What Is Firmographic Data?

Firmographic data is the set of attributes that describe an organisation rather than an individual: industry, headcount, revenue, location, age, ownership and funding. The word is a deliberate parallel to demographics. Demographics segment a consumer market by age, income and household; firmographics segment a business market by size, sector and geography.

The practical definition is narrower than the dictionary one. A firmographic field is an attribute a company has whether or not anybody is selling to it, and whether or not it is doing anything in particular this month. Headcount is firmographic. Industry is firmographic. "Downloaded a pricing guide last Tuesday" is not, that is behavioural. "Runs Snowflake" is not either, that is technographic. Keeping those three families separate is what stops a scoring model from double-counting the same signal.

Firmographics answer four questions that gate every B2B qualification decision before a rep ever picks up a phone. Is this company in a sector we sell to? Is it large enough to need what we sell and small enough to buy it without a twelve-month procurement cycle? Is it in a geography we can service and lawfully contact? And is it structured in a way that means the buyer we want actually sits here, rather than at a parent entity two countries away? A list that cannot answer those four is a directory.

The reason the category is worth a page of its own is that firmographics are the cheapest data in B2B and the most load-bearing. Contact data decides whether you reach a person. Firmographic data decides whether reaching them was worth doing. Teams routinely spend an order of magnitude more on the former while leaving the latter half-populated, then wonder why their ICP score does not predict anything.

The standard set has six families: identity, industry, size, location, age and structure, and funding. Almost every provider ships some version of these, and the differences between vendors show up in coverage and freshness rather than in which fields exist.

Identity. Legal name, trading name, primary web domain, LinkedIn company URL, logo, and any aliases or former names. The domain is the field that matters most operationally, because it is the only one of the six that reliably works as a join key. Company names are ambiguous (there are thousands of businesses called Apex), and matching on them is the single largest source of wrong-company enrichment.

Industry classification. SIC and NAICS codes, plus the provider's own taxonomy and, increasingly, the LinkedIn industry label. These three rarely agree. NAICS is a US-centric government scheme designed for economic statistics, so it slices manufacturing finely and treats most of modern software as one code. LinkedIn's taxonomy is self-declared by the company. A provider's proprietary taxonomy is usually the most useful for GTM and the least portable between vendors.

Size. Employee headcount, headcount band, headcount growth over a trailing window, annual revenue, revenue band. Headcount is usually the most reliable size signal in B2B because it is observable from the outside. Revenue mostly is not, which is covered further down.

Location. Headquarters country, state or region, city, postal address, and the list of additional office locations. HQ country is the field that drives territory assignment and, in practice, compliance routing, because whether GDPR or CCPA governs an outreach depends on where the person is rather than where the company is incorporated.

Age and structure. Founding year, ownership type (public, private, private-equity-backed, non-profit, government), parent company, subsidiaries, number of locations, and whether the entity has been acquired. Structure is the family most teams skip and the one that most often explains a lost deal, because selling to a subsidiary of an account you already serve is a very different motion from a cold start.

Funding. Total raised, last round type and date, investors, and whether the company is bootstrapped. This family sits at the edge of the definition: it is stable and company-level like a firmographic, but it also functions as a timing signal, which is why it appears in both firmographic and signal-based playbooks.

On Cleanlist AI, an enriched company record comes back with up to 180 firmographic properties spanning those families, and the six that are exposed as structured Company Search filters are industry, employee count range, funding stage, location, company name and company domain.

Each family has a different origin, and the origin is what predicts its reliability. There is no single register of company facts, so every provider assembles firmographics from a mixture of public filings, the company's own web presence, employment data and inference.

Identity and domain come from registries and from the web. Company registrations (SEC filings in the US, Companies House in the UK, provincial and state registries elsewhere) give legal names and incorporation details. Domains come from crawling and from DNS and WHOIS records. This family is the most accurate available because it is the most documented.

Industry classification comes from three separate places and inherits the flaws of each. Government codes come from what the company told a registrar, often years ago and often chosen for tax reasons. The LinkedIn label comes from what the company chose from a dropdown. Provider taxonomies come from a model reading the company's website, which is the most current source and the most likely to misread a holding page or a rebrand.

Headcount comes overwhelmingly from professional network profile counts: a provider counts the people who list the company as their current employer. That method is fast, current and systematically biased. It undercounts companies in sectors where staff do not maintain professional profiles (hospitality, field trades, manufacturing floor roles) and overcounts companies where former employees have not updated a profile. Filings-based headcount is more accurate and available for a small minority of companies.

