Where does firmographic data come from: the registry, the filing, the profile and the crawl behind every field.

Definition

A firmographics source is the place a company attribute was originally recorded. Where does firmographic data come from in practice? From a government register, a regulatory filing, a statistical program, the company's own website, a job ad, a profile a page admin filled in, your CRM, a crawl, or a partner that contributed records.

Not from one place. Every firmographic field on an account record has its own origin, its own refresh cycle and its own blind spot. Industry, headcount, revenue and location can come from four different sources, even inside a single vendor's record.

That matters because the data source sets the ceiling on what the data field can mean. A registry can prove a legal name. It cannot tell you how many people work at the office you are about to call.

Firmographics, and what B2B teams use the data for

Firmographics describe businesses the way demographics describe people. On this page, a firmographic data set means industry, company size, employee count, annual revenue, location, ownership type, corporate structure and growth stage.

B2B sales and marketing teams use firmographic data for four jobs. Each job leans on a different firmographic field, and each field leans on a different data source, which is why provenance is a practical question rather than an academic one.

• Market definition: firmographic segmentation turns a vague total market into a countable set of businesses in a named industry and size band.
• Ideal customer profile: the business attributes your best customers share, written down so marketing and sales filter the same way.
• Prioritization and scoring: firmographic fit decides which accounts a sales rep works first, before any behavior is observed.
• Routing and territories: industry, revenue band and location decide who owns the account and which team the lead reaches.

A practical way to find your own firmographics is to look backward. The business attributes shared by current customers become the profile, and new customers are then found by matching that pattern of industry, size and growth stage across a market.

In this page's view, firmographic data changes more slowly than contact data, which is its advantage for segmentation. It is also the layer your technographic and intent signals attach to, so a wrong company record spoils every signal joined to it.

None of those four jobs survives a data field whose source nobody can name. A marketing campaign aimed at a segment built on an unverified industry code reaches a market that does not exist in the way the plan assumed.

Firmographic data against the other B2B data types

Firmographic data is one layer of B2B data, and the layers do not share data sources. Knowing which layer a data field belongs to tells you where to go looking when the information turns out to be wrong.

The table is this page's framing. Firmographic information ages slowly, which is why it anchors the record. The other layers attach to it, so a company matched to the wrong business entity poisons every technographic and intent signal joined on afterward.

The wider picture of these layers and the rules that apply to them sits in the B2B data entry, which also covers what public business data you may and may not reuse.

Where does firmographic data come from, source family by source family

This page groups firmographic data sources into eight families. A commercial database is a blend of several of them, and in this page's view the weighting changes by country and company size.

Read that third column first. Control decides the failure mode. A data field the company controls goes stale when nobody updates it. A data field a collector controls goes wrong when the detection misfires.

Public registries and filings

Registries are the strongest firmographic data source for identity and the weakest for commercial detail. They exist to record legal facts, not to describe a sales target.

What a US filing gives you

The SEC publishes company metadata for every EDGAR filer. Microsoft's submissions record returns the registrant name, the Central Index Key, an SIC code, the employer identification number, the state of incorporation, the fiscal year end, the filer category and a business address.

There is no employee count in that metadata, and no revenue figure. On the check date, the record's description, website and LEI fields were also empty, even though GLEIF holds an LEI for Microsoft Corporation. Identifiers do not travel between systems on their own.

The SIC code deserves care. The SEC states that these codes indicate the company's type of business and are also used in the Division of Corporation Finance as a basis for assigning review responsibility for the filings.

Microsoft's record carries 7372, Services-Prepackaged Software, and an ownerOrg value of 06 Technology. One four digit code has to stand for a company with many business lines. It was never meant to segment a market.

What a US state registry gives you

The SBA's registration guide says most states require businesses to register with the Secretary of State's office, a business bureau or a business agency. Delaware's Division of Corporations is a useful example, because its free name search shows what a registry offers and withholds.

The free entity information consists of the entity name, file number, incorporation or formation date, registered agent name, address, phone number and residency. Results return both active and inactive entities, and the Division says this is not an indication of current status.

