[Home](https://stoneforgegroup.com/) · [Research](https://stoneforgegroup.com/research/) · The Leverage Gap 02 – Research # The leverage gap Professional services firms in the 250 to 499 employee band generate about 21% less revenue per dollar of payroll than firms with 5 to 9 employees. Construction firms generate about 3% more. ## Executive summary This study measures revenue against payroll, not revenue against headcount, across seven employment-size bands in four sectors – construction, real estate, professional services, and administrative & support – using the Census Bureau's 2022 Annual Business Survey. The measure is revenue per dollar of annual payroll: how much a sector's firms generate for every dollar they pay out, and how that figure moves as firms get larger. In professional services, the ratio falls about 21% from the 5-to-9-employee band to the 250-to-499 band – a gap far larger than the Annual Business Survey's own sampling error on either band (see Methodology and sources, below); administrative & support falls further, about 40%, the steepest of the four sectors. Construction and real estate hold roughly flat across the same range, though construction's own revenue is not fully comparable to the other three sectors' – about a fifth of it is money paid straight through to subcontractors, with none of the reporting firm's own payroll behind it (see below). Professional services is also the sector where the two measures split: the same firms keep more dollars per head above payroll as they grow, not fewer, even as the ratio falls. Administrative & support is the one sector where both measures move together, falling on both counts. Both readings are real, and both come from the same two underlying numbers; which one matters depends on the question being asked. A falling ratio means a firm keeps less of each payroll dollar as revenue. A rising per-head spread means the same firm still banks more dollars per person above what it pays them. Neither figure is profit – payroll is the only cost this study nets out, and rent, software, administration and owner compensation all remain inside the revenue side of both numbers. This is a snapshot of different firms at different sizes in one year, not one firm followed as it grows, and it cannot distinguish whether the pattern reflects staff seniority, overhead structure, service mix, or genuinely thinner unit economics in the largest band. No public federal dataset currently crosses profit, or even operating expenses, with firm size for a period recent enough to settle that question – see Methodology and sources, below. - Professional, scientific and technical services - Construction - Real estate - Administrative and support services Revenue per dollar of annual payroll by firm employment size. Sector names are labeled directly on the chart, not by color alone. US Census Bureau, Annual Business Survey, 2022. Finance and insurance is excluded – see Method, below. ## In the 250-to-499-employee band, professional services firms generate about 21% less revenue per dollar of payroll than firms with 5 to 9 employees, while construction firms generate about 3% more. In the 2022 Annual Business Survey, US professional, scientific and technical services firms with 5 to 9 employees generate $3.01 of revenue for every dollar of annual payroll. Firms with 250 to 499 employees generate $2.38 – about 21% less, even though those firms are many times larger. Administrative and support services falls further, from $3.71 to $2.22, about 40% less – the steepest decline of any sector measured here. Construction and real estate run the other way. Construction firms generate $5.26 per payroll dollar in the 5–9 band and $5.42 in the 250–499 band, about 3% more. Real estate rises from $5.70 to $5.88, also about 3% more. Both sectors hold their unit economics across the size range; professional services and administrative & support do not. Construction's figures are not fully like for like with the other three sectors here, though. About a fifth of construction's revenue is money paid straight through to subcontractors – other firms doing the work – with none of the reporting firm's own payroll behind it, where professional services reports no subcontracted cost of this kind at all. That inflates construction's revenue per payroll dollar next to a sector that keeps the same work in-house; see Construction's revenue includes subcontracted pass-through, in Methodology and sources, below, for the figures behind that. The change across bands – whether a sector's own ratio moves as its firms get larger – is not distorted by this the same way, since that comparison runs against the sector's own smaller firms rather than against a different sector's. ## Why this is measured against payroll, not headcount Revenue per employee is a weaker measure of leverage than it looks. A firm can