The Census Bureau’s Quarterly Workforce Indicators (QWI) are built from state unemployment insurance wage records linked to Census business data. They count jobs that start and end each quarter. Census defines a separation as follows: “A separation is indicated when a job is present in one quarter, but is not present in the following quarter.” Census also notes that “Voluntary separations (retirement, new job) and involuntary separations (layoff, firing) cannot be separately identified.” This article therefore says “separations,” not “quits” or “turnover.”

Firm age is a property of the whole firm, measured nationally. Census explains: “Firm age and size are defined at the national level, rather than the state level. A national firm may frequently be larger or older than the part of that firm found in any one state.” Firm age is reported for private-sector firms only.

Separation rates by firm age

Census defines the separation rate as “All separations as a percent of average employment,” where average employment is the average of jobs at the start and end of the quarter. Means of four quarterly rates, private sector, all industries, both sexes, all worker ages, not seasonally adjusted:

Source: U.S. Census Bureau, LEHD Quarterly Workforce Indicators, national files by firm age, release R2026Q3, downloaded October 3, 2026. The 2017 to 2019 column is the mean of 12 quarters. Hire rate is all hires as a percent of average employment.

Separation rates fall at each step of firm age in both periods, and all five were lower in the recent window than before the pandemic.

The registered comparison: firms aged 0 to 1 against firms aged 11 or more

The study set one comparison in advance: the separation rate at firms aged 0 to 1 year minus the rate at firms aged 11 or more, averaged over the latest four quarters, at or before 2025Q3, in which every value used had a QWI status of 1 or 6.

The difference was positive in all four quarters. Over the 12 quarters of 2017 to 2019 it averaged 17.3 percentage points. A pooled version, summing separations and average employment across the four quarters before dividing, gives 30.5% and 14.5%, the same 16.0-point difference.

QWI values are protected with noise. Census describes it this way: “all released data are ‘fuzzed.'” Every value carries a status flag, and this analysis used only values with status 1 (OK) or 6 (calculated from other released measures, no significant distortion). Census publishes no sampling error or other margin of error for QWI, and a covariance-aware interval for this difference is not available from the public tables. No statistical significance is claimed, and QWI values can be revised in later releases.

What the separation measure includes, and what it does not separate

A separation here is any job that ended, whatever the reason. When a firm closes, all of its jobs end, and those jobs are counted with quits and layoffs. QWI does not show how much of any firm-age class’s rate comes from closures. Firms aged 0 to 1 also had a higher hire rate than separation rate in the recent window: 41.4 hires and 30.5 separations per 100 jobs per quarter. The table does not adjust for industry, location or worker mix, so nothing in it says that firm age itself causes the difference.

The rate of separations from stable jobs, those that had lasted at least a full quarter, follows the same order: 13.5% at firms aged 0 to 1, 12.9% at 2 to 3, 12.4% at 4 to 5, 11.6% at 6 to 10, and 8.3% at 11 or more, in the recent window. The gap between the youngest and oldest firms is smaller on this measure: 5.2 percentage points, against 16.0 for all separations. The two rates use different denominators, so this comparison is descriptive and does not show how much of the all-jobs gap comes from jobs that never reached a full quarter.

What the numbers mean for an operator

These are descriptive aggregates for the stated federal-data population. They do not identify causal effects, diagnose an individual company, or estimate the result of adopting a practice. The guidance that follows is operating judgment, not a finding from the data.

A company aged 0 to 1 year is operating where, on average, almost a third of jobs end every quarter. Some of that is the company finding its shape, and some of it is cost: every job that ends early carries the cost of hiring, training and lost output, and young companies have the least slack to absorb it. The stable-job figures cannot show where the gap comes from, because the two rates use different denominators. As a matter of operating judgment, the first months of employment are still a practical place to start looking.

For an established firm, the useful number is not the national 14.5% but the firm’s own rate measured the same way. The national figure blends every industry and region.

How to handle it: protect the first ninety days

The steps below are an operating recommendation. This study did not test them, and the Census Bureau does not endorse them.

1. Measure separations the same way. Each quarter, count jobs that ended (for any reason) and divide by the average of headcount at the start and end of the quarter. Track jobs that ended within the first ninety days separately. In operating experience, that second number is often where a young company’s losses sit.

2. Write down the job before hiring for it. Early exits often come from a mismatch between what the role was described as and what it turned out to be. A one-page description of the outcomes expected in the first ninety days, agreed before the search, makes the hiring decision and the first review concrete.

3. Give every new hire an onboarding sequence. A written first-week plan, a named person to go to, and check-ins at thirty, sixty and ninety days. Young companies tend to skip this because they are busy, and that is when it matters most.

4. Separate the separations you chose from the ones you did not. Record whether each exit was the company’s decision or the employee’s, and the reason. QWI cannot make that distinction, but a company can, and the two call for different fixes.

5. Hire to a plan, not to a backlog. A young company that hires in bursts when work piles up, and lets people go when it slows, will show high separations whatever else it does. A twelve-month hiring plan tied to revenue is the operating answer to both.

What this data does not tell you

• It does not separate quits from layoffs or from jobs lost when firms close.
• It does not show why separation rates differ by firm age, or that any practice lowers them.
• Firm age is the age of the national firm, not of a local workplace, and not the worker’s age.
• QWI is noise-protected administrative data with no published sampling error, and values can be revised.
• It is a different source from the federal job openings and turnover survey, and this analysis does not compare its figures numerically with that survey.

Method

Already published by the agency: Census publishes QWI separation, hire and stable-separation rates by firm age every quarter. New here: four-quarter and pre-pandemic averages across all firm-age classes, the registered young-versus-established comparison quarter by quarter, and the stable-job comparison.

Source. U.S. Census Bureau, LEHD Quarterly Workforce Indicators, national firm-age files (rates and counts), release R2026Q3, data through 2025Q3, downloaded October 3, 2026. Selection: private ownership, all industries, both sexes, all worker ages, not seasonally adjusted.

Window rule. The current window is the latest four quarters, at or before 2025Q3, in which every value used had a status of 1 (OK) or 6 (calculated from other released measures, no significant distortion). That was 2024Q3 to 2025Q2. Values with any other status are treated as missing, never as zero. The baseline is 2017Q1 to 2019Q4.

Protocol. The comparison, the window rule and the baseline were fixed in a dated protocol before any rate was computed. The protocol (wave 2 section, frozen October 4, 2026) and its dated amendments are at https://kamyarshah.com/public-data-protocol/

Check. All figures were recomputed from the raw files by a separately written program, built without access to this study’s results. As a second check, the same rates were rebuilt by adding up state-level counts for the 48 states and the District of Columbia that the Census data service returned for each of these quarters. Alaska and Michigan returned no data, so this is a partial check rather than a national reconstruction. The quarterly differences from that rebuild (16.4, 15.8, 16.0 and 15.4 points) agree with the national file to within 0.1 point.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

The IRS Statistics of Income (SOI) division publishes detailed tabulations of corporate tax returns each year. This article uses Table 3.2 of the Corporation Complete Report, titled by the IRS “Returns of Active Corporations, Form 1120S, Total Receipts and Deductions, Portfolio Income, Rental Income, and Total Net Income,” which groups S corporation returns by size of business receipts. The unit is the tax return. The figures are estimates from a sample of returns: as the IRS explains in its documentation of the corporate sample (tax year 2015 edition), the estimates “are based on a sample, they may differ from population aggregates resulting from a complete census of all corporate income tax returns.”

How to read these numbers

Except for return counts and differences in points, every percentage below is an aggregate share: the total dollars of one line, summed across all returns in a size band, divided by the total business receipts of the same returns. Larger returns contribute more dollars, so each share is weighted toward the bigger returns in the band. It is not the margin of a typical or median company, and deficits are netted into the IRS line for net income less deficit. The measure is tax-return income, which follows tax rules rather than the financial statements a company prepares for its owners or lenders.

The registered comparison

Net income (less deficit) from a trade or business, as a share of business receipts, tax year 2022:

The IRS does not publish coefficients of variation for these size tables, so no interval is shown and no significance is claimed.

Selected lines as a share of receipts, by size band

Each line as a percentage of business receipts, aggregate across all returns in the band, tax year 2022:

Source: Internal Revenue Service, Statistics of Income, Corporation Complete Report, Table 3.2, tax year 2022 (file 22co32ccr.xlsx), downloaded October 3, 2026. Selected compensation and benefits is officer compensation plus salaries and wages plus employee benefit programs. It is not a measure of every labor-related cost. Net income is the line for net income less deficit from a trade or business. The band under $25,000 is excluded, because with very small receipts any ratio to receipts is unstable. The lines shown are the seven the study’s protocol lists, plus two sums. They are not every line on the return and do not add up to 100. Protocol Amendment 2 records that clarification.

Three patterns run through the table. Cost of goods sold rises steadily with size, from 13.55% in the smallest band shown to 76.78% in the largest. Officer compensation peaks at 15.88% in the $100,000 to under $250,000 band and falls to 0.56% at $250 million or more. Net income as a share of receipts is highest at $100,000 to $250,000 (17.75%) and falls in every larger band after that. Because the table mixes industries, it cannot show how much of the cost-of-goods pattern reflects industry mix rather than receipts size.

Officer compensation and net income should be read together

The IRS line is compensation of officers. Some officers are also shareholders, but the table does not say how much of this compensation went to shareholder-employees. Where an officer is also a shareholder-employee, tax rules affect how payments are split between salary and distributions. The IRS states that “S corporations must pay reasonable compensation to a shareholder-employee in return for services that the employee provides to the corporation before non-wage distributions may be made to the shareholder-employee.” Because salary decisions can move dollars between the two lines, the last column of the table adds them together. On that combined measure, the $1 million to under $2.5 million band reported 16.08% of receipts, against 9.34% for $10 million to under $50 million.

The same comparison, 2014 to 2022

Net income as a share of business receipts for the two registered bands, each year’s tabulation:

The bands are in nominal dollars and were not adjusted for inflation, so a company can move between bands without changing in real size. Each year is a separate set of returns, not the same companies followed over time. The band definitions were checked to be the same in every year shown. The smaller band’s share was the higher of the two in all nine years.

What the numbers mean for an owner

These are descriptive aggregates for the stated federal-data population. They do not identify causal effects, diagnose an individual company, or estimate the result of adopting a practice. The guidance that follows is operating judgment, not a finding from the data.

The table compares different groups of returns at one point in time. It does not follow companies as they grow. In the aggregate, cost-of-goods shares rise across the bands shown, officer-compensation shares fall after the $100,000 to under $250,000 band, and net income as a share of receipts peaks in that band and falls in every larger one. Officer compensation plus net income falls from about a third of receipts at $100,000 to $250,000 to under a tenth at $10 million to under $50 million.

As operating judgment, the shape of the table matches what owners often experience as they grow: purchasing, staffing and management take a larger role, and the owner’s own work accounts for less of the result. The table does not show that any one company will follow that path, and industry differences are large. It is a reason to plan for the change deliberately rather than to assume the old ratios will hold.

How to handle it: build your own common-size statement and compare it fairly

The steps below are an operating recommendation. This study did not test them, and the IRS does not endorse them.

1. Put your own numbers in the same format. Express each line of the income statement as a percentage of revenue: cost of goods, officer pay, other salaries, benefits, rent, advertising and net income. Do it for at least three years so a single unusual year does not mislead.

2. Add officer pay and net income before comparing. In an S corporation where the owner is also an officer, how the owner is paid shifts dollars between those lines. Comparing the sum avoids reading a salary decision as a change in performance. The salary itself should reflect reasonable compensation for the work, which is a question for the company’s tax adviser.

3. Compare with your industry first, size second. These IRS figures mix all industries. A distributor and a consulting firm of the same size will look nothing alike. Use industry benchmarks for the main comparison and this table to understand how the shape changes with size.

4. Watch the cost-of-goods and labor lines as you grow. As receipts grow, purchasing and staffing become the main levers. Review gross margin and labor cost as a share of revenue every month, and set a target for each before the next stage of growth, not after.

5. Decide what net income is for. Retained profit funds growth, reserves and debt service. An owner who knows how much the business needs to keep, and for what, can set distributions and salary with the company’s tax adviser without starving the business.

