B2B Lead Pipeline — Companies, Decision-Makers & Emails

Building a lead list is four jobs, and most tools do one. Find the companies that match your ICP. Find the right person at each one. Get their work e-mail. Work out which of them is worth writing to first. Doing that by stitching four tools together is a morning's work — and you do it again next week.

This is the four jobs in one call. Describe who you sell to, say which titles matter, and get back a finished list: one row per person, with a work e-mail, deduplicated and scored A–D.

---

What one row looks like

Company	company, domain, website, industry, companySize, headquarters, country
Person	person, firstName, lastName, title, seniority, department, personLinkedinUrl, matchedTitle
Contact	email, emailStatus, emailConfidence, emailVerification, emailPattern, catchAll
Context (optional)	techStack, signals, signalsPresent, signalScore, whyNow, sellTo
Verdict	leadScore, leadGrade, scoreReasons — in English, so you can argue with it

---

Who it is for

• Founders and small sales teams — a real list, from a sentence describing your customer, in one run.
• AI / GTM agents — one tool call, a flat dataset, no login and no key. Hand it an ICP, get leads.
• Agencies — a repeatable pipeline per client instead of a per-client stack of subscriptions.
• RevOps — top up a segment without opening four tabs.

---

How to use it

Describe the customer, name the titles:

{
  "keywords": ["revenue intelligence software"],
  "locations": ["United States"],
  "companySizes": ["51-200", "201-500"],
  "titles": ["VP of Sales", "Head of Revenue Operations"],
  "maxCompanies": 25,
  "maxPeoplePerCompany": 2
}

Already have the accounts? Skip the search:

{
  "companies": ["gong.io", "ramp.com", "lemlist.com"],
  "titles": ["Chief Revenue Officer", "VP of Marketing"],
  "requireEmail": true
}

The full-context version — every lead carries the company's tech stack and why now is the moment:

{
  "keywords": ["logistics software"],
  "titles": ["Head of Operations"],
  "enrichCompanies": true,
  "attachSignals": true,
  "minLeadScore": 55
}

---

The lead score, in plain terms

Four things decide whether a row is worth a sequence, and each is scored on its own terms before being weighted: can you reach them (35) — a confirmed mailbox beats a learned pattern beats a guess; are they the right person (30) — an explicit title match beats a seniority guess; is it the right company (15) — resolved domain, known industry and size; is now the right time (20) — the buying signals, when you attach them.

The total is normalised over the parts the run could actually produce, so a run without signals still reaches an A and rows stay comparable inside a run. Grades: A ≥75, B ≥55, C ≥35, D below. No model call, no API key, and scoreReasons spells out every judgement in English.

---

Honest about e-mails

emailStatus says exactly how much an address has been checked: verified (the mailbox itself confirmed), found (a learned or observed pattern on a live mail domain), guessed (the common pattern for that domain), none. A confident pattern on a domain that cannot receive mail is never better than a guess. emailVerification carries the raw depth, catchAll flags the domains that accept everything, and requireEmail drops anyone without an address.

---

Built to keep running

• Every stage degrades on its own. If the people search comes back empty for a company, that company still gets its reachable and role addresses so you have a way in. If an optional layer is unavailable, the leads still ship and stepsSkipped says what was missing.
• Deduplicated per person, keyed on person + company, keeping the record with the better contact data — so the same human never appears twice.
• Capped on purpose. maxCompanies, maxPeoplePerCompany and maxLeads bound both the output and the cost, and maxSpendUsd is a hard ceiling on the stages.
• Finishes on time. maxRunSecs is a wall-clock budget for the whole run; when it is nearly spent the remaining stages are shortened or skipped and the leads found so far are delivered, with the cut named under stepsSkipped. The run's own timeout is honoured the same way. The form starts small — 3 companies, 2 people each, about three minutes — so the first run is a sample, not a bill; raise the caps once it looks right.
• Every run reports itself. The summary carries what each stage produced, e-mail coverage, the grade split and what the run cost per lead.

---

Notes

• Dataset-only. No UI, no dashboard, no login. Four ready-made views: leads, contact quality, company context, score audit.
• Public data only. Nothing behind a login is ever touched.
• Two or three well-chosen contacts per company beat ten. The default is three.