Ghost Kitchen Setup and Delivery Economics: Who Holds the Permit, and What It Costs to Make the Model Work

A ghost kitchen is not a restaurant with the dining room removed. Taking out the dining room changes who holds the permit you operate under, what the building department sees when it looks at your floor plan, and what "menu price" even means once one to three platforms sit between you and the person eating your food. Three different bodies of rules, none of them answered by a generic how-to-open-a-restaurant checklist.

Five things below, all specific to delivery-only: who holds the permit in a shared kitchen and what happens when the facility gets cited, the occupancy class a production-only kitchen lands in, the commissary-agreement terms that carry more exposure than the rent, what virtual-brand labeling requires, and the delivery arithmetic that decides whether the rest was worth doing. The layers underneath aren't repeated: the general permit sequence is in the permits checklist, plan review and the pre-opening inspection in working with your health department.

Who Holds the Permit — And What Happens When the Facility Gets Cited

In a shared or commissary kitchen, the facility holds a permit for the facility. Whether that permit covers your operation is a health-department-by-health-department call. Some jurisdictions let you work as a registered "user" under the host's permit with a separate registration layered on. Others require every operator producing food in the building to hold an independent food establishment permit. A few have a named permit type for this situation; many have no such category and simply issue you the standard food establishment permit at the host's address.

There is no national answer, and anyone who gives you one is guessing. Ask before you sign, not after — if you need your own permit at that address, the facility's permit status becomes a prerequisite to yours, and a facility not in good standing can stall your application.

Now the part that catches people:

When the facility gets cited or suspended, what happens to you is a contract question, not a permit question. A suspension of the host's permit can shut down every brand cooking inside that building, regardless of whose kitchen practices triggered it. Your food safety record does not protect you from someone else's walk-in.

That risk is manageable, but only in the document. Before signing, get in writing:

• The facility's permit status and inspection history — the actual reports, not a verbal assurance. Many health departments publish results online; look the facility up yourself.
• Whether one tenant's violation can suspend the whole facility, or only that tenant. Ask the facility, then ask the health department the same question, because the answers sometimes differ.
• What happens to your access and live orders if the facility closes mid-service. Rent abatement? A cure period? Any duty to notify you before an enforcement action becomes a locked door? Most agreements are silent, and silence favors whoever drafted it.

The entity layer doesn't disappear because you're a guest in someone else's kitchen. You still need your own entity, EIN, state tax registration, typically your own certified food protection manager, and often your own business license at that address. Work that sequence in the permits checklist; a host's paperwork substitutes for none of it.

One shared-facility surprise: inspectors check equipment listings on gear you bring in, same as a standalone build. If you're supplying your own prep table, undercounter unit, or hot-holding cabinet, the NSF mark matters — how NSF and UL marks work on used equipment covers what a valid listing looks like when the data plate is gone.

Occupancy Classification Without a Dining Room

This is where delivery-only genuinely diverges from a restaurant build. Occupancy classification drives fire separation, sprinkler requirements, egress, and occupant-load factors — design inputs with dollar figures attached, not paperwork you file afterward. A building department applies roughly a two-step test.

Step one: is there public dining or gathering at all? Under the International Building Code framework, Group A (assembly) covers spaces where people gather for civic, social, or food-and-drink purposes, generally at an occupant load of 50 or more; small assembly areas under roughly 750 square feet and accessory to another occupancy typically fold into that base occupancy. A pure production kitchen with no public area generally doesn't trigger Group A — which is why the model avoids the egress, restroom, and accessibility obligations a dine-in build carries.

Step two: for a commercial kitchen not associated with a dining facility, the code draws the line on floor area. ICC's significant-changes guidance for the 2012 IBC cycle puts such a kitchen in Group B (business) at or under roughly 2,500 square feet, and in Group F-1 — moderate-hazard factory/industrial — above it. Above the line you're classified like any other food-processing operation, a different requirement set than most first-time operators budget for. Source: ICC

Jurisdiction check: Every number here is an IBC concept from a 2012-cycle guidance document, and occupancy classification is one of the highest-variance items in this library. The adopted edition, local amendments, and the exact area threshold belong to your building department, and several states run their own code rather than the IBC. Treat 2,500 square feet and the 50-occupant threshold as the shape of the test, not your local rule.

Then the trap. Any public-facing element can pull the space back toward an assembly or mercantile classification and reopen every question the production-only design avoided — a pickup window, a counter selling bottled sauce, a vestibule with two chairs where drivers wait out of the rain. Each gets added late in a buildout as an obvious convenience. So take your floor plan, including anything that functions as a pickup point, to the building department before you finalize a lease: what classification applies, and what changes if you add the window. Free conversation before the lease; expensive one after the drywall.