What this data does not tell you

• These are aggregate-dollar shares weighted toward larger returns, not the margin of a typical company in the band.
• It mixes all industries, so it cannot separate industry mix from receipts size.
• It reflects tax-return accounting, not financial statements prepared for lenders or buyers.
• Officer compensation and net income are shaped by tax rules and owner choices.
• The IRS publishes no sampling error for these size tables, so small differences between bands or years may not be meaningful.

Method

Already published by the agency: the IRS publishes these dollar amounts and return counts by size of business receipts in Table 3.2 each year. New here: the lines shown converted to shares of business receipts for each band from $25,000 upward, selected compensation and benefits, defined as officer compensation plus salaries and wages plus employee benefit programs, and officer compensation plus net income, and the registered comparison for 2014 to 2022.

Source. Internal Revenue Service, Statistics of Income, Corporation Complete Report, Table 3.2 (Form 1120-S by size of business receipts), tax years 2014 to 2022, downloaded October 3, 2026.

Calculation. Within each band, each line’s aggregate amount is divided by the band’s aggregate business receipts and multiplied by 100. Net income is the line for net income less deficit from a trade or business. The band definitions were compared across years before the series was built. Cells not published would be treated as missing, never as zero.

Protocol. The measure, the two bands and the years were fixed in a dated protocol before any ratio was computed. The protocol (wave 2 section, frozen October 4, 2026) and its dated amendments are at https://kamyarshah.com/public-data-protocol/

Check. All figures, including the return counts for each band, were recomputed from the raw files by a separately written program, built without access to this study’s results.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

The Census Bureau’s Annual Business Survey (ABS) asks the owners of employer firms about themselves and their history in business. Census describes the survey’s coverage as “All domestic, non-farm, for profit businesses with at least one paid employee. Additionally, non-profit organizations (excluding universities) that operate in industries identified as potentially having R&D are sampled.” It notes that “The Annual Business Survey excludes the collection of data from nonemployer businesses.” This article uses the owner tables from the 2024 ABS. Census explains that the survey’s name uses the collection year: “The 2024 ABS, for example, covers reference year 2023.”

The unit throughout is the owner, not the firm. One firm can have several owners, and the survey records a separate answer for each owner. Census also notes that the owner tables “include only businesses that were classifiable by sex, ethnicity, race, and veteran status and responded to at least one owner demographic question.” The counts below are therefore owners of responding employer firms.

The question, and who answers it

The question asks about the status of the previous business the owner started most recently, not including the current one. The form records one answer for each owner. The question is about a business the owner started. A business the owner bought, or a business started before the most recent one, is outside it.

That framing matters for one reason above all. Everyone represented in this table owned an employer firm in reference year 2023. An owner whose previous business sold and who had retired, taken a job, or otherwise left employer-firm ownership before then is not represented. The table describes the history of people who were still owners. It is not a census of how businesses end.

Status of owners’ previous businesses

Census estimates that, of 4,212,412 owners who answered the question, 2,051,500 (48.7%) said the current business is their first. The remaining 2,160,912 reported a previous business they had started. Among those owners:

Source: U.S. Census Bureau, 2024 Annual Business Survey (reference year 2023), Characteristics of Business Owners, file AB2300CSCBO, downloaded October 3, 2026. United States, all sectors, all owners of respondent employer firms. Owners who did not answer the question are excluded, as Census excludes them from its count of owners reporting. Census rounds each published estimate separately, so the five individual answers add to 2,160,913, one more than the 2,160,912 calculated as all reporting owners minus first-time owners.

The study’s single registered comparison is the sold share minus the closed share: 5.5% minus 16.5%, or minus 11.0 percentage points. The shares are this study’s calculations from published counts, with owners reporting a previous business as the denominator. Census publishes standard errors for each answer as a share of all owners who answered, but not for these conditional shares or for the sum of the two “purchased” answers. These figures are point estimates, and no interval is shown. The public table includes an “Other” answer, and this article does not guess what it contains.

By sector

The same shares, calculated for each sector in which Census published every needed count (sector of the owner’s current firm, not of the previous business):

Shares are percentages of owners in that sector who reported a previous business. Differences are calculated from unrounded shares, so they can differ by 0.1 from subtracting the rounded figures. “Industries not classified” could not be calculated because a needed count was not published. In 18 of the 19 sectors shown, the closed share was larger than the sold share. In management of companies and enterprises, the sold share was larger. Sector figures are point estimates with no interval, and small differences between sectors should not be read as meaningful.

Context: owner age and when owners acquired their business

Census publishes the age of owners of respondent employer firms directly, with standard errors:

Adding the two oldest groups gives 51.3% aged 55 or over. That sum is this study’s calculation and has no published standard error. The public tables do not cross owner age with the status of the previous business, so this article does not say how older owners’ previous businesses ended.

When owners acquired their current business: 2.1% before 1980, 5.3% in the 1980s, 10.6% in the 1990s, 19.2% in 2000 to 2009, 33.7% in 2010 to 2019, and 4.3%, 4.9%, 4.3% and 1.3% in 2020, 2021, 2022 and 2023 respectively. 14.4% answered “don’t know.” These are Census’s published shares of owners who answered.

What the numbers mean for an owner

These are descriptive aggregates for the stated federal-data population. They do not identify causal effects, diagnose an individual company, or estimate the result of adopting a practice. The guidance that follows is operating judgment, not a finding from the data.

Among owners of responding employer firms whose answer concerned the previous business they had started most recently, closure was reported about three times as often as sale. The data cannot say why. It also misses owners who sold or closed and then left ownership altogether, and the direction in which that tilts the comparison is unknown. The table does not estimate the chance that a still-operating business will eventually find a buyer.

The practical reading is about transferability. A business sells when someone other than the founder can run it, can see why it makes money, and can believe the money will keep coming after the founder leaves. Many owner-run companies are built the other way: the owner holds the key customer relationships, prices from memory, approves every exception, and keeps the numbers in a format only the owner understands. Those businesses can be very profitable for the owner and still be hard to sell, because what a buyer would be paying for walks out of the door with the owner.

How to handle it: build a business someone else could buy

The steps below are an operating recommendation. This study did not test them, and Census does not endorse them.

1. Find out what a buyer would be buying. List what makes the business money: customers, contracts, a process, a location, a license, a team. Next to each, write who controls it today. Any line where the answer is the owner alone is a transfer risk a buyer will discount.

2. Move one relationship or decision off the owner each quarter. Pick the largest customer or the most frequent decision that only the owner handles. Hand it to a named person, with a written standard for how it is handled, and review the results monthly. In operating experience, a business that runs for a month without the owner is easier to sell than one that cannot.

3. Make the numbers readable by a stranger. Monthly financial statements closed on a fixed schedule, revenue broken down by customer and by line of business, and owner compensation separated from what the business needs to pay a manager. A buyer’s first questions are about these. An owner who cannot answer them quickly is negotiating from weakness.

4. Get a valuation before you need one. An outside view of what the business is worth, and why, shows the gaps while there is still time to close them. Doing it three to five years before an intended exit leaves room to act on the answer.

5. Decide what happens if no buyer appears. Closing is a legitimate outcome, and among these owners it was reported about three times as often as a sale. Planning for it, including a review with appropriate legal, HR or financial advisers of any obligations to employees, customers and lenders, gives the owner a fallback if a sale does not come together.

What this data does not tell you

• It does not show why previous businesses closed or sold, or what any practice would change.
• It covers only people who owned an employer firm in reference year 2023. Owners who sold or closed and then stopped owning a business are not counted, and that could move either share in either direction.
• It describes the previous business the owner started most recently, not every business the owner ran or bought.
• The unit is the owner. Two owners of the same previous business can each be counted.
• It does not cross owner age with how previous businesses ended.

Method

Already published by the agency: Census publishes the counts and percentages for every answer to this question, as shares of all owners who answered, in the ABS Characteristics of Business Owners tables. New here: the shares restricted to owners who reported a previous business, the sold-minus-closed comparison, and the same calculation by sector.

Source. U.S. Census Bureau, 2024 Annual Business Survey, reference year 2023, Characteristics of Business Owners (AB2300CSCBO), downloaded October 3, 2026. Selection: United States, all owners of respondent firms for sex, ethnicity, race and veteran status, all sectors, and each two-digit sector. Counts flagged by Census as suppressed or not available are treated as missing, never as zero.

Calculation. Owners with a previous business = “Total reporting” minus “This is the owner’s first business.” Sold = “purchased by another individual” plus “purchased by another company.” Each share = count ÷ owners with a previous business × 100. No interval is reported for these shares, because Census does not publish the covariances needed to compute one.

Protocol. The comparison, the denominator and the sector rule were fixed in a dated protocol before any estimate was computed. Before results, the question was confirmed to be single-answer from the Census questionnaire. The protocol (wave 2 section, frozen October 4, 2026) and its dated amendments are at https://kamyarshah.com/public-data-protocol/

Check. All figures were recomputed from the raw file by a separately written program, built without access to this study’s results. The national counts match the Census API (2023 ABS owner characteristics) exactly. The share of owners aged 65 or over, 23.1%, matches the Census release.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

The Census Bureau’s Annual Business Survey (ABS) describes its coverage this way: “All domestic, non-farm, for profit businesses with at least one paid employee. Additionally, non-profit organizations (excluding universities) that operate in industries identified as potentially having R&D are sampled.” The 2024 ABS covers reference year 2023. This article uses the Census table of business characteristics by employment size of firm. The unit is the employer firm, counted once, which is different from the Census owner tables, where each owner is counted.

What family-owned means here

Census asks whether two or more members of one family owned the majority of the business. A firm with one owner cannot answer yes, even if that owner’s family depends on the business. So majority family-owned, in this article, means exactly that: two or more relatives together hold the majority. It does not mean that firms outside that definition have no family involvement.

Family-owned and single-owner shares, by firm size

Shares of employer firms that answered each question, with Census’s published standard errors. The 95% intervals are calculated here as the published percentage ± 1.96 × the published standard error:

Source: U.S. Census Bureau, 2024 Annual Business Survey (reference year 2023), Characteristics of Businesses by employment size of firm, file AB2300CSCB04, downloaded October 3, 2026. United States, all sectors, all firms. Percentages are Census’s published shares of firms that answered each question. Only the mutually exclusive employment-size classes set in the study’s protocol are shown.

The study’s single registered comparison is the family-owned share at 1 to 4 employees minus the share at 250 to 499: 23.1% minus 40.8%, or minus 17.7 percentage points. Each share has its own interval, calculated from its published standard error. The difference does not, because Census does not publish the covariance between size classes, so no interval is shown for it.

The 500-or-more estimate carries a published standard error of 24.4 percentage points on a 39.2% share. An interval built from it would run below zero. The figure is shown for completeness and should not be used.

How the two ownership measures differ by firm size

Among firms with 1 to 4 employees that answered the number-of-owners question, 69.0% are owned by one person. Among firms with 250 to 499 that answered the number-of-owners question, 22.8% are. Under this definition a one-owner firm cannot be majority family-owned, and one-owner firms make up a much larger share of the smallest firms. The two questions are answered separately, so the table cannot show exactly how the two shares overlap. The higher family-owned share in the 250 to 499 class than in the 1 to 4 class is consistent with that, and the public tables do not allow ownership structure and other size-related differences to be separated. This article does not claim that family ownership itself becomes more common as firms grow, or that family firms grow faster.

The family-owned share does not rise at every step. It is 34.8% at 20 to 49 employees and 33.2% at 50 to 99, and the intervals overlap for 20 to 49 and 50 to 99, for 50 to 99 and 100 to 249, and for 100 to 249 and 250 to 499.

What the numbers mean for an owner

These are descriptive aggregates for the stated federal-data population. They do not identify causal effects, diagnose an individual company, or estimate the result of adopting a practice. The guidance that follows is operating judgment, not a finding from the data.

Two facts from the table provide context for transition planning. First, among firms with 1 to 4 employees that answered the number-of-owners question, 69.0% report one owner. That makes operational continuity worth addressing, but the table does not show whether a successor has been identified. Second, among firms with 250 to 499 employees that answered the family-ownership question, 40.8% report majority ownership by two or more members of one family. That makes family relationships relevant to ownership planning, but the table does not show how those firms make or settle decisions.

Census does not measure succession plans, and nothing here says family firms are more or less prepared. The operating observation is narrower: when two or more relatives own a majority, future ownership and management decisions may involve both governance arrangements and family relationships. The rights and approvals required depend on the firm’s governing documents and applicable law, so owners should consult qualified legal and tax professionals.

