The Firmographic Field Reference: Proof, Source, and Check per Field

This reference is for the RevOps or recruiting-ops owner who writes routing, territory, and ICP scoring rules on top of firmographic fields and has to answer for where each value came from. It has one entry per field: industry code, headcount, revenue, HQ, ownership, founded date, funding stage, and status. Each entry covers what the value proves, where an authoritative version lives, what is known about its error band and decay, how it misroutes accounts, and the one check that catches it. Where nobody has published a number with a method behind it, the entry says so and shows you how to measure it on your own records.

Which firmographic fields are safe to drive routing?

A firmographic field is safe to drive routing, territory, or ICP scoring only when its value is filed (attached to a named filing, a period, and a definition) or has been checked against a filed value inside your refresh window. The field name tells you nothing about this. The provenance of the value decides it.

Firmographic data is the set of company-level attributes on an account record: industry, size, revenue, location, ownership, age, funding, and operating status. Every value in those fields comes from one of four kinds of source, and the four are not interchangeable.

Most pages that rank for "how accurate is firmographic data" answer with fill rate, which is the share of records where a field is populated. Fill rate tells you whether a value exists, not whether it is true. A fully populated headcount field built from profile counts can still be wrong exactly where it matters most, on accounts that sit near a band edge.

The accuracy claims that do exist come from data providers, and none of them publishes a method. One says profile-derived counts for companies over 100 employees are "usually within 20%" of actual headcount. Another says only 10-20% of profiles reflect real, up-to-date employment. I could not trace either claim to a study with a stated method. Treat both as unverified and measure your own fields with the audit procedure below.

Industry code (NAICS and SIC)

An industry code proves what someone, usually the company itself and often at formation, once said its primary activity was. For most private companies no government body assigns the code or keeps a registry of it, so there is no authoritative value to look up.

NAICS (North American Industry Classification System) is the shared US, Canadian, and Mexican standard. SIC (Standard Industrial Classification) is the older US system, and it is also the name the UK uses for its own self-declared codes. The NAICS edition in force contains 1,012 six-digit codes.

• Definition: a code for the primary activity of an establishment, meaning a single business location. NAICS "does not offer guidance on organizing enterprises (companies) composed of multiple establishments", so a single code on a multi-site company's record is a judgement call.
• Who assigns it:
  • The US Census Bureau assigns one code to each establishment for its own statistics. Title 13 of the US Code prohibits it from disclosing those codes, so you cannot look them up.
  • The SEC assigns a code to every company that files with it, and the code is visible on EDGAR.
  • In federal contracting, the contracting officer selects the NAICS code for each contract, and contractors list their own codes in SAM.gov.
  • In the UK, the company declares its own SIC code at Companies House and can change it on the annual confirmation statement.
  • For everyone else there is no central register and no official way to change a code.
• Error band: not published. I found no study that measures how often the codes on commercial account records match what the companies actually do.
• Decay: the code itself rarely changes, and that is the problem. The business changes and the code stays put. On top of that, classification revisions re-map codes. The proposed next NAICS revision consolidates 46 industries into 18, and nobody has published what share of commercial account records that will affect.
• How it lies: a code chosen at formation and never updated keeps routing a company that has pivoted as if it were still in its old industry.
• Check: pull the code from SAM.gov, EDGAR, or Companies House where one exists, and compare it with the company's current main revenue line, meaning what it actually sells. If the two disagree, segment on the revenue line.

On NAICS code accuracy for segmentation, the practical rule is that coarse segments tolerate bad codes better than fine ones. A wrong six-digit code often still lands in the right sector. If a territory depends on six-digit precision, it depends on a value nobody has verified.

Headcount

Headcount means four different things depending on where it came from, and the sources disagree because they count different people. A filed figure counts the employees of one legal entity at a stated time. A profile count counts the people who say they work there.

The UK has a filed headcount for almost every company. Companies Act 2006 s.411 requires the notes to the annual accounts to disclose the average number of persons employed in the financial year, calculated as a monthly average under s.382(6). The requirement covers small and micro companies too, and 32.4% of active companies on the register file micro-entity accounts.

The US has no filed headcount for private companies. Public registrants report persons employed under Reg S-K Item 101(c)(2), as of a date they choose. For private firms the nearest public figure is Form 5500 line 5a, which gives plan participants at the start of the plan year. On health plans that number includes COBRA enrollees and retirees, so treat it as a proxy and never as headcount. Filed Form 5500s are public through the Department of Labor's EFAST2 search, except for one-participant plans.

• Definition: record which of the four you hold. A number without its definition cannot be compared with anything.
• Authoritative source: the s.411 note for UK entities, the 10-K for SEC registrants, and nothing for US private companies.
• Error band: only provider claims exist: within 20% for companies over 100 employees, and 30-40% of self-reported company sizes off by more than one tier. Neither has a published method.
• Decay: profile counts are live, but revisions arrive about twice a year and are applied retroactively. Filed figures are fixed to their period and age one year at a time.
• How it lies: profile counts include contractors, interns, and people who left without updating their profile. Subsidiary staff roll up into the parent's count. Self-reported sizes skew high, because companies are far more likely to inflate a count than to underreport it. Industries where few staff keep profiles come out low.
• Check: use the s.411 note, the 10-K, or Form 5500 line 5a, and store the definition alongside the number.

