[Home](https://stoneforgegroup.com/) · [Research](https://stoneforgegroup.com/research/) · Firm Size and Firm Age 03 – Research # Firm size and firm age In professional services, the share of firms still under 20 employees falls from 79.2% at age 5 to 70.8% at age 26 or more. Mean firm size rises from 19.2 to 25.0 employees – about six more people after twenty-one more years. ## Executive summary This study measures firm size against firm age, holding the population fixed at firms with five or more employees throughout, across eleven age bands in five sectors, using the Census Bureau's 2023 Business Dynamics Statistics. The question is how much larger a firm's size tends to be at age twenty-six or more than at age five, once the comparison excludes the smallest firms that would otherwise distort it. In professional services, mean firm size rises from 19.2 employees at age 5 to 25.0 at age 26 or more, about 30% more, and the share still under 20 employees falls from 79.2% to 70.8% over the same span. The other four sectors each move by more, published in full below. Widen the population to every firm, including the smallest, and professional services appears to rise from about 2.9 employees at age 0 to about 24.9 at age 26 or more, roughly 8.6 times as many – a figure this study also reports and then accounts for. That 8.6-times figure is almost entirely a composition effect, not a growth curve. Younger age bands hold a much larger share of one-to-four-employee firms than older age bands do, so the youngest cohorts' average is dragged down by a population the oldest cohorts barely contain, without any individual firm needing to add a single employee. Hold the population constant at five or more employees, the comparison this study makes throughout, and the same sector's rise is about 30%, not 8.6 times. This is a snapshot of different firms at different ages, observed in the same year, not one firm followed as it gets older, so it cannot show any individual firm's growth path. It does not show that firm size is capped, only that – among firms already past five employees – the typical difference between a young firm and an old one is modest; survivors are not a random sample, and current size is not peak size. Read Why the average firm size appears to rise much faster than this, below, for the full account of the composition effect. - Professional, scientific and technical services - Construction - Real estate - Administrative and support services - Finance and insurance Mean employees per firm by firm age, among firms with five or more employees. Sector names are labeled directly on the chart, not by color alone. US Census Bureau, Business Dynamics Statistics, 2023. Administrative and support's line does not move smoothly through every age band – the underlying counts are in the published CSV, linked below. ## In professional services, firms aged 26 or more average 25.0 employees and 70.8% are still under 20, against 19.2 employees and 79.2% among firms aged 5. This study's population is firms with five or more employees, the same population the companion [leverage study](https://stoneforgegroup.com/research/the-leverage-gap/) uses, and its baseline is age 5, not age 0. At age 0, professional services firms with five or more employees average 32.2 employees – higher than any later age, including age 26 or more – because the population at age 0 includes firms that enter the count large: spinoffs, restructurings and other paths into existence that start a firm with staff already in place, not firms that grew there. Age 5 is the first band where a difference in mean firm size can be read as a comparison between cohorts rather than as an artifact of how firms are born. Among professional services firms with five or more employees, the ones aged 5 average 19.2 employees; the ones aged 26 or more average 25.0 – about six more people, across the twenty-one years this study's age bands span. The share still below the twenty-employee line moves the same way, from 79.2% to 70.8%. Both measures move together, and neither moves by very much. The other four sectors show more separation than professional services does, and construction moves the most of the five. Administrative and support starts and ends the highest of the five, though – as the chart above shows – it does not move smoothly between the two. Mean employees per firm, age 5 versus age 26 or more, among firms with five or more employees, 2023 Sector Age 5 Age 26+ Change Construction 15.2 26.0 71% more Real estate 17.4 23.8 37% more Finance & insurance 16.0 25.7 61% more Professional services 19.2 25.0 30% more Admin & support 26.8 33.1 24% more ## Why the average firm size appears to rise much faster than this Every figure above holds the population fixed at firms with five or more employees. Widen the population to every firm, including the one-to-four-employee band, and the picture looks completely different. In professional services, mean firm size across all firms rises from about 2.9 employees at age 0 to about 24.9 at age 26 or more – about 8.6 times as many. Read on its own, that looks like decisive evidence against any plateau in firm size. It is almost entirely a composition effect. In this dataset, the older age bands hold a smaller share of one-to-four-employee firms than the younger age bands do – new firms are