Revenue is the weakest field in the set. Public companies report it. Everyone else is estimated, usually by modelling from headcount and sector, which means the revenue field for most private companies is a transformation of the headcount field rather than an independent observation. Treat it as a band, never as a number.

Location comes from registrations, from the company website's contact page, and from job posting locations. Office lists in particular are frequently reconstructed from where a company advertises roles, which is why a firm that hires remotely can appear to have offices it does not have.

Funding comes from press releases, regulatory filings such as Form D, and the crowd-maintained funding databases. It is well covered for venture-backed technology companies in North America and Western Europe and thin nearly everywhere else, which is a coverage bias worth knowing before a funding filter becomes the spine of a segment.

Cleanlist resolves these live rather than holding an index. The company data layer behind Company Search is supplied by Crustdata, whose own site describes a real-time API carrying hiring, funding and company event signals, and a flat file it describes as "a fully refreshed dataset monthly, unified from 11 sources" (crustdata.com, read 2026-09-01).

Firmographic data decays far more slowly than contact data, and within firmographics the decay rate varies by an order of magnitude between fields. 6sense puts company data decay at roughly 15% a year. For comparison, Cognism puts overall B2B contact data decay at about 22.5% a year, email address decay sits in a 22.5% to 30% band (Cognism and SparkDBI), SparkDBI puts phone number decay at about 18%, and LinkedIn's Economic Graph puts annual job changes at about 10.9%. The gap between 15% at the company level and 22.5% or more at the contact level is the whole reason firmographics and contacts deserve different refresh schedules.

Ranked from most durable to most volatile, and this ordering is the one that should drive your refresh budget:

Never changes. Founding year. Once resolved, it is resolved.

Changes rarely, and loudly when it does. Legal name, primary domain, headquarters country, ownership type. A rebrand or an acquisition changes several of these at once, and it is usually announced, which is why event-driven refresh beats calendar-driven refresh for this family.

Changes on a quarterly scale. Headcount, headcount band, office list, industry classification as a provider's model re-reads a changed website. A headcount figure six months old is directionally fine for banding and unusable for growth scoring.

Changes unpredictably and matters immediately. Funding stage and last round date. A Series B announced yesterday invalidates yesterday's segment, and this is the one firmographic field whose staleness costs you a timing advantage rather than an accuracy point.

Was never fresh. Revenue for private companies. It is a modelled estimate, so it does not decay so much as remain approximate, and re-pulling it more often does not make it more true.

The practical schedule that falls out of this: reverify contacts monthly if you run daily outbound, refresh firmographics quarterly, and treat funding as an event subscription rather than a field you refresh. A team spending equally on both is overspending on the durable half.

One structural point that is easy to miss. A provider that owns and licenses a static index ships you its age at the moment of query, so an index refreshed annually contains a full year of closed offices, changed headcounts, acquired entities and dead domains on the day you buy it. A provider that resolves at request time has no index to age. Cleanlist AI is in the second category, which is a genuine advantage on freshness and a genuine limitation if what you want is a bulk file to load into a warehouse.

They differ in what they describe and in how long they stay true, and mixing them is the most common modelling error in ICP scoring. Firmographic data describes what a company is. Technographic data describes what it runs. Demographic and contact data describe the people inside it. Intent data describes what it appears to be doing right now.

Firmographic attributes (industry, headcount, revenue band, location, funding) are structural. They change on a quarterly-to-annual scale and they describe fit. Fit is the question of whether this company should ever buy from you.

Technographic attributes (CRM in use, cloud provider, marketing stack, payment processor) describe compatibility and displacement opportunity. They are sourced by scanning a company's public web footprint, job posting requirements and DNS records, which means coverage is excellent for anything with a visible web fingerprint and poor for anything running behind the firewall.

Demographic attributes in a B2B context describe the person: title, seniority, department, management level, tenure, location. They decay fastest of all four because people change jobs, and they answer who to reach rather than whether to reach anyone.

Intent signals (content consumption, review-site activity, hiring surges, search behaviour) describe timing. They are the shortest-lived and the noisiest, and they are worthless without accurate contact data underneath them, because knowing an account is in-market does nothing if the person you reach has left.

The reason to keep them separate in a scoring model is that they answer different questions and should therefore be weighted differently and refreshed differently. A model that adds a technographic point to a firmographic score and calls the total "fit" produces a number that moves when a company swaps a marketing tool, which is not a change in fit at all.

Where Cleanlist AI sits in that grid is worth stating plainly. It ships structured firmographic filters and structured demographic filters on people (seniority, department, management level, title, function). It ships no technographic filter and no intent data of any kind. What it offers instead of a technographic field is an AI column that researches a company's public footprint at the moment you ask and writes back what it finds, which is research output rather than a verified structured field you can filter a search by. If a technology install base or an intent feed is the spine of your targeting, buy a provider that sells one.