raise it simply by employing more expensive people – more revenue lands per head, but so does more cost. The measure says nothing about whether the extra revenue outpaces the extra pay. Professional services shows exactly that pattern. Firms in the 5–9 band pay about $65,000 per employee; firms in the 250–499 band pay about $104,000 – about 60% more. Revenue per employee rises alongside it too, about 27% more, from $195,000 to $247,000. That is why the older measure made the largest band look like it was doing better. Revenue per payroll dollar tells the opposite story: it falls from $3.01 to $2.38 – about 21% less – because the extra pay outpaces the extra revenue it buys. The full comparison by the earlier measure, revenue per employee, is still worth having on record: Revenue per employee by firm size, 2022 Employment band Construction Real estate Professional services Admin & support 1–4 $266,000 $306,000 $202,000 $164,000 5–9 $295,000 $293,000 $195,000 $155,000 10–19 $312,000 $292,000 $203,000 $138,000 20–49 $348,000 $278,000 $212,000 $131,000 50–99 $384,000 $266,000 $227,000 $117,000 100–249 $390,000 $330,000 $245,000 $117,000 250–499 $462,000 $340,000 $247,000 $96,000 ## The same firms keep more per head, at a thinner multiple Revenue per dollar of payroll and revenue minus payroll, per employee, are different arithmetic on the same two numbers, and they can move in opposite directions without contradicting each other. The ratio asks how much revenue a firm generates for each dollar it pays out; the spread asks how many dollars are left over per person once payroll is covered. Revenue minus payroll per employee, and revenue per dollar of payroll, 5–9 to 250–499 employee bands, 2022 Sector Spread, 5–9 band Spread, 250–499 band Spread change Ratio change Construction $239,069 $376,563 58% more 3% more Real estate $241,844 $282,126 17% more 3% more Professional services $130,039 $143,382 10% more 21% less Admin & support $112,936 $52,753 53% less 40% less Professional services is the sector where the two readings diverge: the per-head spread rises even as the ratio falls, because payroll itself is rising faster than the dollars above it are. Construction and real estate carry no such tension – both measures rise together. Administrative & support is the only sector where the two readings agree, and they agree downward: firms in its largest band keep fewer dollars per head above payroll than firms in its smallest band, on a thinner multiple besides. Neither reading is profit. This spread nets out payroll only – rent, software, administration and owner compensation are all still inside it, uncounted as costs – and no current public dataset crosses operating expenses with firm size recently enough to compute a true profit-per-head figure; see Methodology and sources, below. ## What this means for an owner This part is interpretation, not measurement. A falling ratio is consistent with several different stories, and the data here cannot choose between them. If professional services firms in the largest band carry a more senior, more expensive staff mix than firms in the smallest band, revenue per payroll dollar would be lower there even if every added dollar of pay bought real added output. For that reading to hold, the roles inside the 250–499 band would need to be visibly more senior than the roles inside the 5–9 band. If larger firms carry more of their overhead as payroll – internal support functions a five-person firm buys from an outside contractor instead – the ratio would sit lower on paper with no difference in client-facing productivity. For that reading, the payroll in the larger band would need to sit mostly in non-billable roles. If firms in the largest band have a different service mix than firms in the smallest band, more labor-intensive work would push the ratio down for that reason alone. For that reading to hold, larger firms would need to be doing recognizably different work, not simply more of the same work with more people. It is also consistent with genuinely thinner unit economics in the largest band – the plain reading, and the one this study does not rule out. None of these four readings is asserted here; distinguishing between them needs data this dataset does not carry. ## What this does not show This is a snapshot of different firms at different sizes, observed in the same year – not one firm followed as it grows. It cannot show a firm's revenue per payroll dollar changing as that firm adds people and payroll; it can only show what firms of each size looked like in 2022. PAYANN, the payroll figure behind this ratio, excludes owner draws and distributions. In the 1–4 employee band, owners commonly take a draw rather than a salary, which understates payroll and inflates the ratio there – professional services shows $3.66 per payroll dollar in its 1–4 band, higher than its own 5–9 