heavily concentrated in that smallest band, so the youngest age bands are full of them, and the oldest age bands have fewer. That shift in composition alone lifts an age band's average, without requiring any firm counted in it to carry a single additional employee. Hold the population constant at five or more employees instead, the same population this study uses throughout, and professional services' mean firm size is 19.2 employees at age 5 and 25.0 at age 26 or more – about 30% higher, not 8.6 times as many. The table below gives the share still under 20 employees for every sector and firm-age band this study covers, so a reader who wants a different threshold than twenty employees can read it off directly. Share of firms still under 20 employees, by sector and firm age, 2023 Firm age Construction Real Estate Professional services Admin & support Finance & insurance 0 88.7% 85.0% 83.0% 79.2% 81.7% 1 88.5% 84.2% 84.3% 76.7% 87.9% 2 88.2% 84.9% 83.0% 77.5% 88.5% 3 86.6% 82.1% 81.7% 76.1% 85.4% 4 85.2% 82.0% 81.0% 74.5% 85.0% 5 84.0% 81.6% 79.2% 71.8% 84.1% 6–10 81.1% 80.4% 77.1% 70.1% 84.5% 11–15 75.8% 77.3% 74.1% 67.4% 82.3% 16–20 74.5% 78.8% 74.4% 67.1% 80.3% 21–25 73.0% 74.4% 71.7% 64.6% 76.9% 26+66.2% 71.8% 70.8% 61.3% 74.2% ## What this means for an owner This part is interpretation, not measurement. A modest rise in mean firm size is consistent with several different stories, and the data here cannot choose between them. If the firms counted in the older age bands are disproportionately firms that already carried a little more staff at a young age, the difference would reflect which firms are left in each band's count, not growth by any individual firm. For that reading to hold, exit from the five-or-more population – through failure, acquisition, or shrinking back under five employees – would need to be more common among the firms with the fewest employees in their band. It is also consistent with most professional services firms genuinely carrying a little more headcount in the older age bands than in the younger ones – not a large difference, but a real one. For that reading to hold, individual firms would need to show a similar small difference if the same firms were followed over time instead of compared across a single year's cross-section, which this study does not do. And it is consistent with a plainer reading: firm age moving mean size by about 30%, while the share under 20 employees stays a majority at every age band this study covers, suggests firm size is not simply a function of time in business. Most professional services firms, at any age from 5 to 26 or more years, sit closer to twenty employees than to fifty. None of these readings is asserted here; distinguishing them needs firm-level data followed over time, which this dataset does not carry. ## What this does not show **This is a snapshot of different firms at different ages, observed in the same year – not one firm followed as it gets older.** It cannot show one firm's size changing as that firm ages; it can only show what firms of each age looked like in 2023. **Current size is not peak size.** A firm that reached forty employees and fell back to twelve is counted among the smaller firms at whatever age it is today – so these figures show which firms are not past twenty employees, never that they never got past it. **Survivors are not a random sample.** Firms that grow substantially are more likely to be acquired, which removes them from the independent-firm counts this study draws on. The older cohorts here may be biased toward firms that did not grow much, simply because the ones that did are more likely to have already left the count by acquisition. **Records begin in 1978.** A firm older than the underlying data has no true recorded birth year, so the 26-or-more age band is only reliably populated with firms whose age is measured, rather than assumed, from about 2004 onward – the first year a firm could have a real, counted age of 26 or more within the series. **Twenty employees is a chosen threshold, not a natural one.** It roughly matches where professional services revenue crosses $3M: firms in the 10-to-19-employee band average about $2.7M in revenue, and firms in the 20-to-49 band average about $6.3M, both from the companion [leverage study's](https://stoneforgegroup.com/research/the-leverage-gap/) dataset. A reader who wants a different line can read the full distribution – the share of firms in the 20-to-99 and 100-or-more bands, alongside the share still under 20 – in the [published CSV](https://stoneforgegroup.com/assets/data/plateau-by-sector-age-2023.csv) or the table above. **This study cannot rule out subsector composition as the real driver of professional services' rise.