An ICP is a paragraph of plain language: who buys, why, and what disqualifies.
We run that paragraph against 100M+ classified domains — every company that matches, with the qualifying evidence quoted — instead of whatever subset a data vendor's crawler happened to index.
Every GTM motion — ABM, territory design, capacity planning, partnerships — is downstream of one artifact: the list. In most organisations, that artifact is the least rigorous thing in the building.
Someone writes a thoughtful profile: behaviors, commercial model, disqualifiers. Then someone translates it into the only available filters — industry codes, employee bands, revenue bands. The translation destroys most of the meaning.
"Manufacturers that run in-house service teams" becomes an industry code that includes thousands without service teams and excludes thousands with them. "Sites that monetize through marketplace listings" corresponds to no code at all.
The resulting list is full of accounts that were never going to buy, and missing accounts that would. The team prospects a universe shaped by database schema, not by commercial reality.
A licensed database contains what its vendor indexed — thinnest exactly where long-tail B2B lives. If your product sells to specialist workshops or regional operators, a meaningful share of your market has no profile anywhere.
The profiles that actually predict revenue are behavioral and textual. None of them survive translation into standard filters.
“Runs scheduled maintenance programs.” “Sells through distributors but supports end users directly.” “Publishes technical documentation.” These are the qualifiers your win-rate data points to — and they exist only as sentences on websites, never as columns in a schema.
Two companies share an industry code; one is a self-serve product business, the other a project shop with three customers. Your product fits exactly one of them. A code-based list treats them as identical and your conversion data pays for the difference.
In our industrial specimen runs, a fifth or more of confirmed fits had homepages without the category's obvious keywords — companies a keyword-driven index misfiles or never surfaces. For long-tail B2B, the unindexed share of your market is not a rounding error.
The same two-pass architecture we use for acquisition screening, pointed at a commercial question instead of an investment thesis.
Six of the fifteen extraction signals do most of the qualification work when the question is “will this company buy?” rather than “should we buy this company?”
Your ICP text is the instrument: the extraction matches each company's stated services, customers, and commercial model against the profile's own wording — not a proxy code. Soft criteria become screenable: “operates its own field-service organization” is checkable from service pages and technician bios; criteria genuinely invisible on the web get flagged before the run.
Who a company serves matters more than what it calls itself. End markets are extracted from case studies, named customer industries, and application pages — “serving municipal water systems since 1998” qualifies a company no industry code would surface, turned into a filterable field across the whole universe.
Online quoting, customer portals, e-commerce, published pricing, self-service scheduling — the visible surface of how a company transacts. A digitally mature account enters a product-led funnel; a phone-and-fax operation needs field sales. Maturity markers are extracted as observed facts, not assumed correlations with company size.
Named partnerships, distributor authorizations, integration listings, association memberships. Adjacency signals the warmest kind of cold account; deep certification into a rival ecosystem flags switching costs your forecast should know about. Partner pages are among the most reliably maintained on B2B sites.
Service contracts, maintenance agreements, scheduled programs, subscription language. This signal separates accounts with economics compatible with your billing model from accounts that buy once and disappear — extracted from distinctive language like “preventive maintenance programs” and “annual service agreements.”
Open roles read from careers pages at screening time. A company hiring into the function your product serves has budget moving toward your conversation. Hiring posture also serves as an honesty check: a live careers page with dated postings confirms a functioning business.
ICP universes have run across travel technology, data, software, industrial, and telecom categories for 300+ organisations — including a leading airline-metasearch platform and one of the largest European telecom operators.
One industrial category sweep held 367,478 domains globally. Across ten published subverticals, eligible independent US companies ranged from 702 in precision machining down to 93 in surface finishing.
Companies whose homepages lack the category's obvious keywords entirely — yet whose deeper pages evidence exactly the capability the profile wants. Your competitors' SDRs have never emailed them.
A list you can trust shows what it rejected and why. Request one and judge the evidence quality before anything is commissioned.
Not a replacement for everything — a different layer answering a different question. The honest comparison:
The layers compose: run discovery first, enrich with your existing stack, let intent platforms time the outreach. See our guide on databases vs full-web mapping for the coverage arithmetic.
Clear boundaries define where this method is strong and where you should spend elsewhere.
If a criterion is invisible on the public web — internal tooling, contract terms, budget cycles — we tell you during profile articulation, before money moves.
The deliverable is qualified companies with evidence, structured for your CRM. Person-level enrichment is what your existing stack does; we hand it a better universe to enrich.
Existence and fit, yes; timing, no. Nothing in a website says an account will enter a buying cycle this quarter, and we do not dress correlation up as timing signal.
In B2B this is nearly everything, but a business with no web presence is invisible to us. We state the boundary rather than extrapolating across it.
Structured files built for CRM ingestion: one row per qualified account, with profile-fit scoring, extracted fields, and evidence quotes for first-touch personalization.
Load as a dedicated CRM segment. Top-tier to named-account sellers, mid-tier to sequenced outbound, digitally mature accounts into product-led flows.
When reps disqualify accounts for a recurring reason, that reason becomes a profile revision. The included re-run applies it across the whole universe — your list sharpens each quarter.
Treat the universe as a living object: monitoring re-screens surface new entrants, pivots, and dead accounts on a schedule.
Full terms on the pricing page. Every engagement begins with a same-day specimen.
Send the profile as you would brief an analyst. We return a specimen the same day — qualified accounts with the evidence quoted, including the ones we refused to include.
Request the specimen report