band's $3.01. That is exactly why this study anchors its headline comparison at 5–9, where a real payroll is the norm, rather than at 1–4. The decline itself does not depend on that choice: from 5–9 through 250–499 – bands where employees, not owner draws, make up the payroll – the ratio still falls by about 21% in professional services and about 40% in administrative & support. It is tempting to read a falling ratio as the reason a sector's firms stay small, but the data does not support that: administrative and support services shows the steepest fall of any sector measured here, and, in the separate [firm-age dataset](https://stoneforgegroup.com/assets/data/plateau-by-sector-age-2023.csv) (Census Business Dynamics Statistics, 2023) published alongside this study, is also among the sectors with one of the larger movements up the size bands as firms age. If a falling ratio held firms back, administrative and support services should show the opposite pattern. The two do not track. **This study cannot rule out subsector composition as the real driver of the professional-services decline.** NAICS 54 bundles legal services, accounting, architecture and engineering, computer systems design, management consulting and advertising into one code. If firms in the 250-to-499 band are disproportionately drawn from a lower-margin subsector than firms in the 5-to-9 band are, the decline this study reports could be a shift in mix rather than anything about size – and this dataset cannot tell the two apart. Confirmed live against the Census API, abscs publishes employment-size detail only at the 2-digit sector level: querying NAICS 541 or NAICS 5415 returns a single "All firms" row with no size-band breakdown at all (see Subsector composition, in Methodology and sources, below, for the exact query). A published cross-tabulation of subsector by size band would settle this; none exists. **Construction's higher revenue per payroll dollar is also not simply higher productivity.** About a fifth of construction's revenue – 21% in 2022, 20% in 2017, per the Economic Census – is the cost of work subcontracted out to other firms: money that passes through construction's reported revenue with none of construction's own payroll behind it, where professional services reports no subcontracted cost of this kind at all. That makes the level comparison between construction and the other three sectors here not fully like for like. The comparison this study leans on – how a sector's own ratio moves across its own size bands – holds up better, but is not immune: if larger construction firms subcontract a greater share of their work than smaller firms do, part of construction's flat ratio could be pass-through rather than retained economics, and no public dataset crosses subcontract cost with firm size to test that. See Construction's revenue includes subcontracted pass-through, below. None of this is undermined by ordinary sampling error. The Annual Business Survey's published relative standard errors, pulled for the bands behind each headline percentage, put the professional-services and administrative-and-support declines several times larger than the reported error on either band – not a formal significance test, but a check that the gap dwarfs the noise. See Sampling error on the headline comparison, below. ## 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, Annual Business Survey (ABS), 2022 data year. Table abscs, at https://api.census.gov/data/2022/abscs. The universe is employer firms only – businesses with paid employees. Nonemployer businesses (sole proprietors and others with no payroll) are not included in any figure on this page. The ABS is a stratified random sample of roughly 230,000 employer businesses, not a complete census of US employer firms. Every figure on this page carries sampling error for that reason, before any of the judgement calls documented below. Census does publish a relative standard error alongside each of the count and dollar variables this study uses – RCPPDEMP_S, PAYANN_S, EMP_S and FIRMPDEMP_S give the relative standard error, as a percentage, for RCPPDEMP, PAYANN, EMP and FIRMPDEMP respectively. This study's regular pull does not request them (see the exact query, below). The eight cells behind this study's headline percentages – the 5-to-9 and 250-to-499 bands, four sectors – were pulled separately and are checked against them in Sampling error on the headline comparison, immediately below. The other twenty published cells still carry no stated margin of error. See Known limitations. ### Sampling error on the headline comparison The Annual Business Survey is a sample, not a census, so every ratio on this page carries sampling error before any judgement call this study makes. Confirmed live against the Census API on 7 August 2026, the relative standard errors on receipts (RCPPDEMP_S) and payroll (PAYANN_S) – the two variables behind revenue per dollar of payroll – for the eight cells behind this study's headline percentages: Relative standard error on receipts and payroll, 5–9 and 250–499 employee bands, four sectors, 2022 ABS Sector Band RSE, receipts RSE, payroll RSE, employment Combined RSE on the ratio Professional services 5–9 2.9% 2.8% 2.5% 4.0% Professional services 250–499 5.7% 6.5% 4.8% 8.6% Construction 5–9 3.4% 2.9% 2.1% 4.5% Construction 250–499 5.1% 5.2% 5.3% 7.3% Real estate 5–9 4.8% 4.4% 3.2% 6.5% Real estate 250–499 7.4% 6.9% 15.0% 10.1% Admin & support 5–9 3.3% 2.6% 1.5% 4.2% Admin & support 250–499 6.5% 8.5% 10.5% 10.7% The combined column treats receipts and payroll as independent sources of error and adds them in quadrature – a conservative assumption, since the two move together in practice (a firm that under- or over-reports its size tends to do so on both), and the true error on their ratio is smaller than treating them as independent implies. Applying each band's own combined RSE as a range around its ratio: professional services' 5-to-9 band spans $2.89 to $3.13, and its 250-to-499 band spans $2.17 to $2.59. The two ranges do not overlap. Administrative & support, the steeper of the two declines, is more decisive still: $3.55 to $3.87 against $1.98 to $2.46, again no overlap. This is not a formal significance test – it is a check that the gap is much larger than the reported error, not a p-value – but on that check, both headline declines survive. Run the same check on construction and real estate, both "roughly flat" here, and the ranges overlap in both cases – consistent with calling those two sectors flat rather than rising, since sampling error alone could move either sector's ratio in either direction across these bands. That is a property of this particular comparison, not a general license to ignore RSEs elsewhere on this page: the twenty cells outside this headline comparison were not pulled with their RSEs and carry no stated margin of error at all. ### The exact query The literal call, with the key redacted: https://api.census.gov/data/2022/abscs?get=FIRMPDEMP,EMP,RCPPDEMP,PAYANN,EMPSZFI&for=us:*&NAICS2022=54&key=REDACTED Run once per sector, changing only NAICS2022, for four values: 23 (Construction), 53 (Real estate and rental and leasing), 54 (Professional, scientific and technical services), 56 (Administrative and support and waste management and remediation services). Geography is for=us:* – national totals only; see Known limitations. Each call returns one row per EMPSZFI code, eleven of them, not just the seven this study uses – Size band codes, below, lists which rows are kept and which are discarded, and why. The script issuing these calls is scripts/research_pull.py. ### Variable definitions - **FIRMPDEMP** – Census name "Number of employer firms." Firms with paid employees in the cell (sector by size band). Units: count of firms. - **EMP** – Census name "Number of employees." Paid employees at those firms. Units: count of people. - **RCPPDEMP** – Census name "Sales, value of shipments, or revenue of employer firms." Total reported receipts for firms in the cell. Units: thousands of dollars. - **PAYANN** – Census name "Annual payroll." Total annual payroll paid by those firms, excluding owner draws and distributions. Units: thousands of dollars. - **EMPSZFI** – Census name "Employment size of firms code." Which employment-size band a row describes. Units: a categorical code, not a quantity. RCPPDEMP and PAYANN are both reported in thousands of dollars. A row reporting RCPPDEMP of 1,274,548 is reporting 1,274,548 thousand dollars of revenue, not 1,274,548 dollars – multiply by 1,000 to reach the dollar figure. ### Size band codes Seven EMPSZFI codes are used here, all confirmed live against the Census API on 7 August 2026: - 612 – Firms with 1 to 4 employees - 620 – Firms with 5 to 9 employees - 630 – Firms with 10 to 19 employees - 641 – Firms with 20 to 49 employees - 642 – Firms with 50 to 99 employees - 651 – Firms with 100 to 249 employees - 652 – Firms with 250 to 499 employees Four more EMPSZFI codes exist in this dataset and are deliberately not used: - 001 – All firms. The total across every size; summing it alongside the seven bands above would double-count every firm. - 611 – Firms with no employees. Out of scope: this study, like the rest of the ABS, covers employer firms only. - 655 – Firms with less than 500 employees. An aggregate spanning all seven bands above; including it alongside them would double-count every firm already counted once in its own band. - 657 – Firms with 500 employees or more. Outside the size range this study reports; the largest published band is 250 to 499. **EMPSZFI exists in a second Census dataset with a different code space.