** NAICS 54 bundles legal services, accounting, architecture and engineering, computer systems design, management consulting and advertising into one code. If firms surviving to age 26 or more are disproportionately drawn from a subsector that runs larger at any age, the rise this study reports could be a shift in mix rather than firms actually growing with age – and this dataset cannot tell the two apart. Confirmed live against the Census API, the Business Dynamics Statistics time series crosses firm age with employment-size band at the 2-digit NAICS sector level (135 real rows for NAICS 54), but returns none at all for NAICS 541 or NAICS 5415 – every row at that finer level carries age or size, never both (see Subsector composition, in Methodology and sources, below, for the exact counts). A published cross-tabulation of subsector by age by size band would settle this; none exists. This study also carries no sampling error to check the way the companion leverage study's Annual Business Survey figures do – Business Dynamics Statistics is built from a comprehensive administrative database, not a sample (see Data source, below). Its own source of imprecision, disclosure-avoidance noise, works differently: Census does not publish a per-cell magnitude for it the way it publishes a relative standard error for ABS, so this study cannot state whether its headline changes survive it by a stated margin the way the leverage study's do. What can be said is that the published cells are large – the smallest is 526 firms, most run into the thousands or tens of thousands – and noise infusion is designed to protect small cells without materially disturbing totals of that size, but that is a description of the method's intent, not a number this study can check its own figures against. ## Methodology and sources This study is meant to survive an adversarial read. Every figure below is reproducible from the steps stated here, and every judgement call is disclosed rather than left for a reviewer to guess at. ### Data source US Census Bureau, Business Dynamics Statistics (BDS), 2023 data year, time series endpoint at https://api.census.gov/data/timeseries/bds. The universe is employer firms only – the same restriction the companion leverage study uses. BDS is built from the Longitudinal Business Database (LBD), a comprehensive database of US business establishments and firms, not a survey sample. The Census Bureau describes County Business Patterns and BDS together as covering "the practical totality of the private non-agricultural sector." Unlike the Annual Business Survey the leverage study draws on, these figures carry no sampling error for that reason. Census does apply disclosure-avoidance noise infusion to BDS statistics before publication, to protect the confidentiality of the underlying LBD microdata – a different kind of imprecision from sampling error, one that affects small cells more than large ones, and one Census does not publish a per-cell margin for. See Known limitations. ### The exact query The literal call, with the key redacted: https://api.census.gov/data/timeseries/bds?get=FIRM,EMP,FAGE,EMPSZFI&YEAR=2023&NAICS=54&for=us:*&key=REDACTED Run once per sector, changing only NAICS, for five values: 23 (Construction), 52 (Finance and insurance), 53 (Real estate and rental and leasing), 54 (Professional, scientific and technical services), 56 (Administrative and support and waste management and remediation services). Finance and insurance is included here, unlike the leverage study, because BDS counts firms and employees, not revenue – the gross-flow problem that excludes NAICS 52 from the leverage study's revenue ratios has no equivalent in a firm count or an employee count. Geography is for=us:* – national totals only; see Known limitations. The script issuing these calls is scripts/research_pull.py, function plateau_rows(). ### Variable definitions - **FIRM** – Census name "Number of firms." A simple count of the number of firms in the cell (sector by firm age by size band). - **EMP** – Census name "Number of employees." Paid employment: full- and part-time employees, including salaried officers and executives of corporations, on the payroll in the pay period including March 12. - **FAGE** – Census name "Firm age." An employment-based measure of firm age. A firm born before 1976 has no true recorded birth year and is grouped as "Left Censored" rather than assigned an age. - **EMPSZFI** – Census name "Employment size of firms code." Which employment-size band a row describes – the same kind of variable the leverage study reads, from a different Census dataset. See the code-space warning, below. ### Firm-age codes used, and the ones excluded Eleven FAGE codes are used here, confirmed live against the Census API on 7 August 2026: - 010 – 0 years - 020 – 1 year - 030 – 2 years - 040 – 3 years - 050 – 4 years - 060 – 5 years - 070 – 6 to 10 years - 080 – 11 to 15 years - 090 – 16 to 20 years - 100 – 21 to 25 years - 110 – 26 or more years Three more FAGE codes exist in this dataset and are deliberately not used: 001 ("Total," an all-ages aggregate that would double-count every firm if summed alongside the eleven bands above), 065 ("1 to 5 years," an aggregate overlapping the six single-year bands already counted), and 150 ("Left Censored," firms whose true birth year predates the 1976 start of the underlying data, so no real age is known for them). **FAGE=075 ("11+ years") is a trap.