Why Cafes Fail: The Numbers Behind the Closures

The 90 percent number is folklore

You have heard it from a landlord, a banker, or the guy at the next table: nine out of ten places like yours close in the first year. It gets repeated because it sounds like hard-won wisdom, and because nobody ever asks where it came from.

It came from a credit card company. The hospitality researcher H.G. Parsa and colleagues, in Why Do Restaurants Fail? Part III (University of Denver, Daniels College of Business, citing 2010 industry data), name the source and correct it in the same paragraph: "Erroneously, American Express has estimated that 90% of restaurants fail in the first year." Their own earlier study, published in the Cornell Hotel and Restaurant Administration Quarterly in 2005, put first-year failures "in fact under 30%," and a separate 2004 analysis of California data landed under 30 percent as well.

That gap matters more than it looks. If you believe nine in ten close, then a shop that closes is just playing the odds and there is nothing to learn from it. If closures run closer to three in ten, then most shops make it, the ones that do not had reasons, and the reasons are things you can go look at in your own numbers this week.

What the closure data actually says

Three findings, from sources you can open yourself.

First year: under 30 percent, not 90. Parsa and colleagues, above. They also note the National Restaurant Association treats roughly 30 percent as the industry norm for the first year. Note the honest caveat inside that number: their data counts closures and ownership changes together, so some of what gets recorded as a failure was a sale.

Five years: roughly half, and that is not a food business number. The US Small Business Administration's Office of Advocacy, working from Bureau of Labor Statistics Business Employment Dynamics data, reports in its Frequently Asked Questions About Small Business, 2018 that "about half of all establishments survive five years or longer," ranging from 45.4 percent for businesses started in 2006 to 51.0 percent for those started in 2011, with about one third surviving ten years or more. Roughly four in five survive their first year across all industries.

Where the failed ones were bleeding. This is the finding worth taping to a wall. In the same Parsa paper, the section on cost controls reports that restaurants that failed commonly ran food and labor costs together above 60 percent of revenue, and gives a worked case of a restaurant with food cost above 56 percent that closed within two years. For scale on the food half: the National Restaurant Association's 2025 Restaurant Operations Data Abstract puts the limited-service median food and non-alcohol beverage cost at 32.4 percent of sales in 2024.

Cafes are not restaurants, and nobody counts us separately

Everything above is restaurant and all-industry data. I am not going to pretend it is a cafe number, because there is no cafe number. Nobody publishes measured survival or cost medians for single-location coffee shops, and the closest figures come from datasets where a 200 seat dining room and a 12 seat espresso bar sit in the same bucket.

That cuts both ways for you. A cafe has structural advantages the averages hide: food cost on espresso and drip is low, your inventory is 60 items instead of 600, and you sell a $6 product with no line cooks behind it. It also has a disadvantage the averages hide: your ticket is small, so nothing gets fixed by one good night, and a few cents of drift per drink compounds across hundreds of transactions a day rather than dozens.

So read the research for its shape, not its decimals. The shape is that failure clusters where costs run away from revenue quietly, over months, in a business with thin margins and high fixed rent. That describes a cafe exactly.

The failure is arithmetic, not drama

Nobody closes because of one bad decision. The pattern I see from inside my own shop is three ordinary leaks that never announce themselves, which is precisely why they are dangerous.

Food cost drift. You priced the menu once against real ingredient costs. Since then your milk went up twice, your beans went up once, and a pour that was supposed to be 10 ounces became 12 on the busy shifts. Nothing in your register objects. Your prices are still what they were in January while your costs are what they are today, and the gap between those two facts is not visible on any report you currently read. The fix is knowing which drinks moved, which is an exposure calculation rather than an instinct.

Over-ordering. The most common cash problem in a small cafe is not that money was spent badly, it is that money is sitting on a shelf as three extra cases of oat milk and a case of matcha that will last five weeks. It looks like being prepared. It reads on your bank balance as being broke, and some of it will expire before you get through it. Ordering from a count rather than a feeling is a cash-flow fix disguised as an inventory chore.

Unpriced modifiers. This one is specific to us and it is the biggest. A cafe's costs live in swaps: oat instead of whole, an extra shot, a large instead of a small. Your POS records the drink and the money. It has no idea that "make it oat" moved 10 ounces onto your most expensive milk. I have watched exactly this go wrong at scale in my own data: when the modifier mapping broke, my computed oat usage ran roughly three times what the shop was actually pouring, and every cost number downstream of it was wrong until the mapping was fixed. Numbers you trust and cannot verify are worse than numbers you know you do not have.

The four numbers that catch it early

None of this requires an accountant. Four numbers, checked on a schedule you actually keep.

1. Honest food cost, monthly. Not invoices divided by sales. Starting count plus purchases minus ending count, over net sales for the same dates. The difference between those two methods is routinely three or four points, and points are the unit failure is measured in.
2. Cost per drink on your top five sellers, quarterly. Recomputed against current supplier prices. If the drinks carrying your volume have drifted, everything else is noise.
3. What is sitting on the shelf, weekly. Counted, not remembered. My own full count runs 58 items in about 8.5 minutes on a phone, and my manager has run the last 26 of them. If a count takes an hour, it will not happen weekly, and the number that never gets taken is the number that hides everything else.
4. The gap between what sales say you used and what the shelf says you used. That difference is waste plus untracked usage, and tracking it is the whole method.

The research says most shops make it, and the ones that do not were usually running costs well above where the operator believed they were. Both halves of that sentence should be encouraging, because the second half is measurable months before it is fatal.

Pick the first one and do it this month. Two counts, the invoices in between, net sales over the same dates. If that number comes back four points off where you assumed, you have just found the thing the closure statistics are actually describing, while there is a full year left to do something about it.