Segmentation is the primary job firmographic data does, and in practice it means cutting a total addressable market into a serviceable one along three axes: industry, size and geography. Those three carry the most weight because they are the three that actually define whether a company can be sold to, serviced and lawfully contacted.

Market sizing. Counting the companies that match an industry-plus-size-plus-geography filter is how a serviceable addressable market gets a number instead of an adjective. It pays to count more than once: teams that size the market once tend to defend the estimate, when re-cutting it ten different ways is how you find out which cut is real.

Territory design. HQ location plus headcount band assigns accounts to reps, and the two together balance territories by potential rather than by count. Splitting on geography alone gives one rep a hundred enterprise accounts and another a hundred sole traders.

Segment-specific messaging. Industry drives what the first line of an email says, and company size drives which problem it says. The same product sells to a 30-person company on "you have nobody doing this" and to a 3,000-person company on "you have four teams doing this differently", and only firmographics tell you which sentence to send.

Exclusion lists. The underrated use. Firmographic filters are more valuable for what they remove than what they add: current customers, companies below a viable size, sectors you cannot service, geographies you are not licensed in, and competitors. Every one of those is a firmographic filter, and building them once saves more rep hours than any inclusion filter.

The mechanics matter here. On Cleanlist AI, filters combine with AND across keys and OR within a key, so listing several industries widens a search while adding a headcount band narrows it. One trap that costs people entire afternoons: enum values are matched literally, so a funding stage filter takes the lowercase snake_case form (series_b matches, Series B returns nothing), and an invalid enum returns zero rows silently rather than raising an error. If a count collapses to zero right after somebody edited a filter, check the casing before you conclude the segment is empty.

Firmographic attributes are the fit half of a lead score, and fit is the half that should gate routing while behaviour decides sequencing. An ICP score built on firmographics answers whether a company belongs in your pipeline at all; an engagement score built on behaviour answers which of the qualifying companies to call first. Collapsing the two into one number is why so many scoring models rank an unqualified company that opened three emails above a perfect-fit account that has not visited the site.

Building the fit half is a four-step job. First, derive the profile from won deals rather than from a whiteboard: pull the firmographics of your last fifty closed-won accounts and your last fifty closed-lost, and look at which fields separate them. It is frequently not the field the exec team assumed. Second, weight the fields by how strongly they separate, not by how important they feel. Third, set the score to gate routing: above the threshold goes to a rep, below it goes to nurture or nowhere. Fourth, re-derive quarterly, because an ICP that is not re-derived is a snapshot of the market you sold to two years ago.

Routing rules then hang off the same fields directly. Headcount band decides SMB versus mid-market versus enterprise ownership. HQ country decides territory and which privacy regime governs the outreach. Industry decides which specialist takes the account. Funding stage decides urgency. None of these need a model, they need populated fields, which is the actual reason most routing rules misfire: the rule is fine and the field is blank.

The honest limit on any scoring layer is that it is a filter over data you already hold, so its accuracy is capped by the data underneath it. Scoring a database whose headcount field is 60% populated produces a confident number for the 60% and silence for the rest, and the rest is usually where the new logos are.

On Cleanlist AI the sequence is: search, enrich for a credit, score for five. A company record is 1 credit, a verified work email is 1, a phone number is 10, both on the same contact is 11, and an AI qualification is 5 credits per person and writes the agent's reasoning next to the number rather than only the number. A row the agent cannot answer is still charged.

Because they are measuring different things and calling them the same name. This is the single most common surprise in a firmographic evaluation, and it is a methodology difference rather than one vendor being wrong.

Headcount disagreements usually trace to the counting method. A provider counting current professional network profiles is measuring people who claim the company as an employer. A provider reading filings is measuring full-time equivalents on a payroll at a reporting date. A provider modelling from web presence is measuring an inference. For a company with contractors, a recent acquisition, or a workforce that does not maintain profiles, these three produce genuinely different and individually defensible numbers.

Revenue disagreements are wider still, and for a structural reason: for the overwhelming majority of private companies, nobody knows. Vendors publish revenue anyway because buyers filter on it, and the number is almost always modelled from headcount and sector. That means filtering on revenue for private companies is frequently a laundered headcount filter with extra confidence attached. Filter on headcount directly and you get the same segmentation with an honest label.

Industry disagreements come from taxonomy mismatch rather than error. A company can be simultaneously correct as "Software Publishers" in NAICS, "Financial Services" on LinkedIn and "Fintech Infrastructure" in a vendor's taxonomy. When you migrate providers, the industry field does not map, and a segment that worked in one system quietly changes population in another.