Status costs extra. Delaware says additional information can be obtained for a fee, such as a Certificate of Status. It also says it strictly prohibits mining data, and that use of automated tools in any form may result in suspended access.

Two lessons follow. A state registry confirms that an entity exists and who receives legal papers for it. The registered agent address on that record describes the agent, not where the company's people work.

What a UK filing gives you

Companies House holds the registered office, officers, people with significant control and filing history. Companies must review their records and file at least one confirmation statement every 12 months, even if nothing has changed.

The industry code on the register is the company's own choice, taken from a condensed SIC list, and it can be changed on that same statement. The Companies House service itself says it does not check the accuracy of the information filed.

Revenue is where the register goes quiet. A UK company counts as small if it meets any two of a turnover of £15 million or less, £7.5 million or less on its balance sheet, and 50 employees or less.

A small company may choose whether or not to send a copy of the director's report and profit and loss account to Companies House. A micro-entity, which meets any two of £1 million turnover, £500,000 on its balance sheet and 10 employees, can send only its balance sheet.

So for a small private UK company that uses those options, the public record can contain no turnover at all. Any revenue figure you see against it came from an estimate, not from a filing.

Identifiers that tie records together

The Legal Entity Identifier is a 20 character code under ISO 17442 that identifies a legal entity. GLEIF describes it as a public good, available free of charge, and its data answers who is who and who owns whom.

Identifiers are the quiet hero of firmographic work. A name match guesses. A registration number, an LEI or a domain resolves, which is why, in this page's view, careful lead enrichment keys on identifiers rather than on strings.

Your CRM shows the same lesson. Salesforce's standard account matching rule, covered in account based marketing data, compares account names with acronym, edit distance and exact methods at a threshold of 70, but matches websites exactly. In this page's view, a domain is the closest thing many accounts have to a shared key.

What a filing can and cannot tell you

Headcount in a US annual report is a good example of the gap. Item 101(c) of Regulation S-K asks for a description of the registrant's human capital resources, including the number of persons employed, to the extent material to an understanding of the business taken as a whole.

The item does not say how to count. In this page's view, two filers can both comply and still count contractors, part-time staff or subsidiaries differently.

Company websites and job ads

A website is the fastest-moving data source a company fully controls. It carries positioning, office lists, leadership pages, customer logos and, where the company adds it, machine-readable markup.

The schema.org Organization type defines exactly the fields a firmographic record wants. It includes numberOfEmployees, defined as the number of employees in an organization, plus naics, isicV4, duns, leiCode, vatID, taxID, foundingDate and parentOrganization.

When a company publishes that markup, the field is self-reported but structured, dated and attributable. That is a better provenance story than many purchased records offer.

Job ads are the other half. Google's job posting documentation requires datePosted, description, hiringOrganization, jobLocation and title, and recommends properties such as validThrough, employmentType, baseSalary and identifier.

It also says the hiring organization must be the name of the company and not the specific location that is hiring, and that jobs no longer open should be expired, for example with a validThrough date in the past.

• What job ads prove well: that a function is being staffed, in a named place, on a dated posting.
• What they suggest: growth direction, tooling, team structure, and occasionally a budget cycle.
• What they cannot prove: total headcount, revenue, or whether the role was ever filled.
• Where they mislead: agency reposts, evergreen ads left open, and listings under a parent brand.

Job ads are a chronographic signal wearing firmographic clothes. Treat them the way you would treat intent data: as timing, not as an attribute.

Self-reported profiles

Professional network profiles are a large self-reported firmographic data source in B2B. It is worth reading the documentation rather than the marketing, because the field definitions are explicit.

LinkedIn's organization schema exposes a staffCountRange, described as the range of the number of staff associated with the entity. The permitted values are buckets, not counts. The second column below is this page's reading, not LinkedIn's.

The same schema carries an organizationType with values such as PUBLIC_COMPANY, PRIVATELY_HELD, NON_PROFIT, GOVERNMENT_AGENCY and PARTNERSHIP, and an organizationStatus including OPERATING, REORGANIZING, OUT_OF_BUSINESS and ACQUIRED.