Census does not measure succession plans, and nothing here says family firms are more or less prepared. The operating observation is narrower: when two or more relatives own the majority, decisions about the business’s future need agreement among people whose relationship does not end when the business changes hands.

How to handle it: separate the ownership plan from the management plan

The steps below are an operating recommendation. This study did not test them, and Census does not endorse them.

1. For a single-owner firm, write down who could run it. One owner means one point of failure. Name the person who would run the business for six months if the owner could not, list what that person would need to know, and start transferring it. This is the same work that makes the business easier to sell.

2. For a family-owned firm, separate three questions. Who will own the business, who will run it, and who will be paid by it are different questions. Family firms often answer them together by default. Answering them separately, in writing, avoids treating ownership as a reward for working in the business, or management as a reward for being family.

3. Put the agreement in the documents. A buy-sell agreement among the owners, with a valuation method and funding, can decide in advance what happens if an owner dies, divorces, becomes disabled, or wants out. Without one, those events are settled under pressure. A lawyer and an accountant should draft and review it.

4. Set a date for the review, not just for the plan. Ownership and family circumstances change. A plan reviewed every year or two, on a fixed date, stays usable. A plan written once tends to be out of date when it is needed.

5. Bring in someone outside the family for the hard conversations. An outside adviser or board member who answers to the business, not to any one relative, can raise the questions family members find hardest to raise with each other.

What this data does not tell you

• It does not measure succession plans, family involvement short of majority ownership, or whether family firms perform differently.
• The family-owned definition excludes single-owner firms, which constrains how the size pattern can be read.
• The unit is the firm. It is not comparable to Census owner tables, which count owners.
• The difference between size classes has no published interval, and the 500-or-more estimate is too imprecise to use.

Method

Already published by the agency: Census publishes the family-ownership and number-of-owners percentages and standard errors by employment size of firm. New here: the two measures laid side by side across size classes, the registered comparison between the smallest and the 250-to-499 class, and the explanation of how the definition constrains the pattern.

Source. U.S. Census Bureau, 2024 Annual Business Survey, reference year 2023, Characteristics of Businesses by employment size of firm (AB2300CSCB04), downloaded October 3, 2026. Selection: United States, all firms for sex, ethnicity, race and veteran status, all sectors, and the mutually exclusive size classes shown above. Cells flagged as suppressed or not available would be treated as missing, never as zero. None of the cells used was flagged.

Calculation. Shares and standard errors are Census’s published values. Intervals are the published share ± 1.96 × the published standard error, rounded to one decimal. Before any family-owned or one-owner estimate was viewed, Census’s denominator for these percentages was confirmed, using a different question, to be the firms that answered.

Protocol. The comparison and the size classes were fixed in a dated protocol before any estimate was computed. The protocol (wave 2 section, frozen October 4, 2026) and its dated amendments are at https://kamyarshah.com/public-data-protocol/

Check. All percentages and standard errors shown were recomputed from the raw file by a separately written program, built without access to this study’s results, and matched. The family-owned percentages and standard errors also match the Census API (2023 ABS business characteristics) exactly. The difference and the intervals were recalculated from those values.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

This study uses SBA 7(a) loans approved in fiscal 2018 to 2020, excluding loans that were cancelled or never disbursed and loans whose business age was blank or recorded as Unanswered. Selling a company and buying one are decisions most people make rarely. The Small Business Administration’s 7(a) loan records offer one way to see how loans coded as a change of ownership performed in their first five years, next to loans in the age codes this study groups as existing businesses. For change-of-ownership loans, the business-age field identifies the loan as a change of ownership rather than recording the purchased business’s age.

The SBA publishes 7(a) loan-level files that include the business’s age category and whether the loan was later charged off. Starting in fiscal 2018, the records include a “Change of Ownership” category. This study followed 7(a) loans approved in fiscal 2018 through 2020 for five years from their approval date, after excluding loans that were cancelled or never disbursed and loans whose business age was blank or recorded as Unanswered. Loans coded “New, Less than 1 Year old” or “New Business or 2 years or less” (4,886 loans) fall in neither comparison group. After the exclusions, 115,884 loans remain: 14,516 change-of-ownership loans, 74,530 existing-business loans, 21,952 startup loans and those 4,886.

The main comparison

Of 14,516 change-of-ownership loans, 295, or 2.03%, were charged off within five years of approval (95% interval 1.82% to 2.27%). Of 74,530 loans in the age codes this study groups as existing businesses, 2,760, or 3.70%, were charged off in the same window (95% interval 3.57% to 3.84%).

Change-of-ownership loans were lower by 1.67 percentage points (loan-level 95% interval: 1.40 to 1.92 points lower). These are all in-scope loans for those years, not a sample. The intervals describe uncertainty for loans of this kind under a simple binomial model, not sampling error for the listed loans. They are not adjusted for lender clustering, which may make them narrower than a clustered interval would be. This was the single primary comparison, set before the group charge-off rates were calculated. The fiscal 2018 to 2019 sensitivity, the sector breakdown, the dollar comparison, the franchise split, and the loan-size bands were also registered in advance. They are descriptive, not further tests. The existing-business group is defined in Method. It is not an SBA published category, it includes age codes for businesses as young as 2 years, and the age field does not show whether owners were changing.

For context only, 1,030 of 21,952 loans to open a new business, or 4.69%, were charged off within five years. Comparisons between acquisition and startup loans have been published before. The comparison with existing businesses is the focus here.

Four industries, described

Four industry sectors had at least 1,000 loans in both groups. This breakdown is descriptive, not a further test. In each, the change-of-ownership point estimate was lower. The intervals for each group are in the table. Overlap of those intervals is not a test, and no sector difference interval was pre-registered.

Five-year charge-off rate, number of loans and 95% interval (source: SBA 7(a) FOIA files FOIA_7a_FY2010_FY2019_asof_260630.csv and FOIA_7a_FY2020_Present_asof_260630.csv, data as of June 30, 2026, downloaded October 3, 2026):

Dollars, loan size, and franchises

SBA’s data dictionary defines the approved amount (GrossApproval) as “Total loan amount” and the charged-off amount (GrossChargeOffAmount) as “Total loan balance charged off (includes guaranteed and non-guaranteed portion of loan)”. Measured as charged-off amount within five years divided by approved amount, the ratio was 0.93% for change-of-ownership loans (bootstrap 95% interval 0.76% to 1.12%) and 1.07% for existing-business loans (0.98% to 1.16%). The approved amount is not the amount disbursed or outstanding, so this ratio is not a loss rate on money actually advanced. This comparison is descriptive. No interval for the difference between the two ratios was registered or computed, so it does not show whether the ratios differ.

In descriptive breakdowns among change-of-ownership loans, five-year charge-off rates by approved loan size were 2.66% for loans under $150,000 (1,878 loans), 3.10% for $150,000 to under $500,000 (5,031), 1.43% for $500,000 to under $1 million (3,071), and 0.99% for $1 million or more (4,536). Change-of-ownership loans with a franchise code had a rate of 1.88% (3,089 loans) and those without one had 2.07% (11,427). This split is descriptive, and no interval for the difference was computed. Loan-size bands differ in industry and lender mix. Differences by loan size reflect both. Nothing here isolates an effect of loan size. Franchise status comes from the SBA franchise code field, which can be incomplete. The registered band labels (under $150,000, $150,000 to $500,000, $500,000 to $1 million, and over $1 million) were ambiguous at the exact dollar boundaries. The analysis code, written before any rate was computed, used half-open bands: under $150,000, $150,000 to under $500,000, $500,000 to under $1 million, and $1 million or more. A dated protocol amendment, made after the results were seen, recorded that clarification. The bands and the calculation did not change.

Limits that matter

A charge-off is a loan outcome, not a business failure. Loans charged off more than five years after approval are not counted, so these are not lifetime loss rates. The clock starts at approval, not at disbursement. The exploratory box below checks whether that matters. Payment relief under Section 1112 of the CARES Act covered many SBA loans in 2020, including loans approved in fiscal 2018 and 2019 that were then in repayment, so no cohort here is free of that relief. A sensitivity check on fiscal 2018 and 2019 loans only was registered in advance. It is reported because it was registered. It is not a remedy for that relief. Excluding fiscal 2020, the rates were 2.20% for change-of-ownership loans (211 of 9,572) and 4.25% for existing-business loans (2,327 of 54,779), the same direction. Loans from those two fiscal years could still have received Section 1112 payments while in repayment.

Business age was blank or recorded as Unanswered for 20,446 approvals, so those loans could not be assigned to either comparison group. If those loans differ in charge-off outcomes, excluding them could affect the reported gap. The exploratory box below describes a selection-risk check for them. These are SBA-guaranteed loans only, not all business purchases. Change-of-ownership loans finance a purchase, while existing-business loans finance other uses such as working capital or equipment, and the two can carry different SBA underwriting requirements. They may also differ in loan size, industry, and lender. The rates here are not adjusted for any of that. This study does not compare loan-size mix, or charge-off rates by size, between change-of-ownership loans and the existing-business group. It cannot say whether size mix accounts for any of the gap. The gap describes how these loans performed. It is not proof that buying a business is safer.

Exploratory, not pre-registered. Reason included: outside reviewers asked whether change-of-ownership loans take longer to disburse, which would give them less time at risk under an approval-based clock. SBA’s FirstDisbursementDate field is filled for 14,511 of 14,516 change-of-ownership loans and 74,520 of 74,530 existing-business loans. The median time from approval to first disbursement was 31 days in both groups. The mean was 42.6 days for change-of-ownership loans and 50.0 days for the existing-business group. Counting five years from first disbursement, for loans first disbursed by June 30, 2021 (14,500 and 74,463 loans), the charge-off rates were 2.11% (95% interval 1.89% to 2.36%) and 3.83% (3.69% to 3.97%). The gap remained in the same direction when the five-year clock started at first disbursement. That does not rule out every timing difference.

The same box also covers the 20,446 loans excluded because business age was blank or recorded as Unanswered. Their five-year charge-off rate, counted from approval, was 4.85% (991 of 20,446 loans, 95% interval 4.56% to 5.15%), higher than either comparison group, and their median approved amount was $50,000. Which group these loans belong to is unknown. As a simple reassignment scenario, if every one of them had been a change-of-ownership loan, that group’s rate would have been 3.68%, about the same as the existing-business group’s 3.70%. This is not a worst case, because group membership among these loans could be related to their outcomes. This box is descriptive.

What the numbers mean for buyers and sellers

This section and everything after it, up to Method, are interpretation and guidance from operating practice. They are not findings of this study.

Among disbursed 7(a) loans with a usable business-age code, change-of-ownership loans were charged off less often within five years than loans in the existing-business group, 2.03% against 3.70%. Measured by dollars, the two ratios were much closer, 0.93% against 1.07%. The figures describe loans, not businesses, and they cannot show that buying a business is safer than running one. The 20,446 loans without a usable age code are a reminder that the answer depends on which records are counted, as the exploratory box above shows.

For a buyer, the honest reading is narrow. Among loans with a usable age code, purchases financed through the 7(a) program did not show a higher five-year charge-off rate in these years than the existing-business group. That is useful context. It is not a reason to relax diligence, because a loan that is repaid can still sit on a business that disappoints its new owner.

For a seller, the figures say something indirect. Buyers can finance purchases through this program, which widens the pool of buyers who can pay a fair price. A business is easier to finance when its earnings and operations can be verified by someone who has never worked in it.

How to handle it: make the business easy to finance and easy to hand over

The steps below are an operating recommendation. This study did not test them, and SBA does not endorse them. Lending terms, eligibility, and tax treatment change, so each step should be confirmed with a lender, an accountant, and transaction counsel.

For sellers, two to three years before a sale:

1. Separate the owner’s role from the business. Write down what the owner does each week and move as much of it as possible into documented procedures and named roles. A business that depends on one person’s memory is harder to value and harder to finance.
2. Clean the financial record. Consistent monthly closes, reconciled accounts, and a clear line between personal and business spending make earnings verifiable. Lenders and buyers discount what they cannot verify.
3. Plan the handover. Decide in advance how long the seller will stay involved after closing and which relationships, with customers, suppliers, and staff, need a personal introduction.

