PDL. First provider to return a valid address wins; the rest are skipped.

6 · Draft — Claude icebreaker

Claude Sonnet writes a 50–65 word observation email personalised to the visitor's role bucket. No product pitch, no demo ask — just a segmented pain and an open question.

7 · Review — Slack HITL

The full lead context (identity, intent signals, drafted email) posts to #alerts-gtm with Approve / Skip / Snooze buttons and an edit affordance. The workflow waits up to 24 h for a human decision.

8 · Send — Push to Lemlist

On approval the lead is added to the Lemlist campaign with the (possibly edited) icebreaker attached as a custom variable. A Lemlist pre-check prevents double-adding existing contacts.