** This study reads EMPSZFI from the Annual Business Survey (abscs). The companion firm-age dataset published alongside this study reads EMPSZFI from the Business Dynamics Statistics (BDS) time series instead, confirmed live to use a coarser breakdown for the same range of employees – BDS code 640 means "Firms with 20 to 99 employees" as one band, 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 Three ratios, computed per NAICS sector and employment-size band: - Revenue per dollar of payroll = RCPPDEMP ÷ PAYANN. Both variables are already reported in thousands of dollars, so the units cancel and the result is a plain ratio, not thousands of anything. - Revenue per employee = RCPPDEMP × 1,000 ÷ EMP. Converting RCPPDEMP out of thousands is required here, because EMP is a plain headcount, not a dollar figure. - Payroll per employee = PAYANN × 1,000 ÷ EMP. Same conversion, for the same reason. Display rounding: the ratio (revenue per dollar of payroll) is shown to two decimal places – $3.01, not the fuller division. Revenue per employee and payroll per employee are shown rounded to the nearest thousand dollars – $65,000, not the fuller figure. The published CSV (linked below) is finer-grained than the page but is not itself an unrounded, full-precision figure: it stores revenue per employee and payroll per employee rounded to the nearest whole dollar, and the ratio rounded to two decimal places, the same as displayed here. A reproduction computed directly from Census's raw RCPPDEMP and PAYANN integers, before any rounding, may differ from the published figures at the third decimal place or beyond. Percentage changes quoted in the prose above (21%, 40%, the two instances of 3%, 60%, 27%) are (new figure − old figure) ÷ old figure × 100, rounded to the nearest whole percent. The four ratio-based percentages use the ratio already rounded to two decimals in the table below (for example 3.01 to 2.38). The two headcount-based percentages (60%, 27%) use the dollar figures as displayed, rounded to the nearest thousand ($65,000 to $104,000; $195,000 to $247,000); recomputed from the CSV's finer, whole-dollar figures – 64,854 to 104,038 dollars per employee, and 194,893 to 247,420 dollars per employee – they land on the identical whole-percent result. That is a property of these particular numbers, not a rule that would hold for any pair of figures. ### Inclusion and exclusion rules - **Under 100 reporting firms.** Any sector-and-band cell with fewer than 100 firms is dropped before the ratio is computed. This is this study's own threshold, not a number Census publishes as a minimum – below it, a handful of firms can move the ratio by more than the sample can support. The smallest of the 28 published cells has 733 reporting firms (real estate, 250 to 499 employees), comfortably clear of the floor; none of the published figures were affected by this rule, but a sector or band added later could be. - **Negative reported values.** A row is dropped outright if firms, EMP, RCPPDEMP or PAYANN comes back negative – a Census suppression or disclosure-avoidance code, not a real value. - **Zero or negative employment.** A row is also dropped if EMP is zero or negative, since dividing by zero employees is undefined. - **Zero or negative payroll, guarded separately.** Where PAYANN is zero or negative but the row otherwise passes, the row is kept for revenue per employee, but payroll per employee and revenue per dollar of payroll are left blank rather than computed. This did not affect any of the 28 published cells, all of which report positive payroll, but it is why the pull script checks this case on its own rather than folding it into the employment guard above. **Finance and insurance (NAICS 52) is excluded entirely**, not narrowed to a comparable sub-industry. Insurance carriers (NAICS 524) report 1,164,243 dollars of receipts per employee at the all-firms level; securities, commodity contracts and other financial investment firms (NAICS 523) report 660,719 – both confirmed live against the Census API on 7 August 2026, and both published on this page rounded to the nearest thousand ($1,164,000 and $661,000). Both sit far outside the $96,000-to-$462,000 range spanned by the four sectors this study does cover, because insurance premiums and securities transaction values pass through the reporting firm rather than being revenue it keeps for its own services – RCPPDEMP for a NAICS 52 firm is not measuring the same thing RCPPDEMP measures for a construction or professional-services firm. No sub-industry inside NAICS 52 is a substitute; the gross-flow problem is structural to how insurance and securities receipts are reported, not a property of any one size band. ### 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 firms in the 250-to-499 band are disproportionately drawn from a lower-margin subsector than firms in the 5-to-9 band are, the 21% decline this study reports could be a shift in subsector mix rather than