For buyers, before and after closing:

1. Test the earnings, not the story. Ask for the monthly financials, the tax returns, and the customer concentration, and reconcile them to each other. A quality-of-earnings review by an outside accountant is worth its cost on a financed purchase.
2. Map the operation before day one. Identify the routines that keep the business running and who performs them. Anything that lives only in the seller’s head is a transition risk to cover in the purchase agreement and the handover plan.
3. Hold working capital in reserve. A transition usually costs more and takes longer than planned. A buyer who borrows to the limit leaves no room for the first quarter that goes differently than expected.

What this data does not tell you

• It does not show that buying a business is safer than starting or running one, or that any practice reduces charge-off risk.
• It covers SBA-guaranteed 7(a) loans approved in fiscal 2018 to 2020 only, not all business purchases.
• A charge-off is a loan outcome, not a business failure, and loans charged off after five years are not counted.
• Loans without a usable business-age code are excluded, so the comparison depends on which records are counted.

A business sale is an operating event as much as a financial one. About 98 in 100 change-of-ownership loans in this data were not charged off within five years. Building a business that can keep running after its owner changes is the part both sides control.

Method

Already published by the agency: the SBA publishes the loan-level files but no table for this comparison. Private analysts have published acquisition-versus-startup default comparisons. New here: a fixed five-year charge-off window for fiscal 2018 to 2020 loans, comparing change-of-ownership loans with existing-business loans, overall and by sector.

Source. U.S. Small Business Administration, 7(a) FOIA loan-level files for fiscal 2010 to 2019 and fiscal 2020 to present, data as of June 30, 2026, downloaded October 3, 2026. The files are public records. In SBA’s data dictionary, each record is a 7(a) or 504 loan (the Program field is an “Indicator of whether loan was approved under SBA’s 7(a) or 504 loan program”), and BusinessAge is a “Categorical description of the age of the business”. No borrower, lender, or franchise brand is named here.

Population. All 7(a) loans approved in fiscal 2018 through 2020. Loans cancelled or never disbursed (18,229) were excluded. Loans whose business age was blank or recorded as “Unanswered” (20,446) were also excluded. By fiscal year, they were 8,355 loans (13.8% of fiscal 2018 approvals), 8,536 (16.4% of fiscal 2019), and 3,555 (8.4% of fiscal 2020). “Existing business” combines the SBA age codes “Existing or more than 2 years old”, “Existing, 5 or more years”, “Less than 5 years old but at least 4”, “Less than 4 years old but at least 3” and “Less than 3 years old but at least 2”. Startup loans are those coded “Startup, Loan Funds will Open Business”. Other new-business codes are in neither group.

Measure. A loan counts as charged off if its charge-off date falls within 1,826 days of its approval date, which this article calls five years. For loans approved from March 1, 2019 through February 28, 2020, the window crosses two leap days and ends one day before the fifth anniversary of approval. In those records, two existing-business charge-offs fell on that anniversary day and are not counted, which does not change any rate shown at the precision reported. Every loan in the window has at least five years of follow-up as of June 30, 2026, so open and paid loans count as not charged off rather than being dropped. Count rates use Wilson 95% intervals, the difference uses the Newcombe method, and the dollar ratios use a bootstrap with 2,000 draws. Sectors use the first two digits of the NAICS code, combined as NAICS defines them. No multiplicity adjustment was made. Rates are not adjusted for lender clustering.

Protocol. The analysis plan, including the single primary comparison, was fixed in a dated protocol before any group charge-off rate was computed. Later amendments are listed in it: https://kamyarshah.com/public-data-protocol/

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

The Bureau of Labor Statistics measures quits every month by the size of the establishment in its Job Openings and Labor Turnover Survey (JOLTS). An establishment is a single workplace, not a company: a small workplace can belong to a large company, and one company can have several workplaces. BLS assigns each establishment to a size class under its own survey methodology, which may not match a workplace’s headcount in a given month. BLS describes the survey’s coverage this way: “The JOLTS program covers all private nonfarm establishments, as well as civilian federal, state, and local government entities in the 50 states and the District of Columbia.” This article uses the private nonfarm size-class series (BLS “total private”) only.

Quit rates rise, then fall, with workplace size

Average monthly quits per 100 employees, private nonfarm establishments, all six published BLS size classes, seasonally adjusted (source: U.S. Bureau of Labor Statistics, JOLTS, Table 7 series JTS100000000000001QUR through JTS100000000000006QUR, flat files downloaded October 3, 2026, data through August 2026):

BLS publishes 10 to 49 and 50 to 249 as two separate classes, and no combined rate for them is calculated here. August 2026 is preliminary, and BLS revises seasonally adjusted figures each year, so later releases can differ. The middle-class high and the 5,000-or-more low held month by month, not only on average. In each of the 36 months from September 2023 through August 2026, the highest published quit rate estimate among the six classes was at 10 to 49 or 50 to 249 employees, and the lowest was at 5,000 or more. The same was true in each of the 36 months from January 2017 through December 2019. The smallest class did not have the lowest rate. In the recent window, establishments with 1 to 9 employees (1.81) and with 1,000 to 4,999 (1.83) were close. At the precision of the published monthly rates, the 1-to-9 class was higher in 16 months, lower in 15, and tied in 5. In the earlier window, it was higher in 28 months, lower in 1, and tied in 7.

BLS publishes median standard errors for JOLTS by industry and region, not by establishment size class, so this article makes no claim of statistical significance. It reports averages and how consistently the pattern held from month to month.

How this framing came about. The study’s plan, fixed before any size-class figure was computed, set one comparison: the 1-to-9 class against a pooled group of the two largest classes. Outside reviewers pointed out that this comparison hides the shape shown above. After the results were seen, a dated amendment moved the six published classes to the front. The registered comparison is unchanged and is reported in full in the next section.

The registered comparison: 1 to 9 against a pooled 1,000-or-more group

The study set one comparison in advance: the monthly quit rate at establishments with 1 to 9 employees minus the rate at establishments with 1,000 or more. Over the 36 months, the difference averaged 0.32 quits per 100 employees per month. The 1-to-9 group had the higher rate in 86.1% of the 36 monthly comparisons, and the monthly difference ranged from negative 0.21 to positive 1.01. The 36-month average is the primary figure. The month-by-month sign count and range describe how the monthly difference varied around it.

The registered comparison group pools the two largest published classes. Over the 36 months it averaged 1.49 quits per 100 employees per month. The pooled 1,000-or-more figure is not a BLS seasonally adjusted series. Each published series is seasonally adjusted on its own, so adjusted figures need not add up exactly, and this pooled rate, built from already adjusted levels, is not the rate BLS would publish for one combined class. Each month this study reconstructs employment for each of the two classes as the seasonally adjusted total separations level divided by the seasonally adjusted total separations rate, times 100. It then divides the sum of the two classes’ seasonally adjusted quits levels by the sum of that employment, times 100. Because the published levels and rates are both rounded, this reconstructed employment can differ from the denominator BLS used, and it is not a separately published BLS series. The pooled group combines two published classes with very different quit rates, 1.83 and 0.86, so its result depends heavily on the 5,000-or-more class. As in the rest of this article, no claim of statistical significance is made.

Layoffs, hires and separations

The registered design compared four groups on all four measures. The 250 to 999 class was left out of that design before any size-class estimate was computed, and it appears only in the six-class quits table above.

Seasonally adjusted rates per 100 employees per month, private nonfarm establishments, averaged over each window (source: U.S. Bureau of Labor Statistics, JOLTS, seasonally adjusted total private by establishment size class, Table 7 series JTS100000000000001QUR, JTS100000000000002QUR, JTS100000000000003QUR, JTS100000000000005QUR, JTS100000000000006QUR, with the matching quits, LD, HI and TS level (L) series and LD, HI and TS rate (R) series, flat files downloaded October 3, 2026, data through August 2026. Inputs are the published classes. The 1,000 or more column is not a BLS series and is not in Table 7. It is this study’s pool of the seasonally adjusted 1,000 to 4,999 and 5,000 or more levels and rates):

The 1 to 9, 10 to 49, and 50 to 249 columns are averages of published BLS seasonally adjusted rates. The 1,000 or more column, including the earlier window and the change row, is not a BLS series. It is this study’s pooled calculation from the published 1,000 to 4,999 and 5,000 or more classes (see Method). Table values are rounded to two decimals. Unrounded values are in the study’s reproducibility files. Changes are the later window minus the earlier window, calculated from unrounded window averages and then rounded to two decimals, so they can differ by 0.01 from subtracting the rounded values shown. Total separations also include other separations, such as retirements, transfers and deaths, which is consistent with quits plus layoffs and discharges being less than total separations for every size group in each window.

Quit levels relative to total separations levels

Quits are one category of separation from an establishment’s payroll. Here, the quit-to-separations ratio is the sum of seasonally adjusted quits levels divided by the sum of seasonally adjusted total separations levels over September 2023 to August 2026, so it is not the ratio of the averaged rates in the table. Because seasonally adjusted series are not strictly additive, this ratio is approximate. On that basis, the ratio was 56.0% at establishments with 1 to 9 employees, 64.9% at 10 to 49, 64.5% at 50 to 249, and 54.4% at 1,000 or more.

JOLTS counts events, not people. It does not record worker demographics, the reason for a quit, or the kind of job involved. Size classes differ in industry and occupation mix. Differences by size reflect both. Nothing here isolates an effect of size.

What the numbers mean for an owner

This section and everything after it, up to Method, are interpretation and guidance from operating practice. They are not findings of this study.

In every month of both periods studied, the highest published quit rate among the six classes was in either the 10 to 49 or the 50 to 249 employee class, and the lowest was at 5,000 or more. The data does not explain why, and size classes differ in industry and occupation mix, so the pattern is not proof that size itself drives quits. For an owner, the more useful reading is practical. A workplace in either middle class that sees about two and a half quits per 100 employees a month is close to the national average for its class over September 2023 to August 2026. JOLTS does not show the spread of rates within a class, so closeness to the average is a benchmark, not a verdict.

That changes the question an owner should ask. A rising headcount often brings layers of supervision, more varied roles, and less direct contact with the owner, and the middle classes are where those changes tend to arrive. Whether they cause the higher quit rates is outside this data. Whether a specific workplace is above or below its own class average is something an owner can measure.

How to handle it: measure first, then fix the causes that are visible

The steps below are an operating recommendation. This study did not test them, and BLS does not endorse them.

1. Measure quits consistently with the BLS definition. Count voluntary quits during the month, excluding retirements and transfers to other locations. Divide by employment for the pay period that includes the 12th of the month, then multiply by 100. Track it every month for at least a year. A single month moves in large steps in a small workplace and says little on its own.

2. Compare against the closest class. Use the published establishment size class that best matches the workplace, not the company as a whole and not the largest classes. BLS assigns classes under its own survey methodology, so treat the match as approximate. For private nonfarm establishments with 10 to 49 employees, the seasonally adjusted average from September 2023 to August 2026 was 2.52 quits per 100 employees per month. A workplace well above its class average over a full year has something specific to find.

3. Ask the people who leave, and the people who stay. A short, consistent exit conversation, written down the same way each time, turns individual departures into a record. A brief conversation with people who stay, asking what would make them leave, often surfaces the same issues earlier and at lower cost.

4. Fix the systems the conversations point to. Common findings are unclear expectations in the first weeks, uneven scheduling, and supervisors promoted without guidance on how to manage. Each has a procedural answer: a written onboarding sequence, a published scheduling rule, and a short management standard for new supervisors.

What this data does not tell you

• It does not show why quit rates differ by size class, or that any practice lowers them.
• It counts quits, not people. JOLTS records no demographics and no reason for leaving.
• Figures are national averages across industries for private nonfarm establishments. An establishment is a single workplace, not a company.
• August 2026 data are preliminary, and BLS revises seasonally adjusted figures each year.

As operating guidance rather than a finding: know the workplace’s own rate, compare it to its class, and build the routines that help people stay once they are hired. The national figures are a benchmark for that work. They do not show which practices reduce quits.

Method

Already published by the agency: BLS publishes monthly JOLTS levels and rates by establishment size class in Table 7 of each release (“Job openings, hires, and separations levels and rates by establishment size class, seasonally adjusted”). New here: 36-month averages for all six published size classes in a current and a pre-pandemic window, a month-by-month count of where quits were highest and lowest, the registered pooled comparison, and quit-to-separations ratios.