What occupancy class doesn't change: producing grease-laden vapors means a Type I hood and fire marshal sign-off, Group B or F-1 alike. In most shared facilities the hood is already installed as house infrastructure — confirm it's rated for what you intend to cook under it, because a hood sized for a two-burner sauté station and one sized for a fryer battery are not the same appliance.

The Shared-Kitchen Agreement — What You're Actually Signing

A commissary agreement is not a lease, and reading it like one means skimming the clauses that matter. A lease is mostly about space and money. This document is mostly about access, liability, and who is responsible when something goes wrong in a building full of other people's food.

Secondary sources commonly cite access at roughly $15–$50 per hour or $500–$3,000 a month for a membership — a band wide enough to be nearly useless as a budget input, varying by city, access level, and dedicated storage. Order of magnitude only; the number that matters is on the term sheet in front of you. What to find in the actual document:

Two of those deserve more than a row. Insurance and indemnification is a recurring cost, not a formality: these agreements commonly require general and product liability coverage naming the facility as an additional insured, plus a clause holding the facility harmless for claims arising from your food, your staff, or your violations of law. No amount is standard — ask what limits the agreement requires and price those. Warewashing is the common surprise: shared three-compartment sinks and dish machines are typical but not universal, and some agreements put sanitation of shared equipment on tenants by rotation. If warewashing turns out to be yours to bring, sanitation and cleaning equipment is a lean, standard buy.

Have a lawyer read the document before you sign, the same way you would a commercial lease. The exposure in an indemnification clause or an exclusivity restriction can exceed the rent by an order of magnitude.

Virtual Brands From One Kitchen — What the Labeling Actually Requires

Two questions get conflated here constantly, and they have different answers from different authorities.

The permit and traceability question. The health permit attaches to the physical kitchen and the operator who holds it, not to the consumer-facing brand name. A kitchen running three virtual brands is, to the health department, one licensed operation — or, depending on the licensing structure above, several registered users of one facility. Whatever name appears in the app, the permit holder of record is what an inspector sees and what the public record shows. Ask the facility how the permit is posted and how a member of the public would identify the holder for a given brand; that sounds like trivia until a bad inspection turns "which entity holds this permit" into the only question. Each brand name may also require its own DBA or fictitious name registration — requirement, process, and fee are set at the state or county level.

The platform-level disclosure question. This is platform policy, not food code, and it's inconsistent. Trade press reports DoorDash has labeled virtual-brand listings as such in its own app since March 2021; other platforms have been less consistent about surfacing that a listing is a virtual concept rather than an independent restaurant. Source: Food On Demand We could not corroborate that against a platform-published policy, and app UI changes without notice — check current in-app behavior yourself before you launch, on each platform you plan to list on.

How virtual brands appear in Google Business Profile — hidden addresses, service-area registration, what a shared kitchen can claim — belongs to the website and online ordering guide; read it before you choose a space.

One edge case worth bounding so you can stop worrying about it: FDA's menu-labeling rule, the calorie-count one, applies to chains of 20 or more locations doing business under the same name with substantially the same menu. Source: FDA That threshold was built for physical multi-unit chains, and one kitchen running two or three virtual brands isn't close. Revisit it only if you license a brand out to enough kitchens to approach it.

One packaging habit worth adopting even where nothing requires it: put the permit holder's business name on your receipts or packaging alongside the marketing brand. It costs nothing at print time and keeps a customer complaint from becoming "who even made this?" Budget takeout packaging with your smallwares rather than discovering it in week one.

The Delivery Economics — Whether the Model Actually Works

A delivery-only kitchen has no dine-in channel to fall back on, so the take rate isn't a cost center — it's the margin structure. The question isn't "marketplace or direct." It's: what does a delivery-only kitchen have to charge to make the same margin a dine-in restaurant makes?

The baseline is thin. In the National Restaurant Association's operations data, the median full-service operator reported income before taxes of 2.8% of sales and limited-service 4.0%. That's what the take rate comes out of, which is why a few points of unbudgeted platform cost isn't a bad month — it's the business. That data and the food-cost bands below are in food cost and prime cost.

The commission number, briefly

Major platforms run tiered commission from roughly 15% to 30% of subtotal by platform, plan, and delivery-versus-pickup. The tier buys in-app visibility as well as a rate, which is why "DoorDash takes 30%" is wrong often enough to unlearn. The platform-by-platform breakdown lives in the website and online ordering guide, along with two jurisdiction items that apply to you — city-level commission caps, less protective than they sound, and marketplace facilitator sales tax rules. Take those numbers as given here.

The marketing fee stack on top of commission

Commission is the headline. It is not the ceiling.