** Queried on its own it returns a real firm total – 307,505 firms, professional services, all sizes combined – but every one of the size-band component rows this study needs (EMPSZFI 620, 630, 640, 649, 657) is simply absent from that response, not zero, not suppressed with a flag, just missing. A script that does not check for this treats the missing rows as zero and silently reports the entire 11-or-more population as having no employees in any size band. This study never uses FAGE=075; the eleven codes above are used instead, none of which show this behavior. ### Size-band codes: the five that are exclusive, and the two aggregates excluded Five EMPSZFI codes are exclusive, non-overlapping size bands, summed to build this study's firms_5plus and employees_5plus columns: 620 (5 to 9 employees), 630 (10 to 19), 640 (20 to 99), 649 (100 to 499), 657 (500 or more). Two more codes exist in the same dataset and are deliberately excluded, because they are aggregates that overlap the five above: 635 ("1 to 19 employees," spanning the excluded 1-to-4 band plus the included 5-to-9 and 10-to-19 bands) and 650 ("20 to 499 employees," spanning the included 20-to-99 and 100-to-499 bands as one). Summing either alongside the five exclusive codes would double-count every firm in the overlap. A sixth code, 612 (1 to 4 employees), also exists and is deliberately excluded for a different reason: this study's population, like the leverage study's, is firms with five or more employees. **EMPSZFI exists in a second Census dataset with a different code space.** This study reads EMPSZFI from the Business Dynamics Statistics (BDS) time series. The companion leverage study reads EMPSZFI from the Annual Business Survey (abscs) instead, and its own methodology appendix documents this warning from the other direction. BDS code 640 means "Firms with 20 to 99 employees," where abscs splits the same range into 641 (20 to 49) and 642 (50 to 99). A code carried from one dataset's query into the other returns no error – it simply matches nothing in that dataset, and the affected band silently drops out of the total rather than the query failing loudly. ### How each figure is computed Two figures, computed per NAICS sector and firm-age band, from the same five exclusive EMPSZFI codes: - firms_5plus = FIRM summed across EMPSZFI 620, 630, 640, 649, 657. - employees_5plus = EMP summed across the same five codes. - mean_size_5plus = employees_5plus ÷ firms_5plus, rounded to one decimal place. - pct_under_20 = (firms in the 5-to-9 band + firms in the 10-to-19 band) ÷ firms_5plus × 100, rounded to one decimal place – the share of the 5-or-more population that has not reached twenty employees. Worked example, professional services, age 26 or more: employees_5plus is 1,089,427; firms_5plus is 43,577; 1,089,427 ÷ 43,577 = 25.00, rounded to 25.0, matching the published mean_size_5plus. The same cell's pct_under_20 is 70.8% – computed from the same 43,577-firm denominator, with the firms in the 5-to-9 and 10-to-19 bands as the numerator rather than total employment. ### The all-firms comparison, verified separately The 8.6-times figure in Why the average firm size appears to rise much faster than this, above, is not in the published CSV – it uses EMPSZFI 001 ("All firms," every size band combined, including the 1-to-4 band this study otherwise excludes), so it sits outside plateau_rows() entirely. Verified live against the Census API on 7 August 2026, professional services, all firm sizes: https://api.census.gov/data/timeseries/bds?get=FIRM,EMP,FAGE,EMPSZFI&YEAR=2023&NAICS=54&EMPSZFI=001&for=us:*&key=REDACTED FAGE 010 (age 0): FIRM 72,341, EMP 211,804, mean 211,804 ÷ 72,341 = 2.93, rounded to 2.9. FAGE 110 (26 or more): FIRM 99,594, EMP 2,475,244, mean 2,475,244 ÷ 99,594 = 24.85, rounded to 24.9. 24.9 ÷ 2.9 = 8.586, about 8.6 times. ### Subsector composition: not testable in this dataset NAICS 54 – professional, scientific and technical services – bundles legal services, accounting, architecture and engineering, computer systems design, management consulting and advertising, along with several smaller subsectors, into one code. If the firms surviving to age 26 or more are disproportionately drawn from a subsector that runs larger at any age, the 30% rise this study reports could be a shift in subsector mix rather than firms actually growing with age, and this study cannot rule that out. Confirmed live against the Census API on 7 August 2026, the same firm-age-by-size cross-tabulation this study's own pull uses exists at NAICS 54: querying FIRM,EMP,FAGE,EMPSZFI for NAICS 54 returns 135 rows where both FAGE and EMPSZFI are a real band, not the 001 ("all") aggregate. The identical query against NAICS 541 or NAICS 5415 returns zero such rows – every row at that finer level carries a real firm-age band crossed only with the all-sizes aggregate, or a real size band crossed only with the all-ages aggregate, never both at once: https://api.census.gov/data/timeseries/bds?get=FIRM,EMP,FAGE,EMPSZFI&YEAR=2023&NAICS=541&for=us:*&key=REDACTED → rows with FAGE != "001" AND EMPSZFI != "001": 0 of 28 returned https://api.census.gov/data/timeseries/bds?get=FIRM,EMP,FAGE,EMPSZFI&YEAR=2023&NAICS=5415&for=us:*&key=REDACTED → rows with FAGE != "001" AND EMPSZFI != "001": 0 of 28 returned Census publishes the firm-age-by-size cross-tabulation only at the 2-digit sector level; at 3-digit and 4-digit NAICS it publishes each dimension's marginal totals separately, never the two crossed together. What would settle the composition question is a published cross-tabulation of subsector by firm age by size band, or restricted-access microdata that could reconstruct one; neither is publicly available. This study reports the aggregate rise and cannot attribute it to age versus subsector mix. ### The 300-firm threshold Any sector-and-age cell with fewer than 