Source. U.S. Bureau of Labor Statistics, Job Openings and Labor Turnover Survey, flat files (series for total private, seasonally adjusted, by establishment size class), downloaded October 3, 2026. Data run through August 2026, which BLS marks preliminary. BLS notes: “Establishment size class data are produced for the total private sector only.”

Measures. Rates are the number of events during the month as a percent of employment, which BLS states as per 100 employees. The 1,000-or-more group combines the 1,000 to 4,999 and 5,000-or-more classes. For each class and month, implied employment is the total separations level divided by the total separations rate, times 100. The pooled rate for each measure is the sum of the two classes’ levels divided by the sum of their implied employment, times 100. It is not an average of their rates. Because published levels and rates are rounded, the pooled figures cannot be reproduced exactly from rounded table values. Window figures are simple means of the monthly rates. The quit-to-separations ratio is total quits divided by total separations over the window, both in levels. No multiplicity adjustment was made. BLS revises seasonally adjusted data each year, so later downloads can differ slightly.

Protocol. The analysis plan, including the single primary comparison and the windows, was fixed in a dated protocol before any size-class estimate was computed. The six-class framing of this article was adopted after the results were seen, by a dated amendment (Amendment 8) that keeps the registered comparison and reports it in full. All amendments are listed in the protocol: https://kamyarshah.com/public-data-protocol/

Check. The July and August 2026 monthly rates for quits, layoffs and discharges, and hires in the downloaded files matched BLS Table 7 of the September 29, 2026 release for each of the six published size classes. The pooled 1,000-or-more figures are this study’s calculation and do not appear in Table 7.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

Adopting a tool and changing how work gets done are two different things. In a supplement to its Business Trends and Outlook Survey, collected from November 17, 2025 to February 8, 2026, the Census Bureau asked employer businesses that used AI in any business function what they had changed in order to use it. Figures below come from the Census table “Employment Response Estimates” and its standard errors, downloaded October 3, 2026. Census describes the survey as “representative of all employer businesses in the U.S. economy, excluding farms”.

Census has already published AI use by company size and reported that larger companies use AI more. In this supplement, 17.6% of employer businesses with 1 to 4 employees reported using AI in any of its business functions in the last two weeks, against 30.6% of those with 250 or more employees. This study looks at the changes AI-using businesses reported making in order to use it.

Training current staff: an 18-point gap

Among businesses that used AI in any business function, 13.4% of those with 1 to 4 employees said they trained current staff to use AI. Among businesses that used AI in any business function and had 250 or more employees, 31.4% did. The smaller group was lower by 18.0 percentage points (approximate 95% interval: 15.9 to 20.1 points lower). This was a pre-registered comparison, and the interval excludes zero. The item asks about training current staff. The table does not show whether that item means the same thing in a business with 1 to 4 employees as in a business with 250 or more. The 18.0 point gap is a difference in answers, not a measured difference in training effort. Size classes differ in industry and occupation mix. Differences by size reflect both. Nothing here isolates an effect of size.

On the same question, respondents could select more than one change, so these shares are not parts of one whole and can add to more than 100%. Every answer Census published is shown below, so no answer was picked out after the results were seen. Apart from training current staff, these figures are descriptive.

Changes made in order to use AI, among businesses that used AI in any business function (source: U.S. Census Bureau, BTOS AI supplement 2025 to 2026, sheet “Employment Response Estimates”, downloaded October 3, 2026. Answer wording is Census’s.):

Cited lack of knowledge: no clear difference between the smallest and largest

The study’s second pre-registered comparison asked about businesses not planning to use AI in the next six months. Among businesses not planning to use AI in the next six months, 21.2% of those with 1 to 4 employees cited a lack of knowledge about AI’s capabilities, against 20.0% of those with 250 or more employees. The difference, 1.2 percentage points, has an approximate 95% interval of negative 1.3 to positive 3.7 points, which includes zero. The pre-registered comparison does not show a clear difference between businesses with 1 to 4 employees and businesses with 250 or more on this reason. The size classes in between are outside that comparison. Among businesses not planning to use AI in the next six months, the middle size classes’ point estimates for citing lack of knowledge about AI’s capabilities run from 23.0% to 25.0%, which is descriptive only. This result does not show that the share is the same at every size.

On the reasons question, respondents could select more than one reason, so these shares are not parts of one whole. Every answer is shown below. Apart from lack of knowledge, these figures are descriptive. The most common answer in both size classes was that AI is not applicable to the business.

Reasons for not planning to use AI, among businesses not planning to use AI in the next six months (source: U.S. Census Bureau, BTOS AI supplement 2025 to 2026, sheet “Employment Response Estimates”, downloaded October 3, 2026. Answer wording is Census’s.):

Employment

Census asked businesses that used AI in any business function how, in the last six months, the use of AI affected the business’s total employment. Among businesses that used AI in any business function, 95.7% of those with 1 to 4 employees and 94.8% of those with 250 or more answered no change. This is the businesses’ own report, not a measured change in jobs.

What the numbers mean for a small business using AI

This section and everything after it, up to Method, are interpretation and guidance from operating practice. They are not findings of this study.

Among businesses with 1 to 4 employees that used AI in any business function, 65.5% said they made no changes in order to use it, and 15.0% said they developed new workflows. In practice, that pattern often means AI is being used as a tool on top of existing work rather than as a reason to change how the work is done. The data cannot show whether that use is productive. It does show that most of the smallest AI users did not report adjusting anything around it.

The training gap reads the same way. Only 13.4% of AI-using businesses with 1 to 4 employees reported training current staff, against 31.4% of AI-using businesses with 250 or more employees. In a business of one to four people, the person using AI is often the owner, so “training” can look like nothing at all. The practical question is not whether a formal course happened. It is whether anyone has written down how the tool is used, so the use is repeatable when someone else does the work.

Among businesses with 1 to 4 employees that were not planning to use AI in the next six months, 63.3% said it is not applicable to their business, and 21.2% cited a lack of knowledge about AI’s capabilities. Those two answers are not exclusive, and the second is worth noticing. A business that has not looked closely at what the tools do is in a weak position to decide they do not apply.

How to handle it: treat AI as a process change, not a purchase

The steps below are an operating recommendation. This study did not test them, and Census does not endorse them.

1. Pick one workflow, not the whole business. Choose a task that repeats every week, such as drafting quotes, answering routine customer questions, or reconciling a report. Write down how it is done today, step by step, and how long it takes.

2. Change the workflow on purpose. Decide which steps the tool takes over, which steps a person still checks, and what the finished output must look like. That written version is the new standard procedure. Without it, each use of the tool is less likely to leave a repeatable process behind.

3. Train whoever does the work, including the owner. Training in a small business can be a one-page procedure and a short walkthrough. The test is simple: someone other than the person who set it up can follow the procedure and get the same quality of result.

4. Measure before and after. Time per task, error rate, or turnaround are enough. If the numbers do not move after a month, the tool is not yet changing the work, and the procedure needs another look.

5. For businesses that decided AI does not apply, test that conclusion once. List the tasks that consume the most hours each week and check each against what current tools can do. The conclusion may hold. It is still a decision worth making with the task list in hand rather than by default.

What this data does not tell you

• It does not show that using AI, changing workflows, or training staff improves results.
• Answers are the businesses’ own reports, collected from November 17, 2025 to February 8, 2026.
• The comparisons cover businesses with 1 to 4 and 250 or more employees. The size classes between them are shown only where stated.
• Respondents could choose more than one answer, so shares do not add to 100%.

Using AI and changing the work around it are different decisions. Among AI-using businesses with 1 to 4 employees, 65.5% reported making no changes in order to use AI. As operating guidance rather than a finding: the change in the work, written down and repeatable, is where an owner should expect the value to come from. The survey does not measure whether it does.

Method

Already published by the agency: Census has published AI use by company size and reported that larger companies use AI more. New here: two pre-registered comparisons by company size, one among businesses using AI and one among businesses not planning to use it, with descriptive figures on changes made and reasons given.

Source. U.S. Census Bureau, Business Trends and Outlook Survey, AI supplement 2025 to 2026, file AI_Supplement_Table_2026.xlsx released June 18, 2026 (Census tip sheet CB26-TPS.38), sheets “Employment Response Estimates” and “Employment Standard Errors”, downloaded October 3, 2026. Size classes are 1 to 4, 5 to 9, 10 to 19, 20 to 49, 50 to 99, 100 to 249, and 250 or more employees. The comparisons use the smallest and largest classes. The middle classes appear once, as a descriptive range.

Who answered which question. Use of AI in the last two weeks was asked of all employer businesses. Changes made and employment effects were asked of businesses that used AI in any business function. Reasons for not planning to use AI were asked of businesses not planning to use it in the next six months.

Statistics. Each difference uses the two published standard errors. Census does not publish the covariance needed for these contrasts, so the approximate intervals assume zero covariance between size-class estimates, with z equal to the difference divided by the square root of the sum of the squared standard errors. Intervals are approximate 95% intervals. No multiplicity adjustment was made. Answers Census marked as suppressed were not used.

Protocol. The analysis plan, including the two primary comparisons, was fixed in a dated protocol before any size-class estimate was computed. Later amendments are listed in it: https://kamyarshah.com/public-data-protocol/

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

Each year, the Bureau of Labor Statistics publishes National Compensation Survey estimates of the share of private industry workers with access to common benefits. The reference period is March. Estimates are by the size of the establishment, meaning the workplace, where they work. BLS defines who is in scope this way: “private industry excludes workers in private households, the self-employed, workers who set their own pay (e.g., proprietors, owners, major stockholders, and partners in unincorporated firms), and family members paid token wages.” So the figures exclude the self-employed and workers who set their own pay, including the listed owner categories.

The four benefits shown were fixed in the study protocol before any figure was computed, and no other benefit was examined. The March 2026 estimates show lower access at establishments with fewer than 50 workers for each of the four.

Share of private industry workers with access, March 2026 (source: U.S. Bureau of Labor Statistics, National Compensation Survey, Employee Benefits flat files, series NBU20700000000000233087 and NBU20700000000000733087 (paid sick leave), NBU21500000000000228178 and NBU21500000000000728178 (medical care), NBU29000000000000228319 and NBU29000000000000728319 (all retirement benefits), NBU20500000000000233030 and NBU20500000000000733030 (paid vacation), downloaded October 3, 2026):

Standard errors are from the BLS table set “Employee Benefits in the United States, Standard Errors for News Release”, March 2026 (Table 6 for paid leave, Table 1 for retirement, Table 2 for medical care), private industry access columns. Only the paid sick leave interval belongs to the registered comparison. The other three intervals are descriptive.

The pre-registered comparison: paid sick leave

The study set one primary comparison before any figures were computed: paid sick leave. In March 2026, 75% of private industry workers at establishments with fewer than 50 workers had access to paid sick leave, against 91% of private industry workers at establishments with 500 or more workers. The difference is 16 percentage points, with an approximate 95% interval of 12.8 to 19.2 points, so the interval excludes zero.

Medical care and retirement, described

The remaining comparisons are descriptive. Among private industry workers at establishments with fewer than 50 workers, 55% had access to medical care benefits and 55% had access to retirement benefits. Among private industry workers at establishments with 500 or more workers, the figures were 90% and 91%.

These are shares of workers, not shares of employers. Access means the benefit is available to that worker. For paid leave, BLS publishes access only. For medical care and retirement, access is not enrollment. The intervals use the BLS published standard errors for each estimate and assume zero covariance between the two size classes, which BLS does not publish, so they are approximate. The published access rates are rounded to whole percentages, so each difference can be off by up to one point. The standard errors measure sampling error only, not errors from other sources such as nonresponse. No multiplicity adjustment was made.

Size classes differ in industry and occupation mix. Differences by size reflect both. Nothing here isolates an effect of size. Size classes can also differ in other worker characteristics that BLS reports separately, such as full-time or part-time status.

What the numbers mean for a small employer

This section and everything after it, up to Method, are interpretation and guidance from operating practice. They are not findings of this study.

The gap is widest for retirement and medical care benefits. Among private industry workers at establishments with fewer than 50 workers, 55% had access to medical care benefits and 55% to retirement benefits, against 90% and 91% at establishments with 500 or more. For paid sick leave and paid vacation, the shares among the same private industry workers were 75% and 71%, against 91% for each at establishments with 500 or more workers. At the group level, workers at establishments with 500 or more workers had higher access to each of the four benefits. These separate rates do not show the package any one worker receives, but they describe the market a small employer hires into.