What to do about it, in order. Pick one provider as the system of record for each field rather than blending them, because a blended headcount belongs to no methodology and is unreproducible. Prefer the field with the most direct observation (headcount over revenue, domain over name). Store the source and the date alongside the value so a disagreement is debuggable six months later. And test before you buy: take 200 to 500 companies you know well, run them through each candidate, and check the fields against reality yourself. Coverage claims are the easiest thing for a vendor to publish and the least predictive of whether the record is usable.

Evaluate on refresh method, fill rate on your own list, and cost per usable record, in that order. Database size is the number vendors lead with and the one that predicts the least, because an index of 50 million companies refreshed once a year contains a year of decay at the moment you query it.

Refresh method first. Ask how a record gets updated and how often, and accept "not published" as an answer that tells you something. As of readings taken on 2026-09-01, Crustdata states its flat file is fully refreshed monthly from 11 sources and its API carries real-time hiring, funding and event signals; Diffbot prices refresh as an explicit metered action; Clay resolves at request time across its partner network; Coresignal and The Companies API publish no cadence on their pricing pages; ZoomInfo and Cognism publish none on the pages read.

Fill rate on your list, not theirs. Coverage is not uniform. A provider strong on North American venture-backed software is routinely thin on privately held European manufacturers, and the vendor's headline fill rate is an average over a population that is not yours. Run 200 to 500 of your own accounts through a trial and measure per-field fill: what percentage came back with a headcount, an industry, an HQ country. Then check a sample by hand, because a match is not an accuracy.

Cost per usable record. Compute it from the published credit rate and the per-record credit cost rather than from the plan price. Readings from 2026-09-01: Coresignal's $49 Mini plan carries 2,500 credits with a company record costing 10 to 20 credits, so roughly $0.20 to $0.39 each at that tier; Diffbot's free plan carries 10,000 credits a month against 25 credits per entity export, so about 400 free exports; The Companies API's $95 Startup plan carries 50,000 credits; ZoomInfo and Cognism publish no price at all, and ZoomInfo's own pricing explainer says so directly. Cleanlist AI charges 1 credit for a company enrichment, so a company record is about $0.065 on Starter ($49 per seat for 750 credits) and about $0.059 on Pro ($89 per seat for 1,500), with 25% off on annual billing. Re-check every one of these against the vendor's own page before you sign anything, because data pricing changes without notice.

Then the awkward questions. Where did the data come from, and can the vendor tell you per field? What is the deletion and opt-out process, and is it a real mechanism or an email address? Are the underlying sources disclosed as subprocessors in a document your privacy review can read? A provider that cannot answer those clearly is a provider whose compliance posture you are inheriting.

Published rates ranged from free to roughly $0.39 per company record on the entry tiers of the vendors that publish a price at all, read on 2026-09-01, and two of the largest vendors publish nothing. That spread is real, and it means the sticker price is a weak signal on its own.

The pricing shapes you will meet are four. Credit-based per record, where you buy credits and spend them per lookup: the most predictable, and the easiest to model because you can compute a per-record rate. Per seat, where cost scales with headcount rather than usage: cheap for a two-person team and punishing at twenty. Annual contract with no published rate, the enterprise default, which costs a demo cycle to price. Free tiers, which exist mainly for evaluation: Diffbot's permanent free plan at 10,000 credits a month is roughly 400 entity exports, and ZoomInfo Lite offers 10 monthly credits with no card (pipeline.zoominfo.com, read September 1, 2026; zoominfo.com/pricing itself publishes no plan price and no free tier), which is an inspection allowance rather than a working plan.

The hidden costs are where budgets actually break, and there are four of them. A search or export meter, so that finding out how many companies match a filter costs money and teams stop exploring. A charge on a miss, so you pay for lookups that returned nothing. Per-seat charges layered on top of credits. And overage priced above the plan rate, which turns a busy month into a surprise.

Cleanlist's shape, stated so it can be compared like for like: a company enrichment costs 1 credit, a verified work email 1, a phone number 10, both on the same contact 11, a validation 0.5 and an AI qualification 5. A missed email or phone number is not charged. Plans are priced per seat: Starter at $49 a month with 750 credits per seat and Pro at $89 with 1,500 per seat, and every seat's credits go into one shared team balance. Enterprise is custom. Annual billing takes 25% off. The Free plan includes 50 credits a month. A new workspace starts with 14 days of Pro: 250 credits and 3 seats, which is enough to run the fill-rate test described above on your own list before spending anything.

