Outbound quality and outbound volume are usually a trade. We wanted both.
Manual prospecting is 15-25 minutes per company: open the site, work out what they do, decide if they fit, find the angle, write the email. Across a 21,000-company base that is more than three person-years of work, so in practice it never happens.
The standard shortcut is merge-field blasts — "Hi {{first_name}}, love what {{company}} is doing". Those burn sender domains and earn exactly the reply rate they deserve.
A staged pipeline where the expensive model only ever sees the good leads.
Research: every domain gets scraped, free path first. Anti-bot and JS-heavy sites fall back to a pay-per-success fetcher, so dead sites cost nothing. A liveness gate skips unreachable companies before any model spend.
Qualify: a cheap model scores each company against our ICP with hard disqualifiers checked in order — wrong category, holding-company ownership, too big, too small — producing a 0-100 score, an A/B/C tier, a campaign route, and the evidence behind each call.
Personalise: for the tiers we choose to spend on (a checkbox, not a code change), a mid-tier model writes an operational observation of how their current process leaks, and the premium model drafts the sequence that quotes it. Proof points come only from a tagged knowledge base of real cases, so the copy cannot invent a credential.
Qualifying a company costs about a cent, so we can afford to look at all twenty-one thousand. The expensive model only ever runs on the few hundred that earn it.
Drafting is half the job. The other half is 50 mailboxes that never get burned.
Approved sequences push into Instantly across 11 warmed domains and about 50 mailboxes, with rotation and daily caps set per campaign. Warm-up runs continuously, scores sit between 78 and 100 across the estate, and all 11 domains are verified in Google Postmaster with reputation checked daily.
Volume ramped on purpose rather than all at once: 585 sends in the first week, 7,525 in the heaviest. Across 19,840 sends we bounced 162 addresses — 0.8% — because every address is verified before upload and the base is deduplicated so no company is approached twice.
The send ledger is the unglamorous part that makes it safe to run more than one ICP at a time: every touch is recorded against a company, so a lead already in one sequence cannot be picked up by another campaign.
Deliverability is not a copywriting problem. It is an infrastructure problem, and infrastructure is the part you can actually engineer.
Where the first versions failed.
Dead and parked domains poisoned early runs — the qualifier would hallucinate a company from a placeholder page. We added a pre-flight liveness gate: unreachable site plus no enrichment metadata means the company is skipped before any LLM spend.
Anti-bot walls blanked out a chunk of the base. A rendering fallback that bills only on success fixed coverage without unpredictable cost — a failed fetch costs nothing.
The sequence model kept drifting from the configured cadence, inventing an extra step or merging two emails. We stopped trusting the model with structure: steps map onto the configured plan by key, so the cadence is enforced by code and the model only writes the words.
Prospecting economics that make personalised outbound rational.
- Research + qualify, per company15-25 min → under 1 minautomated
- Cost to qualify a companySDR time → ≈1¢ in tokens~99% cheaper
- Personalised sequence2-3 h manual → ≈35¢ draftreview-only
- Weekly sending volume0 → 7,525ramped over 6 weeks
- Bounce rate0.8%162 of 19,840 sends
- Domain estate11 domains · ~50 mailboxesall Postmaster-verified
- Sends without human approval0by design
Every number above is visible in the tool itself.
Cost per company: token usage is logged per run and per stage, and the control panel shows ≈$ per run next to each entry. Qualification averages ~5k cheap-model tokens (≈1¢); a full qualified lead — research, qualify, audit, full sequence on the premium model — lands around 30-40¢.
Deliverability: bounce, warm-up score and Postmaster reputation are pulled per domain, so 0.8% is a measured number rather than an estimate.
We do not publish reply or conversion figures here. The engine is proving out its infrastructure ahead of its targeting, and we would rather show you the panel on a call than quote a number we cannot stand behind.
Nothing reaches a human inbox without a human decision.
Every sequence is a draft until an operator approves it. Push-to-sending-tool is a manual button, per batch, after review.
Every stage that costs money — rendering fallback, contact discovery, review scraping, news intel — ships toggled off and is enabled per run from the panel.
Disqualification reasons are kept, not just the score, so a human can audit why the engine said no and overrule it.