A public CRM community thread shows the problem in both directions. Morgan Stanley was held at 10,000 in the CRM against 85,792 on LinkedIn, and FairPlay Engineering at 200 against 37. Neither side of either pair is a filed figure. When the employee count field is wrong in your CRM, comparing it with another estimate tells you the two disagree. It does not tell you which one is right.

Who a profile count actually includes

In Refolk's index of professional profiles, 243 US profiles hold a Software Engineer title with "contractor" in the headline. In a hand-checked sample of 25, six list Apple (2), IBM (2), Meta (1), or Microsoft (1) as their current employer. Each of those six adds to the headcount of a company that may not employ them.

The reverse happens with offshore staff. In the India sample of 25, drawn from 47 such profiles, Capgemini, TCS, and Coforge appear as employers. Engineers who work on a client's product sit under the outsourcer, so the client's count comes out low. The same person inflates one company's number and is missing from another's.

When a profile count looks wrong, the quickest diagnosis is to look at the people behind it: who lists the account as their employer, and which outsourcer they actually work through. Refolk lets you ask for that set of people in plain English and read each person's stated employer and headline, which is how the contractor samples above were built.

Revenue, HQ, and ownership

These three fields fail the same way: two different things end up stored in one field. A modeled revenue figure sits where a reported one should be, a registered office sits where the operating HQ should be, and a subsidiary inherits its parent's size.

Revenue

• Definition: reported revenue is a figure from a filing, for a stated period, in a stated currency. Anything else is an estimate.
• Authoritative source: the 10-K for public companies, and UK filed accounts where turnover is disclosed. Small UK companies can leave turnover out of their filings.
• Error band: one provider says private-company revenue "is estimated based on employee count, industry multiples, web traffic, technology spend, and other signals", and that these estimates "vary by 30-50% from actual figures."
• Decay: a reported figure is fixed to its period. A modeled figure moves whenever its inputs move.
• How it lies: the modeled estimate is stored in the same field as a reported figure, with nothing to tell them apart. Because the estimates lean on employee count, every headcount error described above flows into revenue as well.
• Check: add a provenance flag. If a value has no filing, period, and currency attached, treat it as modeled.

On estimated revenue vs reported revenue for a private company, a UK filing can still bound the number when turnover is omitted. A company that qualifies as small meets at least two of three tests: turnover up to £15m, balance sheet up to £7.5m, and no more than 50 employees on average. Those thresholds are revised from time to time, so confirm the current values in the Companies House filing guidance before writing them into a rule.

HQ

• Definition: the place where the company operates and where its leadership and staff sit. It is not the registered address and not the billing address.
• Authoritative source: no filing records an "HQ". The UK registered office is filed, but it "does not have to be the same as your business address, where your company operates or trades."
• Error band and decay: not published.
• How it lies: a registered office, a formation-agent address, or a billing address gets taken for the operating HQ, and the account lands in the wrong territory.
• Check: compare the Companies House registered office with the trading address and with where the company's staff are located.

Ownership and parent/subsidiary

• Definition: who owns the legal entity on the record, and whether that entity is a parent, a subsidiary, or standalone.
• How it lies: profile data typically rolls all subsidiary employees into the parent's headcount, and a subsidiary record that inherits the parent's figure gets upgraded to enterprise and routed to the wrong team.
• Check: the s.411 figure is reported per legal entity, so use the subsidiary's own filing for the subsidiary's size. Deciding which entity owns the buying decision is a separate question, covered in The Subsidiary Ownership Verdict.

Founded date, funding stage, and status

These three fields have the thinnest evidence: I found no published error band for any of them. For founded date, I found no source that settles whether the founded year or the incorporation date should count as authoritative. Store the source with every value, and treat any value without a source as unverified.

• Founded date: incorporation date is a registry fact. A founded year is the company's own claim. The two can differ, and nothing I found rules on which one wins. Keep them in separate fields, each with its source. If a rule depends on company age, have it read the field with the source you can check.
• Funding stage: this usually comes from public Crunchbase records or company announcements, and it is third-party or self-reported data. A stage without a round date cannot tell you whether it is still current. Store the round date and a link to the announcement or filing, and treat any stage older than your refresh window as unverified.
• Status: active, acquired, renamed, or dissolved. A dead or absorbed account that is still routing wastes a rep's time on every cycle. Check the registry record before status gates anything, and run The Quarterly Account Status Sweep so the field doesn't drift between audits.

How firmographic fields go wrong

Most misroutes trace back to one of eight failure modes. Nearly all of them share a root cause: a value from a weak source stored in the same field, with the same weight, as a filed one.