Cleanlist AI returns up to 180 firmographic properties on an enriched company record, resolved live across a pool of 25+ providers at the moment of the query rather than read out of an index Cleanlist AI owns. The descriptive set is what you would expect from the six families above: name, domain, industry, headcount, headquarters and office locations, founded year, ownership and funding history among the rest.

What is exposed as a structured, filterable field is a smaller set than what is returned, and the distinction matters when you are planning a workflow. Company Search filters on industry, employee count range, funding stage, location, company name and company domain. Filters combine with AND across keys and OR within a key. You can pivot straight from a company result set into People Search across those domains, filtered by seniority, department, management level, job function or title, and you spend only on the rows you decide to enrich.

What Cleanlist AI does not have, stated plainly because a page that only lists capabilities is a brochure:

No technographic dataset and no technology filter. You cannot search for companies by the software they run. An AI column can research a company's public footprint and write back what it finds, which is research output rather than a verified structured field.

No intent data of any kind. No topic surge feed, no anonymous website de-anonymisation, no predictive in-market model.

No org charts and no reporting lines. You can find the people and their seniority; you cannot read who reports to whom.

No headcount-growth filter on the public API. Company Search exposes industry, employee count range, funding stage, location, company name and company domain as structured filters, and an enriched company record returns a revenue_range value among its properties. Any documentation or diagram that draws literal API filter chips has to reflect the API, not the portal.

No bulk file. Because records resolve at request time, there is no static company file to license and load into a warehouse. If warehouse-scale ingestion is the requirement, a bulk provider is the right purchase and Cleanlist AI is not.

No SOC 2 Type II or ISO 27001 certification. If your security review makes either a hard requirement, that is a real reason to buy elsewhere.

Firmographic data is the wrong purchase in three situations, and recognising them saves a wasted quarter. First, when your qualification problem is timing rather than fit: if you already know exactly which 400 accounts you want and the question is when to call them, you need signals, not attributes. Second, when your differentiator is technical compatibility: if the deal turns on whether they run a particular platform, a technographic provider answers that and a firmographic one does not. Third, when the segment you sell to is defined by something nobody publishes, such as an internal process or a regulatory exposure that no taxonomy encodes. In that last case the honest answer is that no dataset holds your ICP, and research beats filtering.

Firmographics are also the wrong first spend when your contact data is broken. An immaculately segmented account list whose emails bounce produces nothing, and the ordering that works is: fix deliverability, then segment, then score.

On compliance, firmographic data is meaningfully lower risk than contact data because most of it is not personal data at all. A company's headcount, industry and HQ are attributes of an organisation, and GDPR and CCPA govern personal data. That distinction has real limits. A sole trader's business is frequently identifiable with the individual. A named founder attached to a funding record is personal data. And the moment firmographics are joined to a named contact, which is the entire point of the workflow, the joined record is personal data and the whole regime applies.

What that means in practice for a B2B team. In the EU and UK, outreach to a named business contact typically relies on legitimate interest rather than consent, and legitimate interest is a documented assessment rather than an assumption, so write the Legitimate Interest Assessment down. Keep a record of where each field came from, because a subject access request asks exactly that. Provide a real opt-out in every message and honour it across the whole database rather than one campaign. Under CCPA, be ready to disclose sources and honour deletion. Under CAN-SPAM, include an unsubscribe and honour it within 10 business days. And ask any provider for its subprocessor disclosure before the contract rather than after, since that document is the first thing a data protection review will request and the providers behind a waterfall are part of your processing chain whether or not you chose them individually.

Firmographic segmentation means dividing a market of businesses into groups that share company-level attributes, then treating each group differently in targeting, messaging and rep coverage. It is the B2B counterpart of demographic segmentation, and the unit being segmented is the account rather than the person.

Four cuts do most of the work. A usable segment definition is one you could hand to somebody else and get the same list back.

By industry. Example: NAICS 5415, computer systems design and related services, at 50 to 500 employees. Good for message relevance, because industry decides which problem the first line of an email names, and for excluding sectors you cannot lawfully or profitably service. Weak as a standalone cut, because the three industry taxonomies rarely agree on the same company.

By company size. Example: 200 to 1,000 employees. Good for choosing the motion as well as the message: a 30-person company buys because nobody is doing the job, a 3,000-person company buys because four teams are doing it differently. Headcount bands are the most dependable segmentation variable in B2B because headcount is observable from the outside.

By revenue band. Example: $10M to $50M in annual revenue. Good for pricing tiers and for a budget floor. Read the section above on why providers disagree before you build on it, because for private companies revenue is usually modelled from headcount, so a revenue cut is often a headcount cut with more confidence attached.