The data cannot say why the gap exists, and it is not a measure of what any one employer should offer. It does show that, benefit by benefit, 90% to 91% of private industry workers at establishments with 500 or more workers had access, measured separately for each of the four. That is a benchmark of worker access, not a count of large workplaces offering each benefit.

How to handle it: close the gap in the order that costs least

The order below is an operating recommendation. This study did not test it, and BLS does not endorse it. Costs, tax treatment, and legal requirements vary by state and by employer, so each step should be checked with a benefits broker, an accountant, or employment counsel before it is adopted.

1. Write down what is already offered. Many small employers provide some time off or flexibility informally, and workers cannot count what they do not know about. A one-page benefits summary, given at hiring and reviewed once a year, costs little and turns informal practice into a stated policy. It also makes the next decisions easier, because the starting point is on paper.

2. Check the sick leave rules that already apply. Several states and cities require employers to provide paid sick leave, often with rules that differ by employer size. Before designing a policy, confirm what is already required where the business operates. For employers already covered by a local rule, a written policy that meets it is a requirement before it is a benefit decision.

3. Look at retirement options available to small employers. Some states run retirement savings programs that employers can join, and plan types such as the SIMPLE IRA are available to eligible employers with 100 or fewer employees. Each option has its own rules on employer contributions and administration. A SIMPLE IRA, for example, requires an employer contribution. Whether a given arrangement counts as retirement access in the BLS survey is a question for BLS definitions, not an assumption, so the goal here is a usable plan for workers rather than a change in a statistic.

4. Treat medical coverage as a budget decision, not a yes or no. Medical care has one of the two widest gaps shown here, and it is the decision with the most options to compare. Options for small employers include group plans and reimbursement arrangements built for employers with fewer than 50 full-time equivalent employees. Which option fits depends on headcount, budget, and state rules, which is exactly why this step belongs with a broker rather than a template.

The principle behind the order is simple. Start with what can be fixed by writing it down, then what the law may already require, then what can be offered at low cost, and leave the most expensive decision for when the cheaper ones are in place.

What this data does not tell you

• It does not show why access is lower at small workplaces, or that offering a benefit improves hiring or retention.
• The figures are shares of workers, not shares of employers, and they cover private industry only.
• Access is not enrollment for medical care and retirement, and access says nothing about the quality or cost of a benefit.
• Establishment size is not firm size. A small workplace can belong to a large company.

In these March 2026 estimates, workers at establishments with fewer than 50 workers had lower access to all four benefits than workers at establishments with 500 or more. As operating guidance rather than a finding: treat benefits as part of the operating system of the business, and close the distance through a sequence of modest, written decisions rather than one large one.

Method

Already published by the agency: BLS publishes these access rates and their standard errors by establishment size in its annual Employee Benefits in the United States release. New here: the four benefits side by side, with the differences and approximate intervals stated.

Source. U.S. Bureau of Labor Statistics, National Compensation Survey, Employee Benefits flat files, access rates for private industry workers at establishments with less than 50 workers and with 500 workers or more, reference period March 2026, downloaded October 3, 2026. Series were selected by exact title. The retirement series is titled “access to all retirement benefits”.

Check. The flat-file class “less than 50 workers” corresponds to the published table class “1 to 49 workers”. For private industry workers in March 2026, all eight figures match the 2026 Employee Benefits release: paid sick leave and paid vacation in Table 6, retirement in Table 1, and medical care in Table 2.

Protocol. The analysis plan, including paid sick leave as the single primary comparison, was fixed in a dated protocol before any size-class figure was computed. Later amendments are listed in it: https://kamyarshah.com/public-data-protocol/

Limits. For medical care and retirement, access is not enrollment. Access says nothing about the quality or cost of a benefit. Establishments are single workplaces, not companies. A workplace with fewer than 50 workers may be part of a larger firm. A firm has at least as many workers as its largest workplace. That does not put either column into a firm size class, and this article does not call either class small firms. Establishment size is not firm size.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

In 2015 and again in 2021, the US Census Bureau asked manufacturing establishments in its Annual Survey of Manufactures mail sample how they run. Do they track key performance indicators? How often does anyone review them? What happens when a production problem appears, and what happens to a manager who keeps under-performing? The answers become a structured management score from 0 to 1, where 1 means the most formal, frequent, and explicit practices.

This study asked one question of that data: did the gap between the smallest and the largest plants change between the two surveys? The population is narrow by design. These are manufacturing establishments, meaning individual plants, not firms, and not retail, construction, or service businesses.

The answer: no clear change

In 2021, plants with 1 to 4 employees scored 0.402 and plants with 2,500 or more scored 0.742, a gap of 0.34 on the 0 to 1 scale (approximate 95% interval 0.31 to 0.37). In 2015, the same size classes scored 0.454 and 0.764, a gap of 0.31 (approximate 95% interval 0.28 to 0.34).

The point estimate of the gap was 0.03 higher in 2021. The approximate 95% interval for that change runs from negative 0.01 to positive 0.07, so it includes zero. The interval uses the registered formula, which assumes zero covariance within and across years. Because Census publishes a sampling standard error of 0 for plants with 2,500 or more employees in both years, its nonzero width comes from the 1 to 4 estimates in 2015 and 2021. This was the study’s single pre-registered comparison, and the data does not show a clear change in the gap. That is not evidence that the gap stayed the same. The sampling interval runs up to 0.07 on the 0 to 1 scale. This interval cannot separate a zero change from a change of that size, and it is not the full uncertainty of the gap.

Scores by plant size on the 0 to 1 scale. Point estimates are Census figures. Intervals are this study’s approximate 95% sampling intervals from the published standard errors, with “(Z)” standard errors set to 0.0005 (source: U.S. Census Bureau, MOPS public tables 2015 and 2021, sheet “Employment”, downloaded October 3, 2026):

Note: Census publishes a sampling standard error of 0 for plants with 2,500 or more employees, so the 95% sampling interval for that class has zero width. That does not mean the score is known exactly.

Census has reported that, in its MOPS tables for manufacturing establishments, larger plants generally show more structured management. In these tables the published point estimates increase with each size class from 20 to 49 employees upward in both years. Below that it is not uniform: in 2015, plants with 1 to 4 employees scored above plants with 5 to 9.

Exploratory, not pre-registered. Reason included: so the class table is not read selectively. Point estimates were lower in 2021 than in 2015 in every size class, and the score for manufacturing establishments in the mail sample, all employment classes combined, fell from 0.549 to 0.532. The 1 to 4 class fell from 0.454 to 0.402 and the 2,500-or-more class fell from 0.764 to 0.742. Those level shifts were not pre-registered, so this study does not test or interpret them. What is new here is one pre-registered contrast, the change in the gap between the smallest and largest classes, and that contrast shows no clear change.

Size classes differ in industry and occupation mix. Differences by size reflect both. Nothing here isolates an effect of size.

The practices plants most often reported lacking or rarely using

The 2021 survey also reports how manufacturing establishments in the Annual Survey of Manufactures mail sample, all sizes together, answered each practice question. Under a rule set before the data was analyzed, this study names the three practices with the highest share choosing the registered absence or low-use answer. Census reports these as a percent of respondents:

• 55.1% of responding establishments said under-performing managers were rarely or never reassigned or dismissed.
• 50.1% said non-managers received no performance bonuses.
• 48.6% said they did not have any display boards.

These shares cover all plant sizes together. The public tables used here do not report them by size, so this study does not claim that small plants skip these practices more often. The full list of 12 absence or low-use shares appears at the end of this article.

What the numbers mean for a plant owner

This section and everything after it, up to Method, are interpretation and guidance drawn from the figures above and from operating practice. They are not findings of this study.

The estimated gap was about a third of the scale in both surveys: 0.31 in 2015 and 0.34 in 2021. The approximate interval for the change includes zero, so the data do not show the gap clearly closing or widening. The two surveys are snapshots of different samples, so they cannot show how any single plant changes as it grows.

The response shares show which listed practices plants most often reported as absent or rarely used. They do not rank priorities, but they are a reasonable place for an owner to start looking. Among responding manufacturing establishments in the 2021 mail sample, all sizes together, 9.5% reported having no key performance indicators. In the same group, 26.0% said non-managers never review them, 48.6% had no display boards, and 55.1% rarely or never reassigned or dismissed an under-performing manager. These are separate questions, not one measure, and the shares do not show whether the same plants gave these answers. Read side by side, reporting at least one indicator was much more common than reporting display boards, non-manager review, or prompt action on an under-performing manager.

From operating practice rather than from this data: in a plant of five or fifteen people, the owner often carries monitoring, targets, and consequences personally. That arrangement can work while the owner is present. If those routines live only in the owner’s habits, the plant can lose them whenever the owner steps away. Writing them down makes the standard less dependent on one person.

How to handle it: build structure in sequence

An analysis of earlier MOPS data groups the survey’s management questions into three areas: monitoring, targets, and incentives. That analysis, by Bloom, Brynjolfsson, Foster, Jarmin, Patnaik, Saporta-Eksten, and Van Reenen, published in the American Economic Review in 2019, found that more structured management was associated with higher productivity. That is an association, not proof that adopting a practice raises output. The order below is an operating recommendation. This study did not test it, and Census does not endorse it.

1. Monitoring first. Choose a short list of indicators, review them on a fixed weekly rhythm, and put them where the floor can see them. The answer “3-9 key performance indicators” was chosen by 48.5% of responding plants in that same 2021 group, more than any other answer. Write one standard response for production problems: fix it, find the cause, and record what changes so it does not happen again. The survey’s most structured answer adds a continuous improvement process that looks for problems before they occur, and 55.5% of responding plants chose it.

2. Targets second. A target only works if people know it exists. The survey’s widest awareness answer is that all managers and most production workers know the targets, and 46.9% of responding plants chose it. Set a mix of short-term and longer-term production targets, then post them beside the indicators so progress and target sit in one view.

3. Incentives last. Write down what happens when performance falls short and how long the review takes. The survey’s most structured answer is reassignment or dismissal within six months of identifying under-performance, and 27.2% of responding plants chose it for managers. That share is descriptive. It does not establish that six months is the right window, and any personnel rule should be checked against employment law and the plant’s own circumstances. A written process can still protect the team as much as the business, because it replaces surprise with known expectations. Tie bonuses to measures people can already see. A bonus paid on numbers nobody reviews risks rewarding luck rather than performance.

The reasoning behind the order is practical. Targets need indicators to measure against, and incentives need targets to be fair. A plant that installs bonuses before it has visible indicators has little basis for explaining who earned them.

What this data does not tell you

• It does not show that any practice causes better results at a particular plant.
• It covers manufacturing establishments in the Census mail sample only, not retail, construction, or service businesses.
• The practice shares cover all plant sizes together. The public tables do not show them by size, so they do not show which practices are rarer at small plants.
• Scores and answers are self-reported by the plants.

In both surveys, the smallest and largest size classes were separated by about a third of the scale. The practices the survey describes are concrete enough to install one at a time, which is how structure is usually built: one routine, reviewed until it holds, and then the next.

Method

Already published by the agency: Census has published structured management scores by plant size, state, industry, and plant age for 2015 and 2021, and has reported that, among these manufacturing establishments, larger plants generally show more structured management. New here: the change in the gap between the smallest and largest size classes from 2015 to 2021, with a pre-registered interval.

Source. U.S. Census Bureau, Management and Organizational Practices Survey (MOPS), public data tables for 2015 and 2021, sheets “Employment” and “Response Shares”, downloaded October 3, 2026. Census describes the source this way: “The Management and Organizational Practices Survey statistics are collected from the ASM mail panel only.” These tables cover reference years 2015 and 2021. Census notes that MOPS began as a pilot in 2010. That wave is not used here, and this study makes no claim about it. Data as of those releases.

Measure. Each plant’s score is the unweighted average of 16 practice questions, each normalized from 0 (least structured) to 1 (most structured). Table scores are the averages Census publishes for each size class. The size classes and the scoring rule are the same in both years.