1. Industry code frozen at formation. A self-chosen code that was never updated keeps routing a company that has changed what it does as its old industry. Check: compare the SAM.gov, EDGAR, or Companies House code with the current revenue line, and re-map after each NAICS revision, since the proposed one collapses 46 industries into 18.
2. Headcount from profile counts. Contractors list the client brand (6 of 25 US contractor profiles in Refolk's index), while offshore staff sit under outsourcers such as TCS and Capgemini, which makes the client's count low. Check: use the s.411 note, the 10-K, or Form 5500 line 5a, and record the definition.
3. Inflated headcount bands. Self-reported ranges skew high, as in the FairPlay Engineering case (37 against 200). Check: never route an account sitting on a band edge without a filed figure.
4. Modeled revenue stored as reported. Check: a provenance flag. With no filing, period, and currency attached, the value is modeled.
5. Registered office taken for HQ. A formation-agent address puts the account in the wrong territory. Check: compare the registered office with the trading address and with where staff are located.
6. Subsidiary inherits parent size. The subsidiary gets upgraded to enterprise. Check: use the per-entity s.411 figure.
7. Keyword matching that counts the wrong thing. "Contractor" matched employer names in 20% of the India sample. Check: hand-classify a sample before trusting any count derived from a keyword.
8. Contact decay applied to firmographics. Check: cite the firmographic figure of roughly 20-30% obsolete per year, which is attributed second-hand to a large business-data provider. The widely quoted 22.5% per year and 2.1% per month are the same single finding, they are not independent, and they describe contact data.

Failure mode 7 deserves its own table, because keyword-derived counts feed many ICP models. The same keyword misfires at different rates in different markets.

The India misses are concrete: 4 of the 25 profiles list Elec-Tech Electrical Contractors and 1 lists CSW Contractors. The keyword matched the employer's name, not the person's role. That is why every count from Refolk's index in this guide comes with a hand-classified sample, and why I would not let any keyword count into a scoring model without one.

A field checked against another estimate has been compared, not verified.

How to audit firmographic fields before they gate rules

Audit only the fields your rules actually read. Check each one against its primary source on a stratified random sample, then report accuracy per field with a confidence interval. Fill rate and agreement between vendors don't count as accuracy.

In step 2, one workforce-data provider's advice is worth adopting word for word: compare like with like, using "the same data source, the same employee definition (official headcount vs. estimated headcount), and consistent reporting intervals." A filed monthly average and a live profile count will disagree even when both are honest.

In step 4, keep "unverifiable" as its own result. Don't count it as correct. For a US private company's headcount, unverifiable is often the true answer, and that share tells you how much of a rule is resting on values nobody can check.

How big the sample needs to be

Sample size depends on the smallest error rate you need to detect. The figures below follow from binomial math, n = ln(0.05) / ln(1 - p), and they apply per field and per stratum.

Budget the time honestly. The practitioner estimate is 5-10 minutes per account per field, and that figure is unverified. At that rate, 100 accounts across all eight fields comes to roughly 65 to 135 analyst hours. Step 1 exists to cut that down: audit only the fields a rule reads.

Refresh cadence and gating

Set refresh cadence for each field according to how fast it changes and whether it gates a rule, then change the rules so they cannot read an unchecked value. Published guidance disagrees on cadence, and no per-field window is established publicly.

What the guidance says:

• One provider says firmographics change more slowly and should be refreshed quarterly.
• Another says to re-check moderately stable firmographics periodically and to re-verify volatile fields at the point of use.
• A third recommends refreshing the lead pool before each routing cycle.

I would combine all three, using the two questions in the matrix below.

Gating only works if the provenance travels with the value. Add these fields beside every firmographic value that a rule reads.

<field>_value:
<field>_provenance: filed | registry_self_declared | profile_derived | modeled
<field>_source: e.g. UK s.411 note, 10-K Item 101(c), Form 5500 line 5a, Companies House, profile count
<field>_definition: e.g. average persons employed, monthly average over financial year
<field>_period: period end or point in time the value describes
<field>_checked_on: date of last primary-source check
<field>_check_result: correct | incorrect | unverifiable

Then write the gate into the rule itself.

IF headcount_provenance = filed
   AND headcount_period is within REFRESH_WINDOW
THEN route on headcount_value
ELSE IF headcount_check_result = correct
   AND headcount_checked_on is within REFRESH_WINDOW
   AND headcount_value is not within BAND_EDGE_MARGIN of a band boundary
THEN route on headcount_value
ELSE send to verification queue before routing

Keeping this reference current

Three things move, and each one should trigger a re-check:

• NAICS revisions. Read the OMB notice on each revision to find the reference period it applies from, then re-map every industry code that feeds a rule. The proposed revision merges 46 industries into 18.
• UK size thresholds. Re-read the Companies House filing guidance whenever a rule depends on the small-company tests.
• Your own error rates. Re-run the audit on the same strata at your chosen cadence, and treat any field whose error rate has risen beyond its margin as profile-derived until it is checked again.