Part 3 of "Who Decides Where You Eat"
Before arguing about this one, it is worth looking at the only time anybody actually tested it.
Between 2020 and 2022, a great deal of money was spent trying to remove location from the restaurant business altogether. That experiment has now finished. The results are worth going through carefully, because they ended up explaining something nobody set out to measure.
What happened to ghost kitchens?
Delivery-only operations in cheap industrial space — no dining room, no sign, no walk-ins, a dozen brands cooking out of identical stainless steel bays — attracted more than three billion dollars in venture capital. The logic was clean and hard to argue with. Prime retail rent is the single largest fixed cost a restaurant carries. Foot traffic is expensive to buy. If customers now find you through an app, why pay for a corner?
The scale of the bet was serious. Travis Kalanick's CloudKitchens raised around $850 million in late 2021 at a $15 billion valuation. Kitchen United raised $100 million in July 2022 with grocery giant Kroger participating. The commercial real estate firm CBRE projected that ghost kitchens would account for more than a fifth of the U.S. restaurant industry by 2025.
Then the results came in.
Kitchen United closed all eight of its in-store Kroger locations in November 2023, about sixteen months after that funding round, and by February 2024 had shut its remaining physical sites entirely and pivoted to licensing software. Its assets were absorbed into another company that spring. CloudKitchens was reported to be running at roughly 50 percent occupancy by 2023, closing warehouses and cutting staff; at five of its facilities, 41 of 71 restaurant tenants closed within a single year. Wendy's abandoned a partnership that had once promised hundreds of delivery-only locations. Uber Eats removed thousands of virtual brands from its platform.
Fairness requires noting that location was not the only thing that killed this. Delivery platform commissions of 15 to 30 percent against restaurant margins in the single digits would strain any model. Many operators launched five virtual brands before proving one. The capital arrived with expectations that forced expansion ahead of evidence.
But something more specific happened, and it is the useful part.
What did the experiment actually prove?
Not that location is magic. It proved that "location" was never one thing.
It was a bundle — several distinct functions that had always arrived together, so nobody had ever been forced to itemize them. Remove the street and they all vanish at once, and only then does it become clear how many there were.
Walk-ins are free customers. A restaurant on a real street receives a steady trickle of people who did not choose it in advance. In a warehouse, every single customer must be acquired, and acquired through a platform that charges for the privilege. The cheap rent was real. It was also, in a fairly direct sense, being paid to the delivery app instead of the landlord.
Reviews have to start somewhere. This connects back to the first part of this series. Ranking well requires review volume, and early review volume comes overwhelmingly from people who showed up without having read anything. A place with no walk-ins has no bootstrap. It is asking to be discovered by a system that rewards having already been discovered.
Occasions are geographic. Pre-theater dinner exists because theaters exist in a particular set of blocks. The lunch rush exists because offices do. These demand patterns are produced by the ground, and no amount of marketing manufactures one where the ground does not support it.
And a place with no address cannot be recommended by a person. This is the one nobody priced in. You cannot tell a friend to go to the third bay of an unmarked building off the highway. There is nothing to tell. The single most durable form of recommendation in this entire history — the one the first part of this series ended on, the one no institution ever managed to replace — requires a place that a human being can name and another human being can go to.
A virtual brand also has no reputation to protect. If the reviews turn bad, it can be deleted and relaunched under a new name by Friday. That sounds like an advantage. It is the opposite: the accountability that makes a recommendation worth anything comes from having something to lose, and an address is the most basic thing a restaurant can have to lose.
But hasn't search made location less important?
Yes. Genuinely, and this deserves to be said without hedging, because the ghost kitchen wreckage tempts people into overcorrecting.
Search dismantled one specific function of location, and it was for a century the most important one: being found. A restaurant on a bad corner used to be invisible, and invisibility was fatal regardless of how good the cooking was. That is simply no longer true. A place on a second floor, down a side street, in a neighborhood nobody walks through, can now be a destination. People will find it, and they will travel.
The practical consequence is real. The premium a restaurant should rationally pay for a prime address has fallen, because part of what that premium used to buy — discovery — is now available by other means and considerably cheaper. Money that once had to go into rent can go into photographs, into a menu that machines can read, into the slow work of becoming the place people mention.
But "less important for one job" is not "unimportant," and the distinction is where operators lose money in both directions.