By geography. Example: headquartered in the United States or Canada, excluding Quebec. Good for territory assignment, language, timezone coverage and, decisively, for which privacy regime governs the outreach.

The cuts compound. Industry plus size plus geography is the standard three-way definition that turns a total addressable market into a serviceable one, and every additional axis sharpens the message and shrinks the count at the same time.

The segment that pays for itself fastest is the one defined by what it removes. A saved definition of everything you will never sell to, meaning current customers, companies under the viable size, unlicensed geographies and unservable sectors, is written once and returns rep hours every week afterwards.

Firmographics are the descriptive characteristics used to classify and segment organisations: industry, size, location, performance, structure and age. The word is the plural noun for the attribute set, and firmographic data is the data that populates it. In practice the two are used interchangeably by every vendor and every textbook, so an answer to one is an answer to the other.

It is pronounced fur-mo-GRAF-iks, stress on the third syllable, and it is built from firm in the business sense plus the -graphics of demographics. The word entered marketing usage as a direct analogy: demographics describe the people who make up a consumer market, firmographics describe the companies that make up a business market.

Synonyms for the same idea, which is worth knowing because they change which search results you get, are company attributes, company demographics, business demographics, company profile data and, when the subject is which supplier to buy from, company data. An analyst building a segmentation model usually types the first. A buyer looking for a vendor usually types the last. The fields underneath are the same.

Where the term earns its keep is in what it excludes. A firmographic attribute is true of a company whether or not anybody is selling to it. It does not describe an action, an installed technology or a person. Downloading a pricing guide is behavioural. Running a particular data warehouse is technographic. A job title is demographic. Holding that boundary is what lets a scoring model weight fit and timing separately instead of adding them together and calling the sum a score.

The six variables the term conventionally covers are set out immediately below. The field-by-field breakdown, with where each one is sourced and how fast it goes stale, is earlier on this page.

Six variables make up the standard firmographic set: industry, company size, location, performance and revenue, status and structure, and age. That is the list market research and B2B marketing converge on, and it is the list Google's AI Overview printed under the heading "Key Firmographic Variables" when we measured the term firmographics on September 7, 2026. It maps onto the field families described earlier on this page, with identity split out separately there because a join key is an operational field rather than a descriptive one.

Industry. The sector a company operates in, expressed as a SIC or NAICS code, a professional network label, or a vendor taxonomy. Good for message relevance and for sector exclusions. Reliability is middling: the code is accurate as a filing and frequently wrong as a description, because it was chosen at registration for tax reasons and never revisited.

Company size. Employee headcount and headcount band, occasionally market share. Good for segmentation, pricing tier and rep assignment. The most reliable variable in the set, because headcount is observable from outside, with a known undercount in sectors where staff do not maintain professional profiles.

Location. Headquarters and operating regions. Good for territory design, language, timezone coverage and privacy regime. Reliable for HQ country, much less so for the office list, which is often reconstructed from where a company advertises jobs rather than from where it has desks.

Performance and revenue. Annual revenue, revenue band, financial stability. Good for budget qualification and for a size floor. The least reliable variable in the set, because for private companies it is modelled rather than observed.

Status and structure. Ownership type, legal status, parent and subsidiary relationships. Good for working out whether the buyer sits at this entity or at a parent two countries away, and for spotting an account you already serve under another name. Reliable, and the most under-used variable of the six.

Age. Founding year and years in operation. Good as a rough proxy for how formal the buying process will be. The most durable variable of all, because it never changes once it is resolved.

Free firmographic data exists in three places: national company registries, the aggregators that normalise them, and the company's own public surface. All three are real sources, and none of them builds a target list on its own.

National company registries. The primary record, published by the government that holds it. The UK's Companies House publishes a free monthly company data snapshot and a free URI service for basic company details (gov.uk, read September 7, 2026), and runs a Public Data API against the live register, rate limited to "up to 600 requests within a five-minute period" and requiring a registered application and key (developer-specs.company-information.service.gov.uk and developer.company-information.service.gov.uk, both read September 7, 2026). In the United States, the SEC states that "Access to EDGAR's public database is free", serves submissions and XBRL company facts as REST APIs on data.sec.gov, states that "our current maximum access rate is 10 requests per second" and asks callers to "declare your user agent in request headers" (sec.gov, read September 7, 2026). State and provincial registries cover the entities that never file federally, mostly behind search forms rather than APIs.

Aggregators over registries. OpenCorporates assembles company records from over 140 government registries and other official sources around the world (opencorporates.com, read September 7, 2026), which solves the part that makes registries painful, namely that every jurisdiction publishes a different shape.

