Firmographic Data: The Company Attributes Your Targeting Actually Uses
Firmographic data covers company-level attributes such as industry, headcount, revenue, location and ownership structure. It defines who is eligible to buy and draws the outer boundary of a target audience. It cannot tell you when to make contact, because it describes a company's standing condition rather than anything that has recently changed.
Key takeaways
- Firmographics describe what a company is, so a segment built on them alone contains no information about whether anything has recently changed there.
- Every field arrives self-reported, filed in a public register, or inferred by a model, and the route decides how much weight it can carry.
- An industry label is a judgement call encoded as a field, and two providers classifying the same company will often disagree.
- Treat a firmographic filter as a boundary rather than a measurement, because fine distinctions near the edge of a band sit inside the error.
Firmographic Data: The Company Attributes Your Targeting Actually Uses
Firmographic data is the set of descriptive attributes that belong to a company rather than to a person: industry, headcount, revenue, location, ownership structure, company age and legal entity type. It is the organisational analogue of demographics, and it is what almost every business-to-business targeting filter is built out of. When a target audience is described as "software companies, fifty to two hundred employees, United States", all three conditions in that description are firmographic attributes.
The word exists because the unit of purchase in business selling is an organisation, and the person you write to is only the organisation's representative in the conversation. Before there was a term for the account layer, audiences were described by the individual alone, and "VPs of Sales" is a description that spans a fifteen-person startup and a listed manufacturer with four thousand staff. Naming the company attributes separately forced two questions apart that had been answered as one: which companies could plausibly buy this, and which person inside them should hear about it.
Where each attribute comes from, and how much to trust it
Every firmographic field arrives by one of three routes, and the route determines how much weight the field can carry.
- Company profile size bands, industry picklists, self-written descriptions
- Chosen by whoever set the profile up, often years ago
- Bands are coarse by design, so a range hides a wide spread
- Nobody updates a size band when they make redundancies
- Cheap, wide coverage, low precision
- Incorporation records, regulatory filings, licence registers
- Structured, dated and traceable to a document
- Coverage collapses outside jurisdictions with open registries
- Private companies file far less than listed ones
- Slow to change but genuinely authoritative where it exists
- Industry from site copy, size from staff pages, revenue from models
- Available for companies that appear in no register at all
- Quality depends on a model nobody publishes the error rate for
- Reruns can silently change a value you already filtered on
- Fills the gaps, and fills them with estimates
Headcount is worth separating out, because it is the attribute most filters lean on and the one that behaves worst. It is usually stored as a band rather than a number, the band is often self-selected, and the underlying reality moves faster than any refresh schedule. A company that has grown from forty to ninety staff and a company that has shrunk from two hundred to ninety can both sit in the same fifty-to-two-hundred bucket while representing opposite situations. The band is accurate and tells you almost nothing about which of the two you are writing to.
Revenue is worse again for private companies, because in most cases it is modelled rather than reported. A revenue figure attached to a private company in a data product is frequently an estimate derived from headcount and industry, which means filtering on both revenue and headcount can be filtering on the same underlying number twice.
Industry codes, and why two systems never agree
Industry is the attribute that feels most objective and is least so. There are standardised code systems: NAICS is the North American standard, and it is reviewed and revised every five years, in years ending in 2 and 7, through the federal Economic Classification Policy Committee (Census Bureau NAICS implementation timeline). SIC codes are the older lineage still embedded in a great deal of business data. On top of both sit the proprietary industry taxonomies that data vendors and professional networks use, which are not derived from either standard and do not map cleanly onto them.
The practical result is that the same company can carry a statutory code assigned at incorporation and never revisited, a vendor industry label produced by a classifier, and a self-selected profile category picked from a dropdown by a marketing hire. A company that started as an agency and now sells software will often still be registered as an agency. Nothing is wrong in any of the three systems; they were populated at different moments, by different parties, for different purposes.
So an industry label is a judgement call encoded as a field. It looks like a fact because it renders as a short string in a column, and it is treated as one because filters are built from equality tests. It is closer to an opinion with a timestamp.
How a filter set becomes a target list
- Step 1Describe the audience
Industry, size band, geography, and any structural attribute such as independent versus group-owned
- Step 2Translate into filters
Each condition becomes one provider's field with that provider's taxonomy and that provider's definition of the field
- Step 3Pull the company set
The result is every company the provider believes matches, which is not the same as every company that matches
- Step 4Resolve people to companies
Contacts are selected inside the accepted accounts, adding a second layer of coverage and accuracy questions
- Step 5Review the edges
Read the rows at the boundary of each filter, because that is where the taxonomy disagreements are visible
The step most often skipped is the last one. Filters are evaluated by their output size, which is the one property that reveals nothing about whether they selected the right companies. Reading fifty rows at the edge of a size band, or fifty rows from an industry label you are unsure about, surfaces taxonomy problems in minutes that would otherwise surface as a flat reply rate a fortnight later.
How firmographic quality is measured
Two numbers describe the state of a firmographic set, and most teams track neither.