300 firms in firms_5plus is dropped before pct_under_20 or mean_size_5plus is computed. This is this study's own choice, not a number Census recommends – three times the leverage study's 100-firm floor, because this study's percentages split the population three ways (under 20, 20 to 99, 100 or more) rather than computing a single ratio, so each bucket has a thinner sample than 100 firms would give. All 55 published cells clear the floor; the smallest is finance and insurance's age-0 band, at 526 firms. ### Reproducing this study The pull script is scripts/research_pull.py, function plateau_rows(). Run from the repo root, with the Census API key passed as an environment variable, never hard-coded – the script aborts immediately if CENSUS_API_KEY is unset: CENSUS_API_KEY=your_key_here python3 scripts/research_pull.py It writes two files in the same run: [assets/data/plateau-by-sector-age-2023.csv](https://stoneforgegroup.com/assets/data/plateau-by-sector-age-2023.csv), the table this page draws from, and assets/data/leverage-by-sector-size-2022.csv, the companion leverage study's table, referenced above for the $2.7M and $6.3M revenue-per-firm figures. scripts/research-provenance.py runs on every deploy and checks every figure published on this page against a manifest of declared sources, offline – it never calls Census, so a deploy never depends on the API being up. A separate --refresh mode, run manually and not part of the deploy, re-pulls from Census live and reports any drift between the declared figures and what Census now returns. ### Known limitations The limitations that apply to every study on this site – current size is not peak size, survivors are not a random sample, this is a snapshot of different firms rather than one firm followed over time – are stated in full on [the research hub](https://stoneforgegroup.com/research/) and in What this does not show, above, and are not repeated here. Specific to this methodology: - **Disclosure-avoidance noise, not sampling error.** BDS applies noise infusion before publication (see Data source, above) to protect confidentiality. That is a different kind of imprecision than the Annual Business Survey's sampling error – it affects small cells more than large ones – and Census does not publish a per-cell margin for it, so this study cannot state one, and cannot state whether its headline changes survive it the way the companion leverage study's ABS-based figures can be checked against a stated RSE. See What this does not show, above. - **Subsector composition cannot be tested.** NAICS 54 bundles several subsectors with plausibly different size profiles into one code, and BDS crosses firm age with size band only at the 2-digit sector level – confirmed live against the API (0 of 28 rows at NAICS 541 and at NAICS 5415 carry both dimensions; see Subsector composition, above). The rise this study reports could partly reflect a shift in subsector mix rather than age alone. - **National totals only.** Every call runs for=us:*. None of the published figures show state or metro variation, though the API supports both. - **The 500-or-more band is frequently absent from the age-crossed data.** The exclusive 500-or-more employee code (657) exists in Census's overall EMPSZFI code space, but is absent – not zero, simply missing from the response – from most firm-age-crossed queries this study runs, consistent with small-cell suppression rather than an error. Professional services, age 26 or more, is one example: firms_5plus sums to 43,577 using only the 620/630/640/649 codes, because the 657 row did not appear in that query at all. Where this happens, this study's totals count that cell's 500-or-more firms as zero. The missing population is a small fraction of the tens of thousands of firms in most published cells, but a reader reproducing this table from Census's own API should expect the same absence, not treat it as their own error. - **The 300-firm threshold is a judgement call.** It is this study's own choice, not a number Census recommends. A reviewer could reasonably argue for a higher or lower cutoff; none of the 55 published cells sit close enough to 300 for the choice to matter here, but that is a property of this particular dataset, not a guarantee built into the rule. - **The CSV is rounded, not exact.** The published dataset stores mean_size_5plus rounded to one decimal place and pct_under_20 rounded to one decimal place, the same as displayed here. A reproduction computed directly from Census's raw FIRM and EMP integers, before any rounding, may differ from the published figures at the second decimal place or beyond. Published in full as [CSV](https://stoneforgegroup.com/assets/data/plateau-by-sector-age-2023.csv), including every sector, firm-age band, and the pct_20_99 and pct_100_plus columns not shown in the table above – along with firms_5plus and employees_5plus, the raw counts mean_size_5plus is computed from. ## Citing this work Stoneforge Consulting Group, *Firm Size and Firm Age*, 2026. Published 7 August 2026. Analysis of US Census Bureau Business Dynamics Statistics, 2023. https://stoneforgegroup.com/research/firm-size-and-age/ ## More research This study is part of Stoneforge's ongoing analysis of federal business data. See [all published studies, sources and limitations](https://stoneforgegroup.com/research/).