Statistics. The primary comparison is the 2021 gap between plants with 2,500 or more employees and plants with 1 to 4 employees, minus the same gap in 2015. Its standard error combines the four published standard errors under the registered assumption of zero covariance among the estimates within and across years. Census publishes a standard error of 0 for the 2,500-or-more class in both years, and that value is used as published, so the calculated interval comes entirely from the published standard errors for the 1 to 4 class. It is an approximate 95% sampling interval under the registered zero-covariance assumption. Census does not publish the covariances, so the appropriate interval could be wider or narrower. The published standard errors measure sampling error only and do not capture nonresponse or measurement error. Where Census reports a standard error as “(Z)”, meaning above zero but below 0.0005, this study used 0.0005. Intervals are approximate 95% intervals. No multiplicity adjustment was made. All other figures are descriptive.

Protocol. The analysis plan, including the single primary comparison and the call-out rule, was fixed in a dated protocol before any size-class estimate was computed. Later amendments are listed in it: https://kamyarshah.com/public-data-protocol/

Limits. Manufacturing establishments in the Annual Survey of Manufactures mail sample only. Establishments, not firms. Survey self-reports. The 2015 and 2021 surveys rest on different Annual Survey of Manufactures samples, because Census redraws that sample every five years. These are two cross-sections, not a panel, and a plant can change size class between them.

2021 absence or low-use shares, all sizes, percent of responding manufacturing establishments in the mail sample (source: U.S. Census Bureau, MOPS 2021 public tables, sheet “Response Shares”, downloaded October 3, 2026): under-performing managers rarely or never reassigned or dismissed, 55.1%. No performance bonuses for non-managers, 50.1%. No display boards, 48.6%. Under-performing non-managers rarely or never reassigned or dismissed, 44.6%. No performance bonuses for managers, 38.6%. Key performance indicators never reviewed by non-managers, 26.0%. Never reviewed by managers, 9.8%. No production targets, 9.7%. No key performance indicators, 9.5%. No action taken on production problems, 1.8%. Non-manager promotions based mainly on factors other than performance and ability, 1.7%. Manager promotions on the same basis, 1.4%. The full answer distribution for all 16 questions is in the Census table.

The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.

Questions about this analysis can be sent through https://kamyarshah.com/contact/.

The Hiring Bottleneck Is Not the Hire

The NFIB August survey shows a net 17 percent of owners plan to create new jobs, down 3 points from July. Of the 56 percent hiring or trying to hire, 82 percent found few or no qualified applicants. The constraint is not the candidate pool. The constraint is that a hire fails when the founder never wrote down what success looks like.

A scarce hire becomes expensive when onboarding lives in the founder’s head. The new person asks questions, the founder answers them, and the cycle repeats until the founder realizes the hire is not working. The hire was not given a system. That hire was given access to the founder, and access does not scale.

A hiring pause is the window to build the process. The process names the role, the first thirty days of work, and the owner of each step. It states what the new hire must do and when. It tells the manager how to confirm each step happened.

The Anti-Pattern Is Paperwork Disguised as Process

Most founder-led companies treat onboarding as orientation and paperwork. A new hire signs forms, attends a meeting, and receives login credentials. Founders assume the hire will figure out the rest. That hire does not figure it out, and the founder blames the hire.

Paperwork is compliance. Orientation is a welcome. Neither is a process. A process assigns ownership, defines sequence, and measures completion.

Paperwork tells the hire what the company requires. A process tells the hire what the hire must do to succeed. Chaos is disguised as flexibility. Founders believe that every hire is different and that a rigid process will not fit.

Every hire needs the same structure in the first thirty days. Structure frees the hire to ask the right questions instead of guessing what the founder wants. Scrambling to answer the same questions for every new person is waste that conceals the absence of a system.

Diagnose the Gap Before Writing the Process

Start by listing every task the last three hires completed in their first thirty days. Include tasks the founder handled directly and tasks the hire discovered alone. Classify each task by type: compliance, access, training, output. Compliance is paperwork and legal requirements.

Access is systems, tools, and introductions. Training is instruction on how the company operates. Output is the first deliverable the hire must produce. If compliance and access fill the first week while training and output have no owner or deadline, the list shows the gap on one page.

Assign an owner to each task. An owner is not the founder. An owner is the person who confirms the task closed. If no one other than the founder can confirm a task, the task is a founder bottleneck and must be reassigned before the next hire.

A Calm Rule Separates Process from Reaction

A process is not a checklist. A checklist is a list of tasks. One process is a sequence with owners, deadlines, and confirmation steps. Checklists assume the hire knows what to do.

A process assumes the hire does not and removes ambiguity. Build the process using a framework that names the first thirty days as four distinct phases. The first week covers access and compliance. The second week covers training on systems and tools.

That third week covers shadowing and observation. The fourth week covers the first output under supervision. Each phase has a clear deliverable and a named owner who confirms it. This framework removes the founder from the first seven days entirely.

Founders do not answer questions about login credentials, benefits, or where to find the supply closet. The process assigns those questions to the owner of access and compliance. Founders enter on day fifteen, after the hire has context.

The Framework Fix Is a Thirty-Day Balanced Scorecard

A thirty-day onboarding balanced scorecard measures whether each step closed on time. It lists every task, the owner, the deadline, and the confirmation method. Managers review the scorecard at the end of each week and flag any task that missed its deadline. An audit is not a performance review of the hire.

An audit is a diagnostic of the process. Run it on the last three hires. For each task, record the owner, the deadline, the close date, and who confirmed it. Engagements that start with this audit usually find the same gap first: nobody owned training.

Training is where the gap often hides. The founder assumes the manager will handle it. The manager assumes the founder will handle it. A step with two assumed owners has no owner, and the scorecard shows it never closed.

Assign training to one manager, document the training sequence using a jobs-to-be-done framework, and set a deadline for each step. The next hire then follows a written sequence instead of guessing. The process protects the hire from the chaos that makes a scarce hire fail.

The Purpose Is to Protect Human Capital

A failed hire is a cost the owner already pays. Costs include recruiting, lost productivity, and the time the founder spent answering questions. Costs also include the damage to the team. A failed hire signals to the team that the company does not know how to integrate new people.

Teams become skeptical of the next hire before the next hire starts. Onboarding processes serve the hire and the team. Hires receive clarity. Teams receive proof that the company has a system.

Founders receive time back. Processes are not overhead. Processes are the mechanism that turns a scarce hire into a productive contributor. Kamyar Shah, working as a fractional COO, builds onboarding processes for founder-led companies by starting with the last three hires and auditing what actually happened.

Audits name the gaps. Processes fill the gaps. Scorecards measure whether the gaps stay filled. This work is operational excellence applied to human capital.

Measure Success with Unit Economics

Onboarding processes work when the next hire reaches full productivity faster than the last hire. Full productivity is defined as the hire producing the output the role requires without supervision. A metric is time to first unsupervised output. A target is to cut that time in half.

Track three numbers: time to first output, founder hours spent in the first thirty days, and retention rate at ninety days. If the process works, all three numbers improve. If the numbers do not improve, the process has a gap. Gaps are either missing ownership, missing sequence, or missing confirmation.

A documented process moves the questions of the first thirty days from the founder to named owners. That shift frees the founder to focus on the next bottleneck. It also signals to the hire that the company has systems. Hires trust the company more when the company demonstrates it can operate without the founder in the room.

Porter’s five forces analysis applies to talent markets just as it applies to product markets. Supplier power rises when candidates are scarce. Buyer power falls when owners compete for the same small pool. Onboarding processes shift the competitive dynamic by reducing the risk a candidate perceives in accepting an offer.

The Hiring Pause Is the Window

A hiring pause is not a failure. A hiring pause is the opportunity to build the system that makes the next hire succeed. Federal Reserve action raised its target range to 3.75 percent to 4.00 percent on September 16. A hire financed with credit now costs more than it did a quarter ago.

BLS August data shows average hourly earnings of $37.75, up 3.1 percent over the year. Payrolls rose 162,000 against an average monthly gain of 31,000 over the prior twelve months. Hiring is scarce and expensive. The cost of a failed hire just increased.

NFIB survey data shows 35 percent of owners report job openings they cannot fill. Openings remain unfilled because the process to integrate a hire does not exist. Owners wait for the perfect candidate. Perfect candidates do not exist.

Candidates who succeed are candidates who receive a system. Build the onboarding process now. Document the first thirty days using business process documentation principles. Assign owners to each step and measure time to first output using VRIO analysis to confirm each task adds value.

Every System Built Now Compounds Later

Onboarding processes are not a one-time fix. Processes are structural investments applied to the most expensive recurring event in a founder-led company of 8 to 200 people. Every hire that succeeds because of the process proves the process works. Every hire that fails because the process has a gap reveals where to refine the process.

Processes accumulate value over time. The first hire validates the structure. The second hire refines the sequence. That third hire confirms the owners.

By the fourth hire, the process runs without the founder. Founders built a system that scales human capital. Signaling theory explains why a documented process attracts better candidates: it demonstrates that the company operates as a system rather than as a collection of individual relationships. Candidates assess whether the company can support their growth.

Processes answer that question before the offer is made. Ultimately, the firms that grow past the founder are the firms that document how work happens before the work happens. Onboarding processes are the first test of whether the company can operate as a system. Build the process before the next hire and protect both the hire and the company through shared alignment on what happens in the first thirty days.

The Bottleneck Appeared When the Escape Valve Closed

On September 16 the Federal Reserve raised the federal funds target range to 3.75 percent to 4.00 percent. The 10-year Treasury sits at 5.00 percent, and borrowing costs for small businesses have risen with it. For a founder-led business the operational problem used to be solved by hiring another person or financing another tool, and both of those now require a defensible return.

Meanwhile the August payroll headline of 162,000 sits on a trailing twelve month average of 31,000, which means the aggregate hiring market is far weaker than the headline suggests. Adding fixed payroll before the process exists is the more expensive error. The NFIB Uncertainty Index at 89 against a historical average of 68 means owners defer every decision until the picture clears.

That deferral creates a secondary bottleneck where the business cannot scale because no one will commit to a hire, a lease, or a vendor contract. The constraint is not capital availability or market demand. The constraint is the absence of a system that turns an uncertain environment into a measured one.

That anti-pattern is hiring another person to solve a process gap. A new hire inherits the same chaos the last person struggled with, performs inconsistently, and eventually leaves. The founder concludes the talent was wrong when the real issue was structural.

What the Engagement Examines

A business systems consultant starts with a diagnostic rather than a solution. The first step is mapping where decisions actually happen, not where the organizational chart says they should happen. Most mid-market companies discover that critical operational choices live in email threads, Slack channels, or the founder’s memory rather than in a documented framework.

The second examination is the feedback loop between action and result. In a reactive organization the owner learns about a problem weeks after it occurred, when the damage is already structural. The consultant measures the time between an operational event and the moment the owner receives usable data about that event.

Third comes process documentation, not as a compliance exercise but as a transfer mechanism. The consultant interviews the people who currently execute critical tasks and extracts the decision logic they use. Captured logic becomes a written procedure that a new hire can follow without requiring the founder to train them personally.

The fourth diagnostic is resource allocation, examined through unit economics rather than budget categories. A business systems consultant applies value-based pricing and theory of constraints to identify where the firm spends time or money on activities that do not move the bottleneck. The output is a reallocation plan that increases output without increasing fixed costs.

What the Engagement Produces

The deliverable is not a software implementation. A business systems consultant produces written standard operating procedures, decision frameworks, and measurement architectures. Each SOP describes a repeatable task, the decision points inside that task, and the criteria for choosing between options.

The second output is a balanced scorecard that tracks operational metrics rather than financial lag indicators. A scorecard measures leading indicators like cycle time, error rate, and capacity utilization. These metrics give the owner a diagnostic view of the business rather than a rearview mirror.

The third deliverable is a RACI matrix that assigns accountability for every critical decision. It eliminates the ambiguity that causes delays when no one knows who has authority to approve a vendor, adjust a price, or escalate a customer issue. The consultant interviews stakeholders, identifies decision overlap, and produces a grid that makes authority explicit.

Fourth comes a process improvement roadmap that prioritizes which systems to build next. Sequencing runs by impact and dependency, not by ease or preference. The consultant applies the theory of constraints to identify the one process that, if fixed, will unlock the next layer of growth.

How It Differs from a Technology Project

A technology project assumes the problem is tool deficiency and the solution is a platform. A business systems consultant assumes the problem is process chaos and the solution is documentation.

Consider a mid-market services firm that believed it needed a CRM to fix inconsistent customer follow-up. The real issue was that no one had defined what follow-up meant, who owned it, or when it should happen. Installing software without answering those questions simply automated the chaos.

