Most AI-generated lead lists fall apart the moment someone sends the first email. The company is the wrong size, the "Head of IT" left last year, and half the addresses were guessed from a naming pattern. The list looks finished. It isn't.
Lead Intelligence turns your agent into a careful prospecting analyst instead of a row generator. You give it your website and a market; it gives you a ranked list of accounts and decision-makers where every important field has a source, a date and a confidence level.
Why agent lead lists usually fail
Left to its defaults, an agent optimises for volume. It:
- treats a single directory listing as fact
- guesses emails from name patterns and presents them as real
- ranks a company highly because it matches your industry, even with no sign of need
- merges branches, subsidiaries and similarly named companies
None of that is visible until outreach bounces or reaches the wrong person.
How the skill works
The skill walks your agent through a defined research workflow:
- Understand the seller. It reads your website first — offer, integrations, regions, proof — and works out who actually buys from you.
- Write down the ICP. Must-have filters, positive and negative signals, and buyer personas, before a single search runs.
- Discover and enrich. Several independent sources per account: company sites, registries, job posts, news and technology signals.
- Separate fit from need, intent and timing. A hiring post, a tender and a funding round are different kinds of evidence, and they are scored differently.
- Find and verify contacts. Every email carries a verification status such as
verified_valid,accept_all,not_checkedorunverified_pattern, plus its mailbox type and where it came from. Nothing guessed is ever marked verified. - Exclude, then score and explain. Competitors, suppressed accounts and markets you don't serve are excluded outright. Every other lead gets 0–100 across seven dimensions, with penalties for stale evidence or weak buyer relevance.
What the output looks like
A spreadsheet-ready table sorted by score, with a plain-language reason for every lead. An illustrative row, based on the skill's own worked example:
| Account | Score | Tier | Why |
|---|---|---|---|
| Logistics company, Netherlands, ~220 staff | 88 | A | Runs NetSuite, operates in four countries, hiring a NetSuite administrator; Head of IT email verified |
Rows without usable contacts are kept apart, low-confidence fields are labelled, and every row keeps its source URLs.
Works in any market
It searches local-language and transliterated company names, recognises legal-suffix variants from GmbH to Pty Ltd, and normalises phone numbers to international format, so the same company is never counted twice.
Who it's for
Founders, sales teams and agencies who prospect with Claude or Codex and want a list they can actually send to — especially in B2B services, SaaS and ERP.
What it won't do
It won't invent emails, phone numbers, revenue, headcount or buying signals. It won't bypass logins or anti-bot protection, and it won't promise that a lead converts. When something can't be verified, it says so.