anything about size itself, and this study cannot rule that out. Confirmed live against the Census API on 7 August 2026, abscs publishes employment-size-of-firms detail only at the 2-digit NAICS sector level. Requesting a 3-digit or 4-digit code inside NAICS 54 returns a single row for "All firms," with no EMPSZFI size-band breakdown at all: https://api.census.gov/data/2022/abscs?get=NAICS2022_LABEL,EMPSZFI_LABEL,RCPPDEMP,EMP,EMPSZFI&for=us:*&NAICS2022=541&key=REDACTED → [["NAICS2022_LABEL","EMPSZFI_LABEL","RCPPDEMP","EMP","EMPSZFI","NAICS2022","us"], ["Professional, Scientific, and Technical Services","All firms","2613858701","11153447","001","541","1"]] https://api.census.gov/data/2022/abscs?get=NAICS2022_LABEL,EMPSZFI_LABEL,RCPPDEMP,EMP,EMPSZFI&for=us:*&NAICS2022=5415&key=REDACTED → [["NAICS2022_LABEL","EMPSZFI_LABEL","RCPPDEMP","EMP","EMPSZFI","NAICS2022","us"], ["Computer Systems Design and Related Services","All firms","591640922","2269821","001","5415","1"]] Both queries return exactly one row, coded EMPSZFI 001 – "All firms," the same aggregate code Size band codes, above, excludes from this study's own NAICS 54 pull for double-counting the seven bands it sums. No size-band rows exist at either the 3-digit or 4-digit level to sum instead. What would settle the composition question is a Census cross-tabulation of employment-size band by 4-digit NAICS within 54, or restricted-access microdata that could reconstruct one; neither is publicly available. This study reports the aggregate decline and cannot attribute it to size versus subsector mix. ### Construction's revenue includes subcontracted pass-through A construction firm's reported receipts routinely include money it pays straight through to other firms doing part of the work – subcontracted labor and materials the general contractor never keeps. The Economic Census publishes this directly: ecnbasic, variable CSTSCNT, "Cost of construction work subcontracted out to others." Confirmed live against the Census API on 7 August 2026: https://api.census.gov/data/2022/ecnbasic?get=NAICS2022_LABEL,RCPTOT,CSTSCNT&for=us:*&NAICS2022=23&key=REDACTED → Construction, RCPTOT 2,920,771,250, CSTSCNT 613,241,103 (both thousands of dollars) https://api.census.gov/data/2017/ecnbasic?get=NAICS2017_LABEL,RCPTOT,CSTSCNT&for=us:*&NAICS2017=23&key=REDACTED → Construction, RCPTOT 1,994,166,047, CSTSCNT 401,642,346 (both thousands of dollars) 613,241,103 ÷ 2,920,771,250 = 20.996%, about 21% of construction's total receipts in 2022. 401,642,346 ÷ 1,994,166,047 = 20.141%, about 20% in 2017 – close enough to the 2022 figure that this reads as structural to how the industry is organized, not a one-year artifact. The same query against NAICS 54 returns CSTSCNT of exactly zero: professional services reports no subcontracted construction cost at all, which is expected – CSTSCNT is a construction-specific variable – but it is also exactly the asymmetry that makes the two sectors' revenue figures not directly comparable. This does not mean construction's revenue per payroll dollar is wrong; it means a fifth of what it counts as revenue is money that never touched construction's own payroll, which the ratio's numerator does not distinguish from revenue the firm actually earned providing labor. Professional services carries no equivalent pass-through, so the two sectors' *levels* are not fully like for like. ecnbasic carries no firm-size dimension – CSTSCNT is published only as a sector total, not broken out by employment-size band – so whether larger construction firms subcontract a greater or smaller share of their work than smaller ones do cannot be tested with this dataset either, which is why the level comparison is qualified here rather than corrected: there is no published figure to correct it with. ### Why the 5–9 band is the baseline, not 1–4 Two independent reasons, either one sufficient on its own. PAYANN excludes owner draws and distributions. In the 1–4 employee band, an owner working in the business commonly takes a draw instead of a salary, which understates PAYANN and inflates revenue per dollar of payroll for that band alone. Professional services shows this directly: $3.66 per payroll dollar in the 1–4 band, higher than the same sector's own 5–9 band at $3.01, even though 1–4 is the smaller, less mature group of firms. A ratio that rises because its denominator is measured incompletely is not a finding about leverage. Separately, the companion firm-age dataset published alongside this study (Census Business Dynamics Statistics, 2023) already excludes firms with fewer than five employees from its own bands – its smallest band is 5–9. Anchoring this study's headline at 5–9 keeps the two studies comparable at the same starting point, rather than one starting at 1 employee and the other at 5. The 1–4 band is published in the full table below for completeness. It is