Parent links are typed too, as SUBSIDIARY, ACQUISITION or SCHOOL. Specialties are documented as admin-defined tags, and pages carry an autoCreated flag. LinkedIn's ads help adds that pages may be created by a page admin or auto-generated using third-party data sources.

How LinkedIn fills size, industry and revenue

LinkedIn's targeting help says company size primarily uses the size a page admin entered. If no size was entered, size is determined by the number of member accounts associated with the page. In this page's view, that fallback depends on how many staff keep a profile.

Company industry primarily uses the page admin's input, and LinkedIn says the list of industries is a proprietary taxonomy developed by its own team. If a company has no industry, LinkedIn uses the industry a member added to their own profile.

Notice what is absent: there is no revenue field in the documented organization schema. LinkedIn's ad targeting does offer company revenue, which its help page describes as estimated annual revenue, and growth rate, which can be inferred from similar companies.

If you filter on industry there, you are filtering on that proprietary taxonomy from the LinkedIn industry list, not on NAICS. The B2B target audience entry shows how to write one rule and map it to each tool.

Government statistics: NAICS, the Economic Census, SUSB and County Business Patterns

Statistical programs are the most methodologically careful firmographic data source and the least usable at account level. The reason is legal, not technical, and it shapes how you can use them in B2B marketing.

NAICS: the classification behind the statistics

The Census Bureau describes NAICS as the standard used by federal statistical agencies in classifying business establishments. It was adopted in 1997 to replace the Standard Industrial Classification system, and the 2022 version has twenty sectors.

The NAICS FAQs make a point every data buyer should know. There is no central government agency that assigns, monitors or approves NAICS codes for establishments, and there is no central register of the official code for a business.

The Census Bureau assigns one code to each establishment based on its primary activity, generally the activity that generates the most revenue. Other agencies assign their own codes for their own programs, so one business can carry different codes in different places.

The Economic Census

The US Economic Census is conducted every five years, for years ending in 2 or 7. Response is required by law under Title 13 of the United States Code, and it tabulates establishments, employees, payroll and a measure of output such as sales, shipments or revenue.

Then Section 9 of that same law makes the responses confidential, usable only for statistical purposes. So this business dataset is published as aggregates and never as a per-company record you can buy.

Statistics of U.S. Businesses and County Business Patterns

Statistics of U.S. Businesses is an annual Census series with the number of firms, establishments, employment during the week of March 12 and annual payroll, by geography, industry and enterprise size. Receipts are added for years ending in 2 and 7.

County Business Patterns is the establishment view: an annual series of establishments, employment, first quarter payroll and annual payroll by industry and establishment size, down to county and ZIP code. The Census Bureau says private businesses use it to analyze market potential and set sales quotas.

Both programs count businesses with paid employees, exclude some industries such as crop and animal production and public administration, and take precautions to avoid disclosing an individual employer. The CBP page also carries a notice that its information is no longer current while the Census Bureau reviews its disclosure approach.

The use column is this page's suggestion. None of these programs names a company. Use them to check whether a vendor's account count for a segment is plausible, not to build a target list.

Firm, establishment and enterprise

The SUSB glossary defines an establishment as a single physical location where business is conducted. An enterprise, or company, is one or more establishments under common ownership or control. Enterprise size is the summed employment of all its establishments.

Paid employment there means full and part-time employees on the payroll in the pay period including March 12, and excludes sole proprietors and partners of unincorporated businesses. That is one specific definition of headcount, and commercial records rarely state theirs.

Commercial surveys and panels

Commercial panels work the same way in miniature, as this page understands them. A research firm surveys a sample, then models the rest of the market from it. The modeled part is an inference, which is fine if it is labeled and poor if it is sold as a fact.

SBA size standards: the federal definition of a small business

"Small business" sounds like a firmographic band. In US federal contracting it is a legal test, and the SBA's size standards show how much definition hides behind a size field in a commercial database.

The SBA says size standards define the largest size a business can be to compete for contracts set aside for small businesses. They vary by industry, are set for each NAICS code, and are generally based on number of employees or annual receipts.