The company's own surface. The website, the about page, the careers page and the job postings. Postings are the most under-used free firmographic source in B2B, because one posting states a location, a team, a seniority level and frequently a technology, and the company published it itself.

Primary research. For attributes no taxonomy encodes, such as an internal process or a regulatory exposure, a survey or a set of calls is the only source. It is free in cash and expensive in time.

What free sources will not do, stated plainly so you can budget honestly. They carry no join key that lines up with your CRM, because registries key on a registration number while your CRM keys on a domain. They carry no industry taxonomy that is consistent across jurisdictions. They carry no headcount, because nobody files it. And they carry no contact data at all. Registry data is the right answer for legal verification and compliance checks, and the wrong answer for building a prospecting list.

A company data API is an interface that returns structured company records to software rather than to a person, and the first thing to establish about any of them is whether you are calling a search endpoint or an enrichment endpoint. A search endpoint takes filters, such as industry, headcount band and country, and returns the companies that match, which is how you build a list you did not previously have. An enrichment endpoint takes an identifier you already hold, such as a domain, a company name or a LinkedIn URL, and returns the full record for that one company, which is how you fill a CRM. Vendors price and meter the two differently, and a plan sized for enrichment can be useless for discovery.

Three families exist, and they answer different questions.

Aggregator and enrichment APIs assemble records from many upstream sources and expose them under one schema. Best for lead generation, CRM cleaning and account research. This is the family most B2B teams mean by the phrase.

Official registry APIs serve the legal record straight from the body that holds it: OpenCorporates across more than 140 registries, Companies House for the UK, EDGAR for SEC filers (all read September 7, 2026, with the terms quoted in the section above). Best for verification, compliance and auditing. They are also the answer to whether a company data API can be free, because EDGAR states in writing that it is and the registry tier is where free access lives, while a commercial aggregator generally is not.

Financial and private-market APIs carry deep financials on private companies for diligence and risk work rather than for outbound.

Where Cleanlist AI sits, stated exactly. The public REST API at api.cleanlist.ai/api/v2 takes structured filters and has no natural-language search endpoint, so your integration composes the filters itself. Natural language happens on two other surfaces: Clu inside the app, and the MCP server, where the model writes the filters for you from a plain-language instruction. A company enrichment costs 1 credit. Company Search takes structured filters on industry, employee count range, funding stage, location, company name and company domain, and enum values are matched literally, so a funding stage of series_b matches while Series B returns zero rows without raising an error. The REST API opens on the Pro plan at $89 per seat a month and the MCP server on Starter at $49, and both are open during the 14-day Pro trial.

In account-based marketing, firmographics do the tiering. They decide which accounts get a bespoke programme, which get a light-touch one, and which get nothing, and that is the whole of their job. Most ABM programmes go wrong by asking firmographics to do the other half as well.

Tier definition. A tier is a firmographic band with a budget attached. Tier one might be 30 accounts above 2,000 employees across two industries, each with named creative and an executive sponsor. Tier two might be 300 accounts from 200 to 2,000 employees running industry-specific content. Tier three is everything else inside the ICP, running scaled campaigns. Every one of those boundaries is a headcount, industry or geography filter, which is why a programme sitting on half-populated firmographics ends up with arbitrary tiers.

Account list construction. Build the list by filtering the market, then have reps review it rather than replace it. Firmographic filters do the first pass, exclusion filters strip out current customers and unservable geographies, and human judgement is applied to what survives. A list assembled by asking reps for names produces a different and usually smaller market than the one you actually sell to. Your best customers are another starting point: an agent can find similar companies that share their industry, headcount band and funding stage, leaving out every company already in your CRM.

Buying committee mapping. Once the account list exists, firmographics stop and demographics start. Company size predicts how many people sit on the committee and how senior the final approver is. It does not name them. Titles, seniority, department and function do.

What firmographics cannot do here. They cannot tell you when. An account that fits perfectly in January fits perfectly in June, and no firmographic field says which month to run the programme. Timing comes from hiring, from funding events, from a renewal date or from an intent feed, and none of those is a firmographic attribute. A programme that treats a fit score as a timing score spreads its budget evenly across a year on accounts that were in-market for six weeks of it.

On Cleanlist AI, Company Search and People Search build the list however many times you re-cut the market, and you spend only on the rows you decide to enrich.