So what is location still doing?
Four things, none of which search touched.
It sets the ratio of first visits to habits. Search produces first visits, and first visits are wonderful. They are also expensive, one-off, and easily won by whoever ranks above you next quarter. A habit is different. Habits form along the paths people already walk — the commute, the route home, the block between the office and the train. A restaurant that sits on somebody's line gets a frequency that no ranking confers.
It determines what occasion you can own. Every restaurant is competing for a small number of situations: the quick lunch, the drink after work, the dinner before the show, the Sunday with family. Which of these are available to you is decided largely by what surrounds you. This is not a marketing variable. It is a geographic fact you inherit with the lease.
It sets the cost structure. Rent as a share of revenue is the number that quietly decides whether a good restaurant survives a bad quarter — the common benchmark for full-service restaurants runs somewhere in the range of six to ten percent. A cheaper address is not a compromise. It can be a strategic advantage, provided the money saved goes somewhere useful.
And delivery never escaped geography either. A delivery radius is a circle drawn on a map. Drive time is drive time. The most digital revenue stream a restaurant has is still bounded by where the building physically sits.
The current data points the same direction, quietly. Through 2026, sit-down dining traffic has held up better than short off-premise visits, which have been falling. In Manhattan, retail demand has been expanding beyond the traditional prime corridors while the supply of quality storefronts keeps tightening — which is to say that after every prediction of its death, physical restaurant space in a dense city is something people are still competing to rent.
Why does this sound familiar?
Because it is the same shape as the conclusion of the first part of this series, arrived at from the opposite direction.
That part ended on a puzzle. A hundred and twenty-five years of institutions tried to replace word of mouth. Each one succeeded at something, and each one was eventually bought — by tire interests, by tourism boards, by a search company, by a bank. The only thing never bought or sold was the original: a person you know telling you to go somewhere.
Location is the other one.
They share a property, and it is the property that explains everything in all three of these articles. Neither of them scales.
A recommendation from someone who knows you cannot be mass-produced; the moment you mass-produce it, it becomes something else, and that something else is advertising. A particular corner, in a particular neighborhood, with a particular set of people walking past it at six o'clock, cannot be duplicated either. You can open a second location. You cannot open the same location twice.
Everything in this history that could be scaled was scaled, and everything that was scaled was eventually sold to somebody. The guide, the survey, the review platform, the feed — all of them ended up owned, and their judgments ended up serving whoever was paying. The two things that resisted are the two things that could not be manufactured: knowing someone, and being somewhere.
That is not sentimentality. It is the reason both of them are still standing after every technology that was supposed to make them obsolete.
So does it still matter where you are?
It matters differently, and arguably more than before — just not for the reason it used to.
For a hundred years, the constraint was being findable. A great restaurant in the wrong place died quietly, and everybody understood that this was unfair and unavoidable. Enormous effort went into solving it, and the effort worked. That problem is now largely solved.
But solving it did not remove the constraint. It replaced it with a harder one.
When nobody could find you, being hard to find was an excuse. Now that everybody can find you, there isn't one. The question is no longer whether people can locate the door. It is whether, having located it, they want to walk through it — whether the room is worth sitting in, whether the second visit happens, whether anybody tells anyone else.
Every layer this series has described — the inspector, the survey, the star rating, the map, the video, the summarized answer — ends at exactly the same place. A street. A door. A room with a specific number of tables and a specific set of people in it on a Tuesday night.
That has not been digitized. Nobody has found a way to sell it, aggregate it, rank it out of thirty, or compress it into a sentence. A hundred and twenty-five years of trying to answer the question of where you should eat, and the answer still has an address.
This concludes "Who Decides Where You Eat."
Notes on sources
Ghost kitchen funding, occupancy, and closure figures: CNN Business and Modern Retail coverage from late 2023; Financial Times reporting on CloudKitchens occupancy, September 2023; Kitchen United's Kroger closures, November 2023, and its exit from physical operations in February 2024; CBRE's 2021 projection for the sector. Delivery commission ranges and tenant churn: industry analyses, 2025–2026. Dining traffic patterns: Placer.ai Dining Index and Dor foot-traffic analyses, 2026. Manhattan retail conditions: REBNY H1 2026 Manhattan Retail Report. Rent-to-revenue benchmarks: standard full-service industry guidance.
Comments
Post a Comment