The difference is diagnostic sequence. A technology vendor starts with the platform and retrofits the process to match the software’s capabilities. A business systems consultant starts with the process, documents it, and only then evaluates whether a tool adds value. The same sequence applies to AI, as the guide on how to hire an AI consultant for small business explains.

The second distinction is ownership of the outcome. A software implementation is complete when the platform is configured and the training is delivered. A systems engagement is complete when the business can execute the process without the consultant present. An AI implementation consultant engagement applies the same exit test: the client team runs the system without the consultant in the loop. When the system does need new software connections, the guide to AI integration services explains how to scope that work before comparing quotes.

When an Owner Builds a System Instead of Hiring Another Person

The decision is arithmetic rather than preference. A full-time operations hire takes months to become productive and requires the founder to train them on undocumented processes. If the business lacks written procedures the new hire will spend the first quarter asking questions and the second quarter making the same mistakes the last person made.

A business systems engagement produces documentation that makes the eventual hire successful faster. The engagement examines the founder bottleneck, extracts the decision logic, and converts it into written procedures. When the business does hire, the new person inherits a system rather than chaos.

The second factor is flexibility. A full-time hire is a fixed cost that persists regardless of revenue. A systems engagement is a project cost that ends when the deliverables are complete.

The third consideration is scope. A full-time operations hire works on the problems the founder assigns. A business systems consultant works on the problems the diagnostic reveals.

How Systems Thinking Protects Human Capital

The purpose of business process documentation is not efficiency for its own sake. The purpose is to protect the team from the chaos that makes good people look unreliable. Most mid-market companies do not have a talent problem.

When a business lacks documented procedures every task becomes a judgment call, and judgment calls create variance. Variance burns out high performers because they spend energy navigating ambiguity rather than executing work. The result is churn, and churn is expensive.

A business systems consultant reduces variance by converting judgment calls into documented decisions. The team gains clarity, execution improves, and the business retains talent longer. The system serves the people rather than the other way around.

A business with written SOPs trains new hires through documentation, which scales without consuming senior capacity. The team grows without the founder becoming a full-time trainer. A documented process gives the team permission to act within defined boundaries.

How the Engagement Builds Shared Confidence

A founder working with a business systems consultant enters a collaborative diagnostic rather than a prescriptive engagement. The consultant asks where decisions happen, who owns execution, and what data the owner needs to make confident choices. Shared examination produces alignment on which process to fix first.

Consider a mid-market firm where the sales team and the operations team blamed each other for missed delivery dates. The consultant facilitated a session where both teams mapped the handoff process together. The real bottleneck was not sales overpromising or operations underdelivering but a missing step where no one confirmed capacity before accepting the order.

That collaborative approach surfaces problems the owner could not see from inside the business. The consultant brings an external perspective that names the friction without assigning blame. The team gains confidence because the solution comes from their own process map rather than from an outside mandate.

That engagement produces a coalition around operational improvement. Stakeholders who resisted change because they did not trust the diagnosis now support implementation because they participated in building the framework. The coalition persists after the consultant leaves because the system belongs to the team.

How the Engagement Differs from Full-Time Operations Leadership

A full-time Chief Operating Officer manages ongoing execution, attends leadership meetings, and owns departmental performance. One business systems consultant designs the frameworks and then transfers them to the internal team. The engagement is bounded by scope and timeline rather than an employment relationship.

A consultant does not manage people. The consultant builds the system that makes people manageable. This distinction matters when capital costs are high.

The second difference is speed. A full-time hire takes months to recruit, onboard, and integrate into the leadership team. A business systems consultant starts the diagnostic in the first week and delivers the first set of SOPs within a month.

The third distinction is the exit condition. A full-time operations leader stays as long as the business needs ongoing management. A business systems consultant leaves when the documentation is complete and the internal team can execute the process independently.

Why the Arithmetic Favors Systems Before Headcount

The Federal Reserve’s rate decision closed the escape valve that allowed businesses to solve operational problems by hiring or financing. At 10 percent to 15 percent the cost of capital makes every fixed expense decision a strategic one. A business that adds headcount before it documents the process pays twice, once for the salary and again for the inefficiency.

The employment data supports the same conclusion. A trailing twelve month average of 31,000 against a headline of 162,000 means the hiring market is weaker than it appears. Adding fixed payroll in a weak market is the higher risk.

The NFIB Uncertainty Index at 89 reflects the fact that owners do not know what to do next. A business systems consultant converts uncertainty into a measured decision by building the diagnostic frameworks that reveal which operational gap to fix first. Kamyar Shah structures engagements to transfer systems and operations leadership to the internal team rather than creating dependency on the consultant.

Every business ultimately scales through systems rather than heroics. Firms that build those systems now will compound the advantage when the market clears. The measured approach protects human capital and preserves flexibility when capital costs are high.

The Bottleneck Is Not the Thesis

Most integration failures do not trace to a weak acquisition thesis. They trace to a capacity constraint inside the acquiring company. The leadership team that built the original business now has to run that business and merge a second one at the same time.

Identifying what to combine is not the problem. The real problem is that one leadership team is doing both jobs with no additional hours in the day. Post-merger integration consulting exists to solve that sequencing problem.

The consultant classifies systems into three categories: conform immediately, conform later, or leave separate. Classification protects leadership capacity and keeps the base business stable. The alternative is attempting to conform everything at once, which is the most common way a defensible deal becomes an operational failure.

The Anti-Pattern of Simultaneous Conformity

The instinct after closing is to integrate as fast as possible. Acquirers treat speed as evidence of control and delay as a sign of weakness. The instinct produces chaos.

Forcing every system into alignment at the same time overloads the acquiring team, destabilizes both businesses, and conceals which changes actually matter. The anti-pattern shows up as scrambling. Finance tries to merge chart of accounts while sales is combining CRM records and operations is conforming vendor contracts.

Each one creates dependencies, and those dependencies churn through the same small group of decision makers. The result is not integration. The result is waste. A calm approach starts with diagnosis.

The consultant audits every major system and asks whether conforming it immediately serves the thesis or just satisfies the instinct for control. Most systems can wait. The few that cannot are the ones that either protect revenue or prevent regulatory exposure.

What the Consultant Actually Does

An engagement opens with a classification audit. The consultant maps every process, system, and structure across both companies. Mapping covers financial reporting, customer-facing workflows, vendor relationships, compliance obligations, and human capital management.

Each element is sorted into one of three categories using a RACI framework that assigns ownership and accountability. Conform immediately means the system creates regulatory exposure or revenue leakage if left separate. Conform later means the system matters but does not create immediate risk.

Leave separate means the system works well enough in its current form and conforming it consumes capacity without delivering value. The classification is not a recommendation. It is a decision framework that the acquiring leadership team can execute without external dependency.

The consultant then builds the sequencing plan. Sequencing names what gets conformed first, what waits, and what stays separate indefinitely.

The plan includes ownership assignments, dependency chains, and rollback protocols. It is designed to be executed by the acquiring team while that team continues to run the base business. Business consulting of this kind treats integration as a capacity problem before it is a combination problem.

The Governing Decision Is Sequencing

The most valuable output is not the integration plan. It is the decision about what not to integrate. Every system left separate frees capacity for the systems that actually matter.

Freed capacity protects the base business and reduces the risk that a good acquisition becomes a distraction. Consider a mid-market services firm that acquired a competitor with a different billing system. The instinct was to conform billing immediately because having two systems felt inefficient.

A balanced scorecard approach measured operational risk against leadership capacity, and the consultant classified billing as conform later. The acquiring company’s billing system worked, the acquired company’s billing system worked, and neither created revenue risk. Conforming billing immediately would have consumed leadership attention during the period when customer retention mattered most.

The firm conformed customer communication workflows and key account management instead. Those changes protected revenue and signaled continuity to shared customers. Billing was conformed later, after the customer base stabilized and the leadership team had capacity to manage the change.

Who Owns Integration Inside the Acquiring Company

At lower-middle-market scale, the acquiring company rarely has a corporate development function or an integration management office. The person who actually merges the two companies is usually the COO of the acquiring business. One person runs the base business, manages the integration, and does both jobs with the same team that existed before the deal closed.

The consultant does not take over integration. Instead, the consultant builds the system that lets the COO execute integration without abandoning the base business. The system includes the classification framework, the sequencing plan, and the decision protocols that keep small issues from escalating into leadership bottlenecks.

Ownership stays with the COO. The consultant provides the structure that makes ownership sustainable. Firms that treat integration as a project separate from operations create a false boundary.

Integration is operations. The same people, the same processes, and the same constraints apply. Integration differs by adding a second set of systems to the same leadership capacity.

The consultant’s job is to protect that capacity by preventing overload. Hiring an interim COO is one way to add capacity when the base business cannot afford to lose its existing operator to integration work.

The Difference Between Integration and Operational Due Diligence

Operational due diligence happens before the deal closes. Integration happens after. Due diligence identifies risks and validates the thesis using Porter’s five forces and SWOT analysis to assess competitive position.

Integration executes the thesis while managing those risks. The two activities use similar tools but serve different purposes. Due diligence asks whether the acquisition makes sense.

Integration asks how to combine the two businesses without breaking either one. Due diligence is diagnostic. Integration is prescriptive.

The consultant who performs due diligence may not be the same consultant who manages integration, but the diligence findings should inform the integration plan. Firms that stay diligence-ready and integration-proof build systems that survive both processes.

When Integration Planning Should Start

Integration planning should start during due diligence, not after closing. The classification audit can happen in parallel with financial and legal review. Starting early does not mean executing early.

It means the acquiring company closes with a sequencing plan already built. The plan protects the earliest weeks, which is when most integration failures begin. Waiting until after closing wastes time and creates pressure.

The leadership team feels obligated to show progress, so they start conforming systems without a plan. A reactive approach produces the simultaneous conformity anti-pattern. Early planning prevents that outcome.

The theory of constraints applies: the binding constraint in integration is leadership capacity, and planning protects capacity by preventing wasted motion. Consider a mid-market brand that acquired a small business and delayed integration planning until after closing.

The acquiring team spent the first month debating what to integrate and in what order. By the time they built a plan, they had already made changes that created dependency problems. Reversing those changes consumed more capacity than the original changes did.

The Structural Fix Is Classification

The framework that governs integration is the three-category classification: conform immediately, conform later, or leave separate. It applies to every system, process, and structure, using an Ishikawa diagram to trace dependencies and root causes. The classification criteria include regulatory risk, revenue risk, thesis alignment, and leadership capacity.

Regulatory risk acts as a gate. If the system creates risk, it gets conformed immediately regardless of capacity. The second criterion is a filter.

If conforming the system does not serve the thesis, it gets classified as leave separate. The third criterion is a sequencer. If the system matters but the team lacks capacity, it gets classified as conform later.

The output is a scorecard that names every system, its classification, its owner, and its sequencing. The consultant reviews the scorecard often at the start and less often once classifications settle. Changes to classification are allowed but require explicit justification.

The Human Capital Question

Integration is not a systems problem disguised as a people problem. It is a people problem that shows up in systems. The acquiring company inherits a second team, a second culture, and a second set of expectations.

Those people need to know what is changing, when it is changing, and why. The consultant’s job includes designing the communication plan that protects trust during conformity. The communication plan is built around the classification framework.

Employees are told which systems are conforming immediately, which are conforming later, and which are staying separate. Transparency reduces anxiety and prevents the rumor cycles that destabilize teams. The plan also names who owns each change and who employees should ask when they have questions.

Servant leadership during integration means protecting the team from chaos while the business changes around them. Firms that treat integration as a technical exercise and ignore the human capital dimension create churn. Employees leave, customers notice, and the thesis erodes.

The Proof Is in Sequencing Discipline

Organizations that execute integration well do not conform everything at once. They conform the systems that matter, sequence the systems that can wait, and leave separate the systems that work. Discipline protects capacity, reduces risk, and keeps the base business stable.

The measure of success is not how fast integration happens. The measure of success is whether the thesis survives execution. Kamyar Shah has observed that the firms which struggle most with integration are the ones that treat speed as the goal.

Speed is not the goal. Coherence is the goal. A slow integration that protects revenue and retains talent is better than a fast integration that destabilizes both.

The consultant’s job is to build the system that lets the acquiring company move at the right pace rather than the fastest pace. Engagements that succeed share a common pattern. They start with classification, they sequence based on capacity, and they communicate transparently.