The SBA's summary says most manufacturing companies with 500 employees or fewer, and most non-manufacturing businesses with average annual receipts under $7.5 million, will qualify as small. It adds that there are exceptions by industry, so check the table for the exact code.

• Employees: the average number of people employed for each pay period over the latest 24 calendar months, counting anyone on the payroll as one employee regardless of hours.
• Annual receipts: total income plus cost of goods sold, averaged over the latest five complete fiscal years for federal contracting.
• Affiliates: the employees and receipts of affiliates are included, and affiliation rests on the power to control, whether exercised or not.

Set that beside a profile bucket or a modeled revenue figure. Three systems can describe the same company as small, mid-market and enterprise, and each is correct under its own definition. The B2B companies entry covers how the SBA and NAICS frame business types.

First-party data: your CRM, forms and customers

The firmographic data source most teams forget is their own. Your CRM, your forms and your billing system hold company information that customers and prospects gave you directly, for a reason, at a known time.

• Form fields: a company size or industry picklist on a demo request is self-reported by the buyer, which makes it current but coarse.
• Sales notes: a rep's discovery call records structure, sites and buying process that no database holds.
• Billing records: the contracting entity, its legal name and billing address, which is the record finance trusts.
• Customer data: the firmographics of accounts that bought, which is the evidence behind an ICP.

First-party records have a provenance problem of their own: they mix with enrichment. Once a vendor overwrites a field the customer typed, the CRM forgets who said what, and the most reliable value is gone.

The fix, in this page's view, is a simple rule: never overwrite a first-party value with an inferred one. Store the enriched value beside it, and let reporting choose. CRM data cleansing covers the cleanup when that rule was missing.

A survey is the deliberate version of this source. Asking customers for industry, headcount band and revenue band produces marketing data you can trace, as long as the bands match the ones you segment on later.

Web crawling and partner contribution

Crawling turns the open web into structured fields. A crawler resolves a domain, fetches pages, extracts markup and text, and writes attributes such as industry keywords, location mentions, technologies and company descriptions.

RFC 9309 standardizes the Robots Exclusion Protocol. It requires the rules to live in a file named /robots.txt at the top level of the service, says crawlers should not use a cached copy for more than 24 hours, and states that the rules are not a form of access authorization.

Two practical consequences follow, in this page's view. Sites that disallow crawling are thin in crawled datasets from crawlers that honor the rules, which quietly biases coverage toward companies with open marketing sites.

And extraction errors are systematic, not random. A holding page, a regional subdomain or a parked domain produces a confident record about a company that barely exists at that address.

Partner and co-op contribution

In this page's view, contribution models are the least visible data source and a common explanation for a data field nobody can trace. Records move between partners, co-ops and resellers, and provenance thins with every hop.

• Data co-ops: members contribute records from their own systems and draw on the pooled set in return.
• Contributory networks: a tool installed by users reports back the companies and fields it observes.
• Licensing and resale: one vendor's database sits inside another vendor's product under a different label.
• Partner exchanges: account lists shared for a joint campaign, then absorbed into a permanent database.

The practical effect is circular agreement. Three vendors can report the same wrong headcount because all three trace back to one contributed record, which reads like confirmation and is not.

It also matters legally. If you cannot name where a record came from, you cannot answer the question every privacy regime asks first, which is covered in the B2B data entry linked above.

Firmographic data sources compared

Firmographics that came from one clean place are rare. Almost every account record in a sales tool is a merge, and the merge is where the disagreements you will argue about later were created.

Comparing firmographic data sources on one axis, such as accuracy, hides the trade. Compare them on what they can prove, how fast they move and where they are blind. The table is this page's summary of the sections above.

Where each firmographic field actually comes from

This is the firmographic table worth keeping. It maps the common firmographic data fields on an account record to the source that tends to produce them and to the reason the value drifts. The mapping is this page's framing.

Two fields in that table are estimates for many private companies. Those two are exactly the fields teams filter on hardest when they build a target account list.

