Small business statistics from 14 federal data studies show three patterns. The largest size classes outscore the smallest on management practices, benefits and AI training, while quits peak at mid-size workplaces. Young firms separate jobs at twice the rate of old ones. Employer-firm owners report closing their most recent earlier venture three times as often as selling it.
A small business statistic without its unit is easy to misuse. The 14 studies drawn on here go back to the federal source each time: the Census Bureau, the Bureau of Labor Statistics, the IRS and the Small Business Administration. Each was published on kamyarshah.com under a public-data protocol that registers the comparison in advance and discloses any later amendment.
The bottleneck is the unit, not the number
The risk in a statistic sits in its unit, not its digits. A share of workers gets read as a share of employers, a workplace as a company, a loan as a business. Each of the 14 studies counts something different, and the differences decide what a figure can support.
Some count establishments, meaning single workplaces. Others count firms, owners, workers, jobs, tax returns or loans. A firm with 300 employees can run ten workplaces of 30.
So the first rule of reading federal data is simple. Name the unit before the number, every time.
The anti-pattern: one headline, no denominator
The failure to avoid is the statistic stripped of its base. A line such as “small firms lose half their staff” says nothing about half of what, measured how, or over which period. A plan built on a figure like that has no fixed reference point. It can justify panic one quarter and complacency the next.
The calm alternative is diagnosis. Read the unit, the denominator, the size measure and the period, then decide what the figure can and cannot carry. Like the Theory of Constraints, the test starts by finding the one limit that governs the result, here the limit on meaning. Apply it to each figure below.
Size does not move every measure the same way
In manufacturing plants, the Census management practices score rises with every size class from 20 to 49 employees upward, in 2015 and 2021 (management score study). Paid sick leave reaches 75 percent of workers at establishments under 50 employees, against 91 percent at 500 or more (benefits access study). Among businesses already using AI, 13.4 percent of those with 1 to 4 employees report training staff, against 31.4 percent at 250 or more. Those last two compare only the smallest and largest classes.
Quits follow a different shape. In the Bureau of Labor Statistics job openings survey, the highest quit rate fell at establishments with 10 to 49 or 50 to 249 employees. That held in all 36 months of each window studied (quit rates study). Establishments with 5,000 or more had the lowest.
Injury rates are mixed. In the agency’s own figures, the national rate peaks at establishments with 50 to 249 employees (injury rates study). Inside the 17 of 652 industries with full data, the peak falls in a middle class in 8 and in the smallest class in 4.
The lesson is structural. Size is not one ladder that every measure climbs, and these are snapshots, not paths that any company follows as it grows. Read each measure by its own shape.
Free 20-Minute Operations Review
Dealing with a specific operational bottleneck? Kamyar Shah works with founders and CEOs to identify the root cause and build a fix.
Young firms separate jobs at twice the rate
Firms aged 0 to 1 separated 30.5 percent of their jobs in an average quarter, against 14.5 percent at firms 11 or more years old (separations by firm age study). On jobs that lasted a full quarter, the gap is 5.2 points, on a different denominator. The source study says this cannot show where the overall gap comes from, and separations include jobs lost when a firm closes.
A second study set three 2023 measures side by side for firms aged 0 to 1 and did not combine them (separations and job destruction study). Across 48 states and DC, it found 129.4 separations over the full year per 100 average quarter-start jobs. On the Census annual base, 16.9 jobs per 100 were destroyed and 9.5 per 100 were lost at closing establishments. Firms 11 or more years old showed 64.2, 11.4 and 3.4.
New hires at firms aged 0 to 1 who stayed a full quarter earned $3,397 a month (new-hire earnings study). That is 78.9 percent of the $4,306 earned by new hires into stable jobs at firms 11 or more years old. Inside young firms, those new hires earn 81.9 percent of what all their stable employees earn, against 62.1 percent inside old firms.
These are nominal covered earnings, not pay rates. The study says they describe earnings differences that mix industry, occupation, hours and location, and should not be used as a pay benchmark.