The first is fill rate, which is the share of rows in which a given attribute is populated at all. Fill rate is attribute-specific and varies enormously across a single pull: location and industry are usually near-complete, while ownership structure, founding year and revenue thin out fast, and they thin out unevenly, because the companies missing a revenue estimate are disproportionately the small and privately held ones. A segment defined partly on revenue therefore quietly excludes the population whose revenue is hardest to model, which is a selection effect nobody chose.
The second is agreement rate, which is the share of companies where two independent sources give you the same answer. This one is worth measuring once, on a sample, before a large build. Take a couple of hundred companies you already know, pull the same attribute from two providers, and count how often they match. Use whatever bands your own filters use, and count a match as landing in the same band rather than as exact equality. The number you get is a property of your market and your providers, so borrowing anyone else's figure is pointless, but having your own changes how you write filters: a low agreement rate on headcount is an argument for a wider band and a narrower reliance on it.
Both numbers should be recomputed when you change provider, change segment, or return to a pool that has been sitting unused. Attribute quality is not uniform across a data product; it is a function of which companies you asked about.
Where the textbook definition breaks
The tidy definition says firmographics describe a company. That is exactly what they do, and it is also the problem.
Firmographics describe what a company is. Outbound usually needs to know what a company is doing. Fit and timing are different questions, and only one of them is answered here. Every company in a well-built firmographic segment could plausibly buy the product, and on any given week almost none of them are in a position to act. This is why a firmographic-only list can look perfectly on-target in review and perform like a phone book in the field: the selection criteria contain no information about whether anything has changed at the company recently. The attributes that carry timing live elsewhere, in a job posting, a funding event, a leadership change or a visible technology switch, and a segment built without any of them is a segment built on a permanent condition.
The same company returns different values from different providers. Pull one mid-sized company from three data sources and you can reasonably expect three headcounts and two or three industry labels. None of the providers is lying. They sampled different evidence at different times through different classifiers. What this means for a filter is uncomfortable but simple: a condition of "five hundred to one thousand employees" is a filter on one vendor's estimate of headcount, not on headcount. Move the same audience definition to another provider and the resulting company set will overlap substantially and not completely, and the non-overlap is not an error anybody can fix.
The honest way to hold this is to treat a firmographic filter as a boundary rather than a measurement. It is good at excluding companies that are clearly outside the range, and unreliable at fine distinctions near the edges of a band. Designing a segment that depends on the difference between four hundred and eight hundred staff is designing a segment on top of a rounding error.
What to do with it
Use firmographics to draw the outer boundary of who is eligible, then use something else to decide who is worth writing to this month. That second layer is what separates a target list from a market map, and the attributes that supply it are covered in the neighbouring entries: technographic data holds the stack attributes, and hiring signals hold the timing attributes.
Write the definition of each firmographic condition down before building anything, including which provider's version of the field you mean and what you will do when two sources disagree. Then check the boundary rows by hand. Two afternoons spent reading edge cases in a segment saves the far more expensive discovery that an entire industry label meant something different from what you assumed.
Finally, keep the segment narrow enough that a single message can be true for every company inside it. If the segment spans companies whose situations differ enough that the message has to hedge, the segment is doing the job of two segments and should be split into two campaigns with their own premises.
Related terms and guides
Firmographic conditions are the raw material of an ideal customer profile, which is the document where those conditions get written down and agreed rather than improvised per campaign. The distinction between a filter and a signal, which is the heart of the timing problem above, is worked through in B2B prospecting.
On the data side, the reason a single provider rarely produces a complete company record is covered in waterfall enrichment, and the comparison of the providers themselves is in data enrichment tools for B2B SaaS. For what a badly drawn firmographic boundary does to the economics of a programme, see cost per lead in B2B. For a worked example of the account layer being drawn properly in a market where the standard industry tag is close to useless, see healthcare lead generation.
If you would rather see a segment built and tested against your own market than argue about size bands in a spreadsheet, see what a first campaign looks like.
Frequently asked questions.
Frequently asked questions- What is the difference between firmographic and demographic data?
- Demographics describe a person: age, role, seniority, location. Firmographics describe the organisation that person works for: industry, headcount, revenue, ownership structure, entity type. Business targeting needs both, because the company decides whether a purchase is possible at all and the individual decides whether the conversation happens. Filtering on one and assuming the other is the most common targeting error.
- Why do two data providers report different headcounts for the same company?
- They sampled different evidence at different times through different methods. One may use a self-selected profile band, another a model built on observed staff records, another a filing. None of them is lying and none is authoritative. The practical result is that a headcount filter is a filter on one vendor's estimate, so wide bands survive provider changes and narrow ones do not.
- Are SIC and NAICS codes reliable for targeting?
- They are consistent, which is not the same as current. NAICS is reviewed and revised every five years, in years ending in 2 and 7, so the standard itself moves. More importantly, a company's assigned code is often set once at registration and never revisited, so a business that has changed what it sells will frequently still carry the code for what it used to sell.
- Can firmographic data alone build a good target list?
- It can build the eligible population, which is a necessary first step and rarely a sufficient one. Every company in a well-drawn firmographic segment could plausibly buy, and on any given week almost none of them are positioned to act. Adding a timing attribute, such as a visible technology change or a hiring decision, is what turns an eligible population into a reason to write this month.