Do 9 in 10 restaurants fail? What the data says

No. In the federal count of new food and lodging businesses, between 79 and 90 in every 100 were still open a year later, depending on the year they opened, and roughly half to 6 in 10 were open after five. That is as good as or better than all private-sector businesses, so the claim that 9 in 10 restaurants fail does not match the data. Restaurants do run on thin margins, though, so this week the useful checks are break-even and prime cost.

Where the 90 percent claim came from

Both a 2003 university release and a 2014 research draft tie it to television in 2003. Ohio State University's release on a Columbus restaurant study, dated September 7, 2003, says the figure was repeated on every episode of an NBC reality show, and the study's author, H. G. Parsa, reports finding no evidence for it in the research. Tian Luo, then at BLS, and the Berkeley statistician Philip Stark describe it as a television advertisement from that year, and the two accounts differ in detail. Either way, neither source points to a count of anything behind it.

Since then, owners have heard it from lenders, relatives and strangers. What follows is what government counts and published studies say instead, with who measured what, because the numbers differ by how they define “fail”.

What the federal numbers say about years 1, 5 and 10

The Bureau of Labor Statistics follows every new private-sector establishment, meaning a single location with employees, and counts how many are still open each March after it opened. Its survival table for accommodation and food services covers establishments that opened in each year ending in March from 1994 to 2025, with data through March 2025. Restaurants, bars and caterers sit in that group with hotels and other lodging, and BLS does not split restaurants out in these tables. Here is a selection of opening years, next to the same count for every private-sector industry.

Reading the whole table, not just these rows:

• After one year, between 79.0 and 90.3 percent of food and lodging establishments were still open, depending on the opening year (31 opening years, 1994 to 2024). No opening year had more than 21 in 100 close in year one.
• After five years, between 49.0 and 60.4 percent were open (27 opening years, 1994 to 2020), against 45.4 to 51.9 percent for all private industries.
• After ten years, between 34.6 and 41.6 percent were open (22 opening years, 1994 to 2015), against 32.4 to 35.3 percent for all industries.
• In 31 of 31 opening years the one-year figure beat the all-industry one, as did the five-year figure in 26 of 27 and the ten-year figure in 22 of 22. That count is mine, from reading the two tables side by side.

Two cautions. First, by year ten, only about 4 in 10 are still open, and for the all-industry count it is about 3 in 10. Starting a business is risky in every industry, not only this one. Second, a closing is not always a failure, which is the next section. Opening years from 2016 to 2020 had their first five years run through 2020 and 2021, and their five-year figures, 57.1 to 60.4 percent, sit at the top of the range. This table cannot tell you why.

A closing is not always a failure

BLS defines a closing by employment: a location that had paid employees and then has none, or goes inactive. It waits three quarters before treating a closing as permanent, so a seasonal restaurant that closes for winter is not counted as dead, and it tracks a business through a change of name or ownership. The data show that a place stopped employing anyone. They do not show whether the owner lost money, retired or sold up.

Other studies count differently, which is why their numbers differ:

Read down the table and the lesson is the definition. A count of bankruptcies gives a tiny number. A count of ownership changes in one Ohio city over three years gives a large one. Luo and Stark, who used non-public BLS records for the West, treat a merger or spinoff as the end of what they can observe, not a death. Their paper is a draft on arXiv and does not show a journal publication, and the authors say their opinions are not the Bureau's.

The studies also disagree. Parsa's three-year figure of 59 percent for Columbus is well above what BLS shows for the nation: reading the same table, between 27 and 40 in every 100 new food and lodging establishments had closed by their third year. Different city, different decade and a different question could explain the gap, and no source here settles which one does.

How restaurants compare with other new businesses

Luo and Stark ranked more than 500 kinds of single-location service startups in the West. Full-service restaurants were the most common, with about 81,500 in the data, and they landed in the middle. Their selected results are below, with the share still open after one year and the median life, the age by which half had closed.

Limited-service restaurants did somewhat worse than full-service ones on both measures. Size at the start matters too. The paper finds survival rises with the number of employees at opening, and restaurants that began with five or fewer employees had a median life of 3.75 years, a little shorter than the 4.0 years for other service businesses that began that small.

Why restaurants close: what the research says

No single cause shows up. Four things do:

• The owner's life. Parsa interviewed Columbus owners. Many who closed named family issues such as divorce, poor health or wanting to retire, and all described how much time a restaurant takes. The Ohio State release says many restaurants close for personal reasons, not because they could not succeed financially.
• Starting small. In Luo and Stark's data, survival rises with startup size. Larger startups may need more capital, they write, and still tended to last longer.
• Low earnings and heavy debt. Luo and Stark's review of earlier research notes that restaurants with low earnings and high liabilities are more likely to go out of business. A 2024 study of publicly traded restaurant companies, from 2000 to 2019, found that the quick ratio (cash and near-cash against short-term bills), operating cash flow and working capital were the most crucial indicators of a bankruptcy filing. Those are large listed companies filing for bankruptcy, so treat it as a hint for independents, not a measure of them.
• Thin margins that costs can erase. The National Restaurant Association's July 2026 analysis describes a typical independent restaurant before the pandemic as spending about 33 cents of each sales dollar on food, 33 on labor and 29 on everything else, leaving roughly 5 cents before tax. It estimates total expenses for an average restaurant rose 36 percent from 2019 to 2026, and reports that 42 percent of operators said their restaurant was not profitable in 2025. It is a trade group, and these are its estimates and its operator survey.

Put that last bullet and the cash-flow finding together and you get the same message from two directions: when only a few cents of each dollar are left, a slow month or a jump in costs can use up the cushion before the bank balance does.

What a thin margin looks like: an example

This is an example, not a real restaurant. Rosa's Taqueria is a counter-service place with $80,000 in sales in an average month. Break-even is the sales a month needs to pay every bill; the busy-but-not-making-money guide walks through why it is the first number to know.

Now two ordinary shocks, one at a time:

• Sales dip 8 percent, to $73,600. Left over is $73,600 × 0.64 − $48,000, which is minus $896. The month loses money.
• Costs per sale rise 3 points, from 36 to 39 percent, with sales unchanged. Left over is $80,000 × 0.61 − $48,000, which is $800. Break-even climbs to about $78,700.

Neither shock needs a bad owner, a bad dish or a bad block. The cushion was only about 6 percent of sales. That is the pattern the research points to, and it is something you can see on your own numbers this week, before a bank balance shows it.

What it means for an owner

The average is not your restaurant

The federal counts say new food and lodging establishments last as long as new businesses in general, or a bit longer. They say nothing about whether yours is healthy. Half of a group lasting five years tells you nothing about which half you are in.

The pattern is money, not mystery

The research that looks at money points to earnings, debt and cash. Those you can measure. Prime cost is food and drink cost plus all labor cost, as a share of sales; it is the part of your costs that moves with every shift you run, and the prime cost guide shows how to work it out and which ceilings to compare it to. Break-even tells you how much room you have if sales dip.

This week

1. 01Enter last month's sales and costs in the break-even calculator. Write down your cushion (sales minus break-even), and compare it with your worst recent month-to-month dip.
2. 02Work out last week's prime cost: food and drink used plus all labor cost, divided by net sales. The prime cost guide has the formula and the ceilings by type of restaurant.
3. 03Add up what leaves your bank account in a normal week: payroll, supplier invoices, rent and loan payments. Divide your balance by that. The result is how many weeks you could pay your bills with no sales at all.
4. 04If the numbers say you are busy but not keeping money, follow the order in the busy-but-not-making-money guide, and fix the biggest gap first.