Statistics of U.S. Businesses (SUSB) Data

It is the Census Bureau's annual series counting US employer businesses by enterprise size, assembled from the Business Register - the same administrative backbone that maintains the universe of American firms - rather than from a survey form asking anyone to remember anything.

Every row crosses three axes: an establishment industry (NAICS up to six digits), an employment-size class of the owning enterprise, and a geography running from the national total through states, metropolitan statistical areas, counties and congressional districts. The measures are firm counts, establishment counts, employment for the week of March 12, annual payroll in thousands of dollars, and receipts in years ending in 2 and 7 when the Economic Census supplies them.

The scale is national-account sized: the 2022 file covers roughly 6.4 million firms and 8.3 million establishments across all industries. The distinctive move is the enterprise lens - a janitorial chain's hundred branches collapse into one firm in a large size class, while its independently owned competitors fill the small classes. For Diversified Support Services that contrast is stark in a single industry line: NAICS 5617 Services to Buildings and Dwellings holds 215,428 firms employing about 2.1 million people in 2022, with 149,120 of those firms below five employees.

One row per industry x size-class x geography cell, exactly as a delivery lands:

The four rows above tell the whole structural story of one support-services industry in 2022. Of 215,428 firms in NAICS 5617, 149,120 - 69 percent - employ fewer than five people, yet that vast cohort runs on $9.7bn of payroll and holds $39.3bn of receipts, about 17 percent of the industry total. Add the next two bands and nine in ten firms sit below twenty employees. Meanwhile the ten-to-nineteen class already out-employs the five-to-nine class, 248,001 jobs against 212,142, from barely half as many firms: concentration starts early even in a famously fragmented trade.

Notice also that firms nearly equal establishments at the total (215,428 against 224,960) but converge almost exactly in the micro class - the multi-site chains live in the upper rows.

Fourteen verified fields form the spine of a row, in three groups:

Identity. STATE pins the geography (00 for the US total, otherwise a two-digit FIPS code), NAICS carries the industry up to six digits ('--' marks the all-industry line), and ENTRSIZE holds the enterprise employment-size class from 01 (total) and 02 (under five employees) through coded classes up to 100,000+. Three description columns - STATEDSCR, NAICSDSCR, ENTRSIZEDSCR - keep every extract readable without a codebook.

Measures. FIRM and ESTB count enterprises and establishments. EMPL carries March-week employment, PAYR annual payroll in thousands of dollars, and RCPT receipts - populated only in Economic Census years ending in 2 and 7.

Precision gauges. EMPLFL_N, PAYRFL_N and RCPTFL_N grade injected noise: G for under 2 percent, H for 2 to under 5, J for 5 percent or more. Values marked D or S are withheld and set to zero. A gap in this data is a disclosure decision, never a zero - treat the flags as part of the measurement, not a defect in it.

The headline files cover the US and states at six-digit NAICS. The rest of the program extends the identical grid downward and outward, and ships with a sample on request: county tables from 2000, congressional-district tables from 2015, metro-area tables from 1997 at three-digit NAICS, island-area tables from 2020, the 2008-2016 employment-change series tracking firm births, deaths and growth among survivors, and the receipt-size variants that re-key the size axis to enterprise receipts in Economic Census years.

Most questions need only one or two of these - a county fragmentation study wants the county files, not the change series - so they sit behind an explicit request note rather than bloating every delivery.

Geographically this is the deepest record on the Diversified Support Services shelf: national, state, metro, county and congressional-district cuts of the same size grid, with territories joining in 2020. Where a product like the Quarterly Services Survey stops at the national line, SUSB goes all the way down to a single district.

Temporally annual tables cover 1992 and every reference year from 1997 through 2022, with the geographic tiers arriving progressively (metros 1997, counties 2000, districts 2015, territories 2020). A separate employment-change series adds dynamics for 2008-2016.

Granularity is the enterprise: establishment industry up to six NAICS digits crossed with the owning enterprise's employment-size class. That is the axis establishment-based products lack - the QCEW NAICS 56 tabulation counts every worksite monthly, while SUSB tells you whether those worksites belong to empires or owner-operators. The two together answer both halves of any support-services market question. The UK mirror image - BRES - plays the same role across the Atlantic.

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

Your cadence runs independently of the annual reference-year rhythm. Take a one-time pull of the full history - 1997 through 2022 plus the 1992 baseline, every geography tier you name - or land a warehouse table that absorbs each new vintage cleanly against the last.

Whatever arrives is normalized to the field dictionary above: NAICS codes typed as codes, size classes decoded from ENTRSIZE into readable bands, noise flags preserved alongside the values they qualify. Joins to employment, wage and revenue records hold on the same NAICS x geography keys, so this record drops straight into a support-services panel assembled from our shelf. Name the industries, geographies and years when you request the sample - it ships shaped to exactly that cut.

• Fragmentation and concentration analysis - read the size distribution of any support-services industry straight from the enterprise axis: 69 percent of building-services firms below five employees, consolidation beginning by the ten-to-nineteen band.
• TAM and market-sizing models - anchor revenue and payroll on measured bases ($230.1bn receipts, $81.9bn payroll for NAICS 5617 in 2022) with federal methodology behind them.
• Roll-up and franchise screening - count the addressable micro-firm base before modelling consolidation, then track how much of it the upper size classes already absorb.
• Territory design - weigh sales coverage by county or district using the same size grid repeated at every geography tier.
• Long-run feature pipelines - train on an annual 1997-2022 panel with stable codes and no re-basing mid-series.
• Economic journalism and coursework - cite the standard federal measure of US business size structure.

Market Researchers & Consultants and Investors & Quants tie at 3/3. The first anchor engagement models on measured size distributions; the second price consolidation theses on the gap between 149,120 micro-firms and the chains in the upper classes. Data Scientists & ML Engineers (2/3) get a long annual panel on typed keys built for feature work; Sales & Growth Teams (2/3) get territory-weighting inputs, minus any named-account rows; Journalists, Academics & Students (2/3) get a citable federal standard. Competitive Intel Product Teams (1/3) draw context only - the same disclosure protection that makes the figures trustworthy guarantees no competitor ever appears by name.

Four things worth settling upfront.

First, the lens is the enterprise, not the site. A national chain contributes one firm in a large size class regardless of how many states its branches span. If your question is about local labor markets, pair this record with the QCEW NAICS 56 tabulation; if it is about market structure, this is the right tool.

Second, flags and gaps carry meaning. G, H and J grade injected noise; D and S mean a value was withheld and zeroed. A vanished cell is a disclosure decision - read it as signal, and keep the flag columns beside the numbers they qualify in any downstream model.

Third, receipts follow the Economic Calendar. The RCPT column populates in years ending in 2 and 7 only, so a receipts time series steps every fifth year while payroll and employment run annually.

Fourth, the identity is typed end to end. Geography, NAICS and size-class codes join cleanly to the rest of our Diversified Support Services shelf - and to the SUSB-vs-BRES comparison if your market crosses the Atlantic. Specify industries, geographies, size bands and years, and the sample arrives pre-cut to them.