Why providers disagree on headcount and revenue

In this page's view, disagreements between two vendors trace to one of six causes. Naming the cause is faster than arguing about which vendor is right.

Cause four is, in this page's view, the most damaging, because it is invisible. The precision of a number like 347 employees implies a measurement, when the underlying record was a band that a joining step converted into a value.

Cause two is the one people forget. NAICS is production-oriented, grouping establishments by similarity in the processes used to produce goods or services. Your CRM is built around a buying entity.

Why the same company gets two industries

Industry is the field most likely to be confidently wrong, because at least four different systems are in play and none of them was designed for prospecting.

• Statistical classification: the Census Bureau assigns one NAICS code per establishment by primary activity, for statistics only.
• Regulatory routing: the SEC says the SIC code on a filing is also used to assign review responsibility for filings.
• Self-selection: the code on a company register and the industry on a profile are both chosen by the company from a fixed list.
• Inference: a crawler reads the site and assigns a vendor's own category, which maps to none of the above.

NAICS is hierarchical, from a two digit sector to a three digit subsector, a four digit industry group, a five digit industry and a six digit national industry. Truncating a six digit code to two digits to fit a picklist destroys the distinction you were filtering on.

The fix is not to find the true industry. It is to decide which system your ideal customer profile is written in, then convert everything else into that one, and store the original code beside it.

Legal entity, operating brand and the address problem

Location disagreements can mean the two data sources are describing different things, both correctly. A worked example makes it concrete, using two public records anyone can open.

In the global LEI index, Microsoft Corporation's legal address is recorded in Tumwater, Washington, at a suite number. Its headquarters address in the same record is One Microsoft Way in Redmond.

The SEC's EDGAR metadata for the same company lists the business address as One Microsoft Way, Redmond. The registry-style address and the operating address are both accurate and point to two different places.

Now imagine that split applied to a company you have never heard of, in a country you do not know, with no headquarters field published anywhere. A location filter can silently drop that account.

• Registered address: where legal documents are served, which can be a registered agent, an accountant or a corporate services provider.
• Headquarters address: where leadership sits, which the company defines itself.
• Site address: where the people you want to meet actually work, sometimes only visible in job ads.
• Billing address: where the contracting entity pays from, which sales discovers last and finance cares about most.

Before a territory or a routing rule depends on a location field, write down which of those four the field holds. That single line prevents many of the arguments that follow.

How to sanity check a field

You do not need a data team to audit a firmographic field. You need one open tab per data source family and a rule about what counts as agreement. This routine was written for this page.

1. State the claim and the unit

   Write the field, the value and the entity it describes. For example: employee count, 340, for the UK subsidiary rather than the global group.

2. Open the registry

   Find the legal entity by number, not by name. Confirm it exists and that the name on your record matches a real registration. Where status is a paid extra, note that you have not confirmed it.

3. Look for a filing obligation

   Decide whether any published financial statement should exist at all. If the company qualifies as small or micro, expect no turnover figure and stop treating revenue as a fact.

4. Read the company's own pages

   Check the about page, office list and any structured markup for a self-reported number. Note the date the page was last changed if you can see it.

5. Count the open job ads

   Search the careers page for the location and function in question. Hiring in a place confirms presence there better than any address field does.

6. Compare the profile bucket

   Check whether the vendor's precise number sits inside the self-reported size band. A value outside the band suggests the two records describe different entities or different dates.

7. Rule on it

   Accept, correct or mark unknown. Unknown is a legitimate outcome and is cheaper than a confident wrong value driving routing.

Run this on a sample, not on everything. In this page's view, twenty accounts sampled from the segment you are about to work will show whether the data field is usable across the rest.

Building a target list from traceable sources

Much B2B lead generation starts from a purchased list and hopes the firmographic data holds. The alternative is to build the list from data sources you can name, in the order that puts the cheapest evidence first. The steps were written for this page.

1. Size the market from statistics

   Use SUSB or County Business Patterns to see roughly how many firms or establishments sit in your industry codes and size bands, so a later list has a number to be checked against.

2. Start from identity, not from attributes

   Pull the businesses in scope from your own customer data, association memberships or a marketplace directory. Then confirm identity by registry number, the field a public source can actually prove.

