US Census County Business Patterns - Insurance Agencies & Brokerages (NAICS 524210)

It is the federal government's count of who sells insurance through an agency or brokerage, geography by geography, year after year. County Business Patterns is the US Census Bureau's annual series of subnational business establishment statistics, and filtering its six-digit NAICS code 524210 isolates Insurance Agencies and Brokerages: every yearly release carries establishment counts, employment during the week of March 12, first-quarter payroll and annual payroll, tabulated by geography and split across nine employment-size classes. In the 2023 county file alone, 524210 rows appear for 2,729 individual counties, each carrying the full measure set.

Two properties make it the industry's measuring stick. First, the definition holds still - one NAICS code, one record layout, one annual grain, reaching back through reference years 1986 through 2023, so a 1998 county and a 2023 county describe themselves in the same language. Second, it counts but never names: no agency appears by identity anywhere in the files, which is exactly why every carrier, consultant and aggregator can treat the numbers as neutral ground. In Datadory's catalog of 1,744 datasets across 159 viable industries, this record scores 10/10 for quality - the anchor of the ten-record insurance brokers slice.

Get a sample of this dataset

One row per geography per reference year, straight from the 524210 slice of the 2023 county file:

# one row per county x NAICS 524210 - reference year 2023
fipstate : 01        Alabama
fipscty  : 001       Autauga County
naics    : 524210    Insurance Agencies & Brokerages
est      : 12        establishments
emp      : 39        mid-March employees      nf G (<2% noise)
qp1      : 405       first-quarter payroll $k nf H (2-<5% noise)
ap       : 1920      annual payroll $k        nf G (<2% noise)
n<5      : 9         n5_9   : 3               size-class counts

# same county, different scale - Baldwin County
fipscty  : 003       est 88   emp 462   qp1 7645   ap 32228
n<5      : 58        n5_9   : 18   n10_19 : 10

# small counties lean on the flags instead of hiding
fipscty  : 005       emp 32   nf J (>=5% noise)   est 10

Three things worth noticing. First, the measures come in threes - employment, first-quarter payroll, annual payroll - and each travels with a noise flag (nf) quantifying how much disclosure control perturbed it: G means under 2% noise, H means 2 to under 5%, J means 5% or more. Autauga County's 12 agencies employed 39 people for roughly $1.92 million in annual payroll; Baldwin County's 88 agencies employed 462 for about $32.2 million. Second, the size classes turn a single establishment count into a market-structure reading: 58 of Baldwin County's 88 agencies employ fewer than five people, a textbook independent-agency footprint. Third, small counties do not disappear when numbers run thin - the third row shows a 10-establishment county whose employment carries the high-noise J flag instead of being silently dropped.

Twenty-three columns in the county-layout record, all verified against the published record layout rather than inferred:

• Geography keys - fipstate and fipscty locate every row to a specific state and county.
• Industry key - naics, fixed at 524210 in this slice, with ------ reserved for all-industry totals.
• Measures - emp (mid-March employees), qp1 (first-quarter payroll, $1,000) and ap (annual payroll, $1,000).
• Disclosure flags - emp_nf, qp1_nf and ap_nf carrying the G/H/J noise scale, plus N for withheld cells elsewhere in the family.
• Structure counts - est plus the nine n* size classes from under-5 employees through 1,000-or-more, with fine splits extending the top class.

Everything outside the county dictionary folds under additional fields on request: legal-form-of-organization splits from recent vintages, the columns unique to the discontinued 2012-2022 combined CBP/NES report, ZIP Code industry detail, and the older record layouts used before 2020. Name what your workflow needs when you request the sample and it arrives resolved against the exact files.

Geography runs seven levels deep: national totals, states, counties, metropolitan statistical areas, combined statistical areas, congressional districts and ZIP Codes - the last as both totals and industry detail - plus Puerto Rico and the Island Areas. One industry, one consistent measure set at every level, which is what lets a national trend be traced down to a single county without changing sources.

Temporal spans reference years 1986 through 2023 in the current archive. Geographies joined the family at different points - MSAs in 1993, ZIP Codes in 1994, Puerto Rico in 1998, the Island Areas in 2008, congressional districts in 2013, CSAs in 2017 - so the earliest usable year depends on the level you cut at. Newer reference years enter the archive on the Census Bureau's own schedule; deliveries reach you at the cadence you set.

Granularity is one annual observation per geography per six-digit NAICS industry, with establishments distributed across nine employee-size classes. There is no quarterly view and no firm-level microdata - if you need named producers to pair with these denominators, the NIPR Producer Database covers that layer.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the channel and set the cadence - a one-time historical pull of the full 1986-2023 panel lands as cleanly as a recurring feed of the newest reference year. Every delivery ships the record layout above attached, with noise flags preserved rather than smoothed over, so aggregation logic in your warehouse can weight low-noise and high-noise counties differently. Say which geographies and years matter when you get a sample of this dataset; the sample goes out first and the schema in the sample is the schema you ship against.

Anywhere the question is how big is the market here, this is the reference measurement:

• Territory design - rank counties by agency count and payroll so sales capacity follows the actual distribution footprint; the bread-and-butter move for sales growth teams.
• Saturation and density models - divide establishment and employment series over population or premium volume to find counties where distribution is thin relative to demand.
• Expansion screening - count incumbents before signing a lease; competitive intelligence teams use it to gauge local crowding precisely because no competitor is named.
• Quant inputs - four decades of annual observations give investors and quants a stable exogenous feature for sector and carrier models (quant backtesting).
• Research and journalism - local insurance-employment trends cited from a continuous federal series, the standard move for journalists and academics (citation-grade research).
• Product plumbing - builders stand dashboards on a record layout that repeats identically year after year.

Three things worth knowing upfront.

First, this is the denominator layer. It sizes the market; it does not identify players. Pair it with a producer register when names matter - NIPR Producer Database for US producers keyed on lifetime NPNs, or the EIOPA Insurance Statistics family when the question moves to the EU and EEA - and the head-to-head comparison of the two shows exactly where each stops.

Second, small cells are protected, not absent. Disclosure control perturbs or withholds values in thin markets - that is what the G/H/J/N flags encode - so single-digit-count counties deserve flag-aware handling rather than blind averaging. The flags ship with the numbers precisely so you can.

Third, the grain is annual and the identity is industrial. Rows describe NAICS 524210 establishments in a reference year, not policies sold or premiums written - headcount and payroll only. Within that frame the series is unusually clean: one layout, one code, four decades. Start from the insurance brokers data hub or the ranked rundown of the best insurance brokers datasets to see how it sits against the rest of the slice.