Sponsored listings are a separate, bid-based cost. A higher commission tier does not buy placement — a restaurant on the top tier can still be outranked by a competitor who also pays for promoted placement. DoorDash's pricing page lists Sponsored Listings and Promotions as opt-in extras alongside the plan tiers, plus a $6/week tablet fee. Source: DoorDash It doesn't publish Sponsored Listings pricing, because it's an auction: what you pay is what your competitors bid. Platform promotions and loyalty-tier subsidies are a second layer — discounts that keep your listing eligible are typically funded in part by you. Refunded and adjusted orders are a third, landing entirely on your side.

Stack all of that and the effective take rate lands meaningfully above the tier rate. Industry and consultant commentary — a secondary-source estimate, not a platform-published figure — puts effective take rates at roughly 28–35% on DoorDash and 29–36% on Uber Eats once marketing spend, subsidies, and adjustments are counted. Your own number depends entirely on how much promotion you buy. The point of citing a range at all: the honest planning number is not the tier rate on your contract, and modeling as if it were is the most common way a delivery-only budget breaks. Model your own the boring way — a full month of platform payouts, net deposits divided by gross order subtotal. Do it monthly. It moves.

What it compounds against

A take rate doesn't subtract from your margin — it subtracts from your revenue, while food cost stays anchored to the original menu price. The two don't offset; the second grows as a share of what's left. At a fixed $12 food cost on a $36 delivery ticket:

Read the last two columns together. The take rate moved 20 points; the money left to run the business fell by 39%. The same unchanged $12 of food went from 39% of net revenue to 51% — which reads like a catastrophic food-cost problem on a P&L, except nothing about the food changed. That gap is why a delivery-only concept can hit textbook food cost against menu price and still not clear.

The honest markup arithmetic

So you raise delivery prices, and almost everyone gets the formula wrong. To net the same dollars from a delivery order as from a walk-in order priced at P, after a total take rate of r, the delivery price has to be P ÷ (1 − r) — not P × (1 + r).

That distinction is the whole section. At a 25% take, the intuitive move is to add 25%: a $15 item becomes $18.75. But 25% of $18.75 is $4.69, so you net $14.06 — ninety-four cents short on every item, forever. The correct price is $15 ÷ 0.75 = $20.00.

The markup grows faster than the take rate itself, because you're solving for the price that survives the cut — not the price that offsets it. From 15% to 35% the take rate a bit more than doubles; the required markup roughly triples.

That's why the common practice of listing delivery prices 10–20% above in-store is frequently short of break-even here. At a true 15% effective take, a 17.6% markup holds. At a stacked 35%, a 20% markup nets you less per order than a walk-in — every order, with no dine-in channel making up the difference. There's a ceiling on the fix, though, and you'll find it: past roughly a third above in-store pricing, conversion suffers and you're trading volume for margin.

What actually closes the gap

Lower labor — honestly stated. This is the model's real advantage and it's smaller than the pitch. No servers, host, bussers, or dining-room cleaning, and a much smaller footprint to heat, light, and insure. The NRA medians give the shape: payroll ran 36.5% of sales for full-service and 31.7% for limited-service. A delivery-only kitchen still carries back-of-house payroll — it sheds the front-of-house share and the dining-room occupancy cost, not the labor line. A few points of headroom plus rent you don't pay, not a free 36%.

Fixed-cost spreading across brands. One rent line, one hood, one dish pit, one utility bill carrying two or three concepts — what that now requires is the next section.

Volume against fixed costs. Operators commonly discuss needing roughly 25–50 orders a day to cover fixed costs — industry commentary, heavily dependent on rent, ticket average, and concept, not a benchmark to plan against blindly. Run your own: monthly fixed costs divided by contribution per order (the "left for labor, rent, equipment, profit" column above, minus variable labor and packaging). If the answer is past what your market can deliver by month three, the problem is the plan, not the execution.

The Multi-Concept Strategy, Realistically

Running two or three virtual brands from one kitchen is still a legitimate way to spread fixed costs. It is no longer the free lunch it was in 2021–2022, because platforms now police duplicate and low-quality listings — and publish the standards. DoorDash's virtual brand quality requirements, in brief:

Miss the performance minimums and the brand can be removed, not merely deprioritized. Source: DoorDash These are the platform's own rules and they change without notice — check the current version before building a menu around them. Uber Eats enforces comparable standards: trade press reported roughly 8,000 virtual listings removed in a crackdown beginning March 2023, targeting duplicate and low-performing brands and requiring menus to differ from a parent restaurant's at the same address by more than half the items. Source: Nation's Restaurant News

So: a virtual brand is a real launch, not a free menu tab. Distinct branding, its own photography, enough menu difference to survive an audit — and if it underperforms it can be delisted within months, wasting the setup cost. Don't launch a third brand until the first two clear the published minimums; a limping brand isn't spreading fixed costs, it's consuming the attention the working brand needed. More brands also means more concurrent orders on the same cold storage, so revisit refrigeration capacity before adding a third concept. The build list and budget belong to the ghost kitchen equipment guide; where to buy used commercial kitchen equipment covers sourcing once licensing is settled.