A firmographic template is a field schema you can paste into a sheet or a CRM object, and a usable one carries five things for every field: the column name, the type, the source family, the decay class and the refresh trigger. The decay class is the column most published templates leave out, and it is the one that decides your refresh bill. The five classes used below are never, rare, quarterly, event-driven and modelled estimate.

company_domain. Text. Source: registry and web. Decay: rare. Refresh on a rebrand. This is the join key, so it goes first and it is the one column that must never be blank.

legal_name and trading_name. Text. Source: registry. Decay: rare. Refresh on a rebrand or an acquisition.

linkedin_company_url. Text. Source: web. Decay: rare. Refresh on a rebrand. The secondary join key.

industry_naics and industry_sic. Code. Source: government registry. Decay: rare. Review annually.

industry_vendor. Text. Source: provider taxonomy. Decay: quarterly. Refresh when the company's site changes. Put the vendor's name in the column header, because this field does not port between providers.

employee_count. Integer. Source: professional network counts or filings. Decay: quarterly. Refresh quarterly.

employee_band. Enum. Source: derived from employee_count. Decay: quarterly. Derive it in the query rather than storing it independently, or the two will disagree.

revenue_band. Enum. Source: modelled. Decay: modelled estimate. Label the column as an estimate so nobody filters on it believing it was observed.

hq_country, hq_state, hq_city. Text. Source: registry and website. Decay: rare. Refresh on a headquarters move.

founded_year. Integer. Source: registry. Decay: never. Resolve once and leave it.

ownership_type. Enum. Source: registry and press. Decay: event-driven. Refresh on an acquisition or a listing.

parent_company_domain. Text. Source: registry and press. Decay: event-driven. The field that tells you an account is already yours under another name.

funding_total, last_round_type, last_round_date. Source: filings and press. Decay: event-driven. Subscribe to the event rather than scheduling a refresh.

Two columns belong beside every field above and are almost always missing: source and as_of_date. Without them, a disagreement between two providers is unarguable six months later and a stale value is indistinguishable from a fresh one.

We ran the six keywords this page targets through Google US desktop results on September 7, 2026 and recorded each AI Overview verbatim. Five of the six resolved a live SERP, all five carried an AI Overview, and none of the five named Cleanlist AI. Here is what they did name, so you can see who Google currently treats as the authority and check it against your own search.

"firmographic data" (320 searches a month, $11.69 CPC, low competition). The answer opens with a definition quoted from pipeline.zoominfo.com and commonroom.io, prints a five-item "Core Attributes" list of industry, company size, annual revenue, location and ownership structure sourced to demandbase.com, and closes by pointing readers at ZoomInfo and Crunchbase. Nine sources cited. Organic one, two and three on the day were hginsights.com, demandbase.com and pipeline.zoominfo.com, and the highest of those is a vendor glossary entry.

"firmographics" (1,900 a month, $31.39 CPC). The definition is quoted from Wikipedia and foundryco.com, and Wikipedia holds organic position one. The centre of the answer is a six-item "Key Firmographic Variables" list co-cited to seven sources at once: surveymonkey.com, about.crunchbase.com, en.wikipedia.org, demandbase.com, hginsights.com, uk.indeed.com and cognism.com. No vendor is recommended anywhere in it.

"firmographic data providers" (70 a month). Six vendors named, each in one sentence: ZoomInfo with "over 100 million company records", Dun and Bradstreet with "500 million records" and the DUNS numbering system, Techsalerator with "195 countries", plus Cognism, People Data Labs and CoreSignal. Three of the six carry a countable scale number. Not one carries a price.

"company data provider" (110 a month, $55.90 CPC) and "company data api" (40 a month, $49.11 CPC). The company data api answer sorts vendors into categories by data type rather than ranking them, and the entries it quotes carry a coverage number attached to a name.

Two things follow. The definitional job on this term is already being done by an encyclopedia and a set of vendor glossaries. The practical job, meaning which fields exist, which are reliable, how fast each decays and what to do with them, is being done by nobody in particular, which is what the rest of this page is for. And the entries Google reproduces run to two sentences, a "Best for" segment followed by a mechanism, with a countable number inside about half of them.

So here is ours in the same shape, with numbers you can check on the pricing page. Cleanlist AI is best for teams that want every rep drawing on one shared credit balance. It runs waterfall enrichment across 25+ providers and charges 1 credit per verified work email and 10 per phone number, with a lookup that returns no data costing nothing, starting at $49 per seat a month, and a new workspace starts with 14 days of Pro: 250 credits and 3 seats.

Method, so this is reproducible: DataForSEO Google Organic Live Advanced, location United States, language English, desktop, depth 50, AI Overview loaded asynchronously, one task per query, collected September 7, 2026. Volumes and CPC are Google Ads figures for the same location and language. AI Overviews are volatile between identical runs, so treat any single-snapshot naming count as a range rather than a point. We will re-run this and update the section rather than publish a second page.