never the band a headline percentage is computed against. ### Why the per-head spread above is not profit The same firms keep more per head, at a thinner multiple, above, reports revenue minus payroll, per employee – not profit. PAYANN is the only cost this study subtracts from RCPPDEMP; rent, software, services purchased from other firms, administrative overhead and owner compensation above payroll all remain inside the revenue side of that figure, uncounted as costs. No current public federal dataset supports a true profit-per-head figure broken out by firm size for a period recent enough to be useful here. The Annual Business Survey this study otherwise draws on carries no profit or operating-expense variable at all – RCPPDEMP and PAYANN are the closest it comes. The quinquennial Economic Census publishes operating expenses by industry, but not crossed with a firm-size dimension, so a firm's costs cannot be split out by how many employees it has. The one federal dataset that does cross operating expenses with employment size of firms is the Economic Census's 2012 Business Expenses Survey component – over a decade old at the time of this study's publication, and not repeated since. Building a genuine profit-per-head series by firm size and sector is not possible from any dataset currently public. ### The full table All seven bands, four sectors, computed from the published dataset: Revenue per dollar of payroll by firm size, 2022 Employment band Construction Real estate Professional services Admin & support 1–4 $5.70 $6.92 $3.66 $4.38 5–9 $5.26 $5.70 $3.01 $3.71 10–19 $5.11 $5.21 $2.80 $3.11 20–49 $5.09 $4.59 $2.60 $2.82 50–99 $5.17 $4.05 $2.53 $2.54 100–249 $5.13 $4.97 $2.59 $2.33 250–499 $5.42 $5.88 $2.38 $2.22 Published in full as [CSV](https://stoneforgegroup.com/assets/data/leverage-by-sector-size-2022.csv), including firm counts, revenue per firm and both per-employee figures not shown in either table on this page. ### Reproducing this study The pull script is scripts/research_pull.py. 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: [assets/data/leverage-by-sector-size-2022.csv](https://stoneforgegroup.com/assets/data/leverage-by-sector-size-2022.csv), the table this page draws from, and assets/data/plateau-by-sector-age-2023.csv, the companion firm-age dataset referenced in What this does not show, above. 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: - **Margin of error stated for the headline comparison only.** The Annual Business Survey is a stratified sample, not a census, so every figure on this page carries sampling error. This study pulled the relative standard error for the eight cells behind its headline percentages and confirmed the professional-services and administrative-and-support declines are several times larger than the combined error on either band – see Sampling error on the headline comparison, above. The other twenty published cells were not re-pulled with their RSEs and carry no stated confidence interval; a reviewer who wants one for those has to pull the remaining fields and propagate the error through the ratio, as this study did only for the headline eight. - **Subsector composition cannot be tested.** NAICS 54 bundles several subsectors with plausibly different economics into one code, and Census does not publish employment-size detail below the 2-digit sector level – confirmed live against the API. The decline this study reports could partly reflect a shift in subsector mix rather than size alone; see What this does not show and Subsector composition, above. - **Construction's revenue includes subcontracted pass-through, untested by firm size.** About a fifth of construction's receipts is cost of work subcontracted to other firms, per the Economic Census, which is not broken out by employment-size band. That qualifies the level comparison between construction and the other three sectors here; see Construction's revenue includes subcontracted pass-through, above. - **National totals only.** Every call runs for=us:*. None of the published figures show state or metro variation, though the API supports both; a sector's ratio could differ by geography in ways this study cannot show. - **The 100-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 28 published cells sit close enough to 100 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.** As stated above, the published dataset stores rounded values, not the full-precision division. A reproduction that starts from Census's raw integers rather than this study's CSV may differ from the published figures at the third decimal place or beyond. ## Citing this work Stoneforge Consulting Group, *The Leverage Gap*, 2026. Published 7 August 2026. Analysis of US Census Bureau Annual Business Survey, 2022. https://stoneforgegroup.com/research/the-leverage-gap/ ## 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/).