3. Attach the self-reported layer

   Add industry, size band and specialties from each company's own profile and site. Keep the band as a band, and store the date you read it.

4. Add dated growth signals

   Job ads, funding records and leadership changes tell you which of those businesses is moving now. Store them as events with dates, never as permanent attributes.

5. Enrich the gaps, then label them

   Use a vendor or a data enrichment API for what is still missing, and mark those values as inferred rather than observed.

This order matters because each step costs more than the one before it. Statistics and registry searches are free to read, self-reported information is cheap, signals take work, and modeled fields cost money and carry the least certainty.

It also gives your sales team something to say. A sales rep who knows a business appears in a register, hires in a city and published a size band has three verifiable facts to use in outreach.

What to record when you accept a field

Provenance is a property of the data record, not a memory in someone's head. Four columns beside each imported data field turn an argument into a lookup. The scheme was written for this page.

If your account object cannot hold it, the fields belong wherever your system of record keeps source metadata, so reporting can group by it later.

The payoff arrives the first time a sales rep disputes a number. You answer with a data source and a date instead of a debate, and you fix the import rather than the record.

What the data source changes about firmographic segmentation

Firmographic segmentation inherits the weaknesses of the data source underneath it. The segment looks equally solid on a slide whether the industry field came from a filing, a company's own picklist choice or a crawler's guess.

Firmographic segmentation that holds up

Good firmographic segmentation starts from your existing customers, not from a data vendor picklist. Export the accounts that became customers, look at which business attributes they share, and record the source of each attribute before you build a segment on it.

A segment based on industry plus company size is a common starting point in B2B marketing, and a source-sensitive one. If half the list is coded on one taxonomy and half on another, the segmentation is arithmetic on two different things.

Revenue based segmentation deserves its own warning. In markets where private businesses publish no financial statements, a revenue segment is a segment of estimates, so treat it as a ranking rather than a hard filter.

The information you want beside every segment is short: how many companies, from which data source, observed when. Marketing plans based on that line survive contact with the sales team.

The practical rule for B2B marketing and sales targeting is to segment on the fields you can trace and to treat the rest as hypotheses. To use round numbers, a segment of 4,000 companies you cannot verify is weaker than 400 you can.

It also changes how you buy. When you compare data providers, ask which sources produce industry, size and revenue in your target market, rather than comparing headline counts of companies in a database.

Common mistakes about firmographic sources

• Treating a data vendor as a source. A vendor is a blend of data sources, and the blend is what you are actually buying.
• Reading agreement between three tools as confirmation, when all three may resolve to one contributed record.
• Storing a midpoint of a range as if it were a count, then filtering on it to the nearest employee.
• Filtering on revenue in markets where private companies file no profit and loss account, so every value is modeled.
• Mapping a six digit industry code onto a two digit picklist and losing the distinction the filter depended on.
• Assuming a NAICS code in a file is official, when no central agency assigns or approves NAICS codes for businesses.
• Calling a prospect "small" without saying whether that means a profile band, an SBA size standard or a vendor segment.
• Using a registered address as a territory key, when it can belong to a registered agent or a corporate services provider.
• Letting enrichment overwrite a value the customer typed into a form, so the CRM loses its only first-party record.
• Refreshing the whole database on one schedule, when registry fields move yearly and hiring signals move weekly.
• Never writing down the data source, which turns every later disagreement into an opinion.

In a sequence

Provenance changes the first line of an email. When a firmographic field came from a dated, public data source, you can name what you saw instead of implying research you did not do. The template below was written for this page and does that and nothing more.

Subject: {{observation}} at {{companyName}}

Hi {{firstName}},

I was reading {{sourceName}} from {{sourceDate}}, and it lists {{observation}} at {{companyName}}.

If that is still current, it can put {{problem}} in front of {{roleArea}}. We help companies like {{similarCustomer}} with {{outcome}}.

Is that yours to solve, or should I be asking someone else?

{{senderName}}
{{senderCompany}}

Frequently asked questions