Nothing in the government's own tables says nine in ten businesses fail. The BLS Business Employment Dynamics table for establishments opened in the year ended March 2019 prints 79.2 percent still reporting payroll one year later, 70.2 percent after two years, and 51.5 percent after five, and the SBA Office of Advocacy reports 49.2 percent of new employer establishments reaching five years across cohorts from 1994 to 2022. So the honest answer to the title's question has two halves. Half of these establishments are gone by year five, which is severe and does not need exaggerating, and the same release adds that 69.5 percent of the ones that reach year five also reach year ten. What predicts getting that far is in none of these tables. The official data describes who survives and measures nothing about why, so the practical reading has to come from what the numbers do show about cash, customers and price.
The number everyone quotes has no source behind it
The claim that nine in ten startups fail circulates in pitch decks, blog posts and conference talks, and Chicago Booth Review checked it against BLS data on new-business survival by industry. They found no industry, restaurants included, with a 90 percent failure rate. Official tables put first-year losses near one establishment in five and five-year losses near half.
The inflated version does work the real one cannot do. A 90 percent failure rate makes entrepreneurship look like a lottery, which flatters the people who made it and excuses the ones who did not. A curve that loses a fifth in year one and half by year five describes a process with steps, and an owner can act on a step.
Exhibit 1
No official table puts small-business failure at nine in ten
The curve, at the figures the table prints
Take the cohort of establishments opened in the year ended March 2019. One year on, 79.2 percent were still reporting payroll. Two years on, 70.2 percent. Five years on, 51.5 percent, and 46.8 percent at March 2025, six years in. For establishments opened in the year ended March 2015, the newest cohort with a full decade of data, 34.7 percent survive ten years.
Read the steps, not the endpoints. The drop from birth to year one is about 21 points. Year one to year two costs about 9. Year two to year five costs under 19 spread across three years. The first year is the steepest step by a wide margin, and after that the curve flattens.
The SBA's summary of many cohorts says the same thing: 67.7 percent of new employer establishments survived at least two years, 49.2 percent reached five, and 33.9 percent reached ten, averaged across cohorts from 1994 to 2022.
Exhibit 2
The first year is the steepest step and the curve flattens after it
A closing is not a failure, and an establishment is not your company
Two definitions spoil a lot of confident writing about these numbers.
BLS counts a closing when an establishment with payroll in one quarter reports none in the next, and it counts a death only when that zero lasts four more quarters, so deaths are a subset of closings. A closing can be a seasonal shutdown, a records change, or a business that stops reporting. The closing rate is not a failure rate.
The other definition is the unit. The survival table counts establishments, one physical location each, not companies. A firm with five offices is five establishments, and a chain opening a branch is an opening, not a new company. When someone says "half of small businesses close in five years" from these files, they have changed the unit, the count and the period at once. These are private-sector establishments on payroll, which tells you nothing directly about a two-person consultancy with no employees.
The curve has held through every cohort worth testing
A cohort curve is useful because it keeps its shape. BLS notes that survival rates follow a similar path regardless of birth year, which is why a curve describes the past and cannot forecast one firm.
The 2009 cohort, opened into a recession, survived one year at 76.7 percent and reached five years at 50.1 percent. Against the 2019 cohort at 79.2 and 51.5, opening in a downturn cost about two points at year one and about one and a half at year five, and that cohort's curve kept flattening after. The pandemic cohort, opened in the year ended March 2020, survived the first year at 80.9 percent and the second at 72.3 percent, both above the 2019 cohort's figures, and stands at 51.4 percent at five years against 51.5. That is parity, and the table prints no cause for the lift.
BLS reports that first-year survival was lowest for establishments born in 2001 and 2008, both recession years, and that most census divisions saw first-year survival rise for establishments born in 2020 before falling in 2021 and 2022. The table moves a little. It does not break.
Your sector moves the number by less than you think
For establishments opened in the year ended March 2019, five-year survival was 50.7 percent in professional, scientific, and technical services, 55.8 percent in construction, 57.4 percent in accommodation and food services, and 53.6 percent in health care and social assistance. First-year survival for the same cohorts was 78.0 percent in professional services, 78.7 in construction, 82.6 in accommodation and food services, and 87.7 in health care and social assistance.
The spread across those four sectors at five years is under seven points. Food service carries a reputation as a graveyard and prints the highest five-year survival of the group. Health care's advantage shows up early, at the first year, and narrows by the fifth.