What to Ask Before You Sign

The guide compressed into a script, grouped by who has the answer.

The facility or commissary operator

• Whose permit covers my operation here, and do I need my own?
• What's your permit status, and can I see your inspection history in writing?
• If your permit is suspended, what happens to my access and to orders already in?
• What insurance do you require, at what limits, and are you named as additional insured?
• Is there an exclusivity or no-outside-kitchen clause, and what does it restrict?
• Which equipment is dedicated, which is shared, and who handles grease trap service?
• How are production blocks assigned at peak?

The building department

• Given this floor plan, what occupancy classification applies, and which code edition and local amendments are you enforcing?
• Does that change if I add a pickup window, a driver waiting area, or a retail counter?

The health department

• Can I operate under this facility's permit, or do I need my own at this address?
• Is there a shared-kitchen-user permit or equivalent registration here?
• Does each virtual brand need separate menu approval, and whose name goes on the inspection report?

Your own math

• What total take rate — commission plus marketing — can I absorb at my real food cost and still hit my margin?
• What delivery price does that require, using P ÷ (1 − r), and will my market pay it?
• How many orders a day covers fixed costs, and is that realistic in month three?

Answer the last group honestly and you'll know whether to sign before you've spent a dollar on equipment.

Frequently Asked Questions

Do I need my own health permit if I'm renting space in a shared commercial kitchen?

It's a jurisdiction-by-jurisdiction answer with no national rule. The facility's permit covers the facility; whether it extends to operators inside varies. Some health departments let you operate as a registered user under the host's permit, some require every operator to hold an independent food establishment permit, and some have a distinct shared-kitchen-user category. Call yours before signing, because the answer changes what you're buying. Either way you'll typically still need your own entity, EIN, tax registration, and manager certification.

What happens to my ghost kitchen if the shared facility's permit gets suspended?

That's a contract question, not a permit question. A suspension of the host's permit can shut down every brand cooking in that building regardless of whose practices caused it, so a clean record of your own doesn't protect you. Before signing, get the facility's permit status and inspection history in writing, ask both the facility and the health department whether one tenant's violation can suspend the whole building, and find out what the agreement says about your access and live orders if the doors close mid-service. Most say nothing.

What building occupancy classification applies to a delivery-only kitchen?

Under the IBC framework, a commercial kitchen not associated with a dining facility is generally Group B (business) at or under roughly 2,500 square feet and Group F-1 (moderate-hazard factory/industrial) above it, and with no public dining area you generally avoid Group A (assembly) entirely. Those figures come from ICC guidance for the 2012 code cycle — the adopted edition, local amendments, and the exact threshold are your building department's call, and several states use their own code. Adding a pickup window or driver waiting area can change the answer, so take your floor plan in before signing a lease.

How many virtual brands can I run from one kitchen?

Platform rules bind before code does. DoorDash's published virtual brand quality requirements cap brands at 10 per address absent an exception and require at least 8 menu items, 50%+ of main items genuinely differentiated from other brands at the same address, half the menu hot or prepared, and real non-AI photography. Performance minimums — roughly 3 orders a week on a 90-day average, a 4.0+ rating, cancellations and missing-order reports under 5% — can get a brand removed outright, and Uber Eats enforces similar standards. Policies change without notice, so check the current version before building a menu around them.

How much does it really cost to sell through DoorDash or Uber Eats once marketing fees are included?

More than the commission tier. Headline commission runs roughly 15% to 30% of subtotal by platform and plan, but sponsored listings are a separate bid-based cost, platform promotions and loyalty-tier discounts are partly funded by you, and refunded orders land on your side. Industry and consultant commentary estimates effective take rates of roughly 28–35% on DoorDash and 29–36% on Uber Eats once that's counted — a secondary-source estimate, not a platform-published figure, and dependent on how much promotion you buy. The reliable version is yours: divide a month of net platform deposits by gross order subtotal.

What should I look for in a shared kitchen or commissary rental agreement?

Read it as a liability and access document, not a lease. Find the access structure and how blocks are assigned at peak; the insurance limits required and whether the facility is named as additional insured; the indemnification clause, which typically holds the facility harmless for anything arising from your food or staff; any exclusivity clause that could cap you at one location; what equipment is dedicated versus shared and who handles grease trap service; and what a 30-day termination notice does to a permit tied to that address. Have a lawyer read it — the exposure in an indemnification clause can exceed the rent.