Consider a young services firm planning its next hiring round. The design goal would be a written onboarding sequence, with a RACI chart that names one manager as owner of the ninety-day check-in. Nothing in the data measures what such a sequence would change.
Closure is reported far more often than sale
The Census owner survey asked about the previous business each owner started most recently. Of owners who reported one, 71.4 percent still own it, 16.5 percent say it closed and 5.5 percent say it was bought (previous business study). Closure exceeds sale in 18 of 19 sectors. The unit is owners, so two owners of one business both count, and owners who left business entirely are not in the survey.
The same survey offers separate context. Of owners of responding employer firms, 51.3 percent are 55 or older. Among firms that answered, majority family ownership runs from 23.1 percent at 1 to 4 employees to 40.8 percent at 250 to 499 (family ownership study). A single-owner firm cannot count as family-owned, which shapes how that pattern reads.
Consider a founder whose company depends on the founder’s own approvals for pricing, hiring and key accounts. A buyer would be purchasing relationships held in one person’s head. Documented procedures, a second signer and a written pricing policy are the kinds of systems a buyer looks for.
Acquisition loans charge off less often by count
SBA 7(a) loans coded as a change of ownership were approved in fiscal 2018 to 2020. They charged off within five years at 2.03 percent, against 3.70 percent for existing-business loans (SBA charge-off study). By dollars the gap is much smaller, 0.93 percent against 1.07 percent, with no interval computed.
If the 20,446 loans with no business-age answer were all change-of-ownership loans, that group’s rate would be 3.68 percent, which would erase the gap. The study states that this is not evidence that buying a business is safer.
A second study ranked 13 sectors by these charge-off rates and by employer-firm death rates. The two rankings broadly matched, a descriptive rank correlation of 0.68 (sector study). It supports a ranking, not a cause.
Advertising is a small line on tax returns
On 2023 sole-proprietor tax returns, aggregate advertising was 1.18 percent of receipts, and contract labor exceeded it in 17 of 18 sectors (advertising share study). On 2022 S corporation returns, advertising is 1.61 percent of receipts in the smallest band shown, $25,000 to $100,000. It is 0.78 percent at $10 million to $50 million.
Sector advertising share did not line up with employer job growth, a near-zero rank correlation of minus 0.09 with no test run. Both tax figures are aggregates weighted by receipts, not what a typical business spends.
The same S corporation returns put officer pay plus net income at 22.12 percent of receipts in the smallest band shown (S corporation study). It peaks at 33.63 percent at $100,000 to $250,000 and is 9.34 percent at $10 million to $50 million. In that band, cost of goods is 63.26 percent, officer pay 2.42 percent and net income 6.92 percent.
These shares mix all industries and are shaped by tax rules on how shareholder-employees are paid. The study advises comparing with one’s own industry first, so they are context for a Balanced Scorecard review, never a benchmark for any one company’s unit economics.
Small firms that skip AI say it does not apply
Among businesses not planning to use AI, “not applicable” was the most common reason. It was cited by 63.3 percent of those with 1 to 4 employees, against 42.2 percent at 250 or more (AI use study). Lack of knowledge showed no clear difference between the two classes, 21.2 percent against 20.0 percent, and the answers could overlap. Among the smallest firms already using AI, 65.5 percent made no changes to use it, against 48.5 percent at 250 or more.
Consider a four-person firm that has tried an AI tool and changed nothing else. The design question is which recurring workflow the tool would sit inside, and who owns that workflow. Classify the work first, then decide whether a tool belongs in it.
From statistics to systems
Read together, the 14 studies set out a few clear shapes. The largest classes score higher than the smallest on management practices, benefits and AI training, while the middle classes carry the highest quit rates. Young firms separate jobs fastest, and closure is reported three times as often as sale.
None of this is a forecast for any single company, and none of it shows what a given fix would do. What the data supports is a reading discipline: name the unit and check the denominator. Which system to build is an operating judgment the data does not test.
That judgment is where fractional COO work begins, with the same reading test applied to a company’s own numbers. Trust in a plan depends on trust in its inputs. Statistics inform the plan, and systems carry it.