BLS warns against judging an industry by survival rates alone. In its study of one national cohort of establishments born in 1998, which found 66 percent still existed two years after birth and 44 percent four years after with little variation by industry, the sector with the lowest survival had stronger employment growth than the sector with the highest. Dying slowly and growing fast are different outcomes, and the table reports one of them.
Exhibit 3
Sector moves five-year survival by under seven points
What the data says, and what it refuses to say
The official tables describe survival. They do not explain it.
Where a study does look for predictors, take it at its strength. A Census Bureau working paper by Brian Headd, published in 2001 for firms opened in the late 1980s and early 1990s, found the traits most associated with survival were having employees, more than $50,000 in starting capital, a college degree, and starting the business for personal reasons. Younger owners, no starting capital, and service or retail trade raised the likelihood of closure. The paper flagged itself as unofficial, and it excluded C corporations. Its own summary of survival is close to today's tables: about two-thirds of employer firms survived at least two years and about half survived at least four.
That is a 2001 working paper describing firms that opened three decades ago. A degree and $50,000 are not decisions available to you in October 2026. Treat Headd as a description of who tended to last, not a checklist.
Exhibit 4
The survival tables describe who lasts and do not measure why
The signals that are in the record for owners today
The Federal Reserve Banks' 2026 Small Business Credit Survey report on employer firms describes the ground small firms are standing on. Just under half were operating at a profit at the end of 2024, with 47 percent at a profit, 19 percent breaking even and 34 percent at a loss. Expectations for revenue and employment growth sat at their lowest since the 2020 survey, with the revenue expectations index down from 39 to 33, and 77 percent of firms reported rising costs.
When costs bite, owners reach for personal money first. The survey lists using personal funds, raising prices, using cash reserves and cutting costs as the most common responses, each near half of firms with financial challenges. This is a self-reported survey of firms with 1 to 499 employees, not a random sample, and it is a picture of conditions at the end of 2024 with no probability of failure attached.
On cash, the JPMorgan Chase Institute report "Cash is King" found the median small business held 27 cash buffer days, meaning enough in the bank to cover about 27 days of outflows with no incoming cash. That came from 470 million transactions at 597,000 small businesses between February and October 2015, so read it as an old snapshot of a real pattern, not a current national number. No page in this record measures the link between buffer days and survival, so the connection is our inference and yours to weigh: the first-year cliff is a cash event for most firms, and 27 days is thin cover for a slow month.
The Census Bureau's Business Dynamics Statistics for 2023 records 5,593,727 firms nationally, 699,980 establishment exits at a printed establishment exit rate of 9.396 percent, and 523,958 firm deaths, which came to about 9.4 percent of the firm count that year, a share we computed from Census counts rather than one Census publishes. The series covers employer businesses, so it says nothing about firms with no employees.
What a service business owner can take from this
Cut the folklore and three practical readings survive.
Cash months on hand is the variable the first-year cliff acts on. A firm that can pay outflows through a slow quarter makes the year-two column. The monthly bookkeeping rhythm that tells you what your monthly outflow actually is, in dollars, is the precondition for knowing how many months you hold, and most firms do not know it.
Concentration is the second. The tables show a steep first step and a flat middle: the danger is early and it is lumpy. Losing one large client in month eight is a different event from losing one of fifteen in year six. Quotes that get answered is the pipeline end of this, and a second reliable customer is worth more than a fifth project at the same margin.
Price is the third, and it is the one the Federal Reserve survey puts next to personal funds: raising prices is one of the most common things owners do under cost pressure, which suggests the pricing decision was late. The research on decision fatigue covers the same territory from the owner's side, including when to price a job, and finding the real bottleneck is how you check whether the constraint is demand at all. Gerber makes a related point about owners who are good at the work and bad at owning it, and his opening failure statistics do not survive the government table either, which is the reason to keep the diagnosis and drop the numbers.
The routine that keeps these three visible is small. A weekly review in 30 minutes puts cash, pipeline and price on one page every week, which is where a firm notices a drift before it is an emergency. Where the numbers are genuinely hard to read, management consulting covers turning the figures that bear on survival, months of cash, client concentration and gross margin per job, into decisions instead of reports.
One caveat on the calendar. BLS publishes one annual set of March-to-March survival figures each year, with the next release scheduled for October 28, 2026, and that release carries annual revisions. Expect the newest cohort and any revised history to move within weeks of it, and treat the figures here as current to the files BLS has published as of this writing.