Case Study 2 — The Ghost-Kitchen Buildout and What Came After: A Sector Arc and One Operator's Unwind
A note on sourcing. The sector arc described in Part One is public record in outline: the COVID-19 dining-room closures beginning in March 2020, the resulting shift of restaurant sales to off-premise channels, the delivery-commission-cap ordinances several cities adopted, the rapid expansion of delivery-only kitchen capacity and virtual brands, and the subsequent contraction — facility closures, consolidation among operators, and retrenchment by chains that had launched delivery-only concepts — as dine-in traffic returned. No company is named and no company's financials, unit counts, or fee schedules appear here, because those are not reliably public and this book does not invent them.
Part Two is a clearly labeled composite — a single operator built from patterns common to many. Every figure in it is constructed and illustrative. It is not any real business.
Part One: the sector arc
The setup
In March 2020, American dining rooms closed. Chapter 28 already gave you the structural consequence: off-premise stopped being a supplementary channel and became, for a period, the entire business. The industry's cost structure was rebuilt around it in a matter of weeks — packaging, curbside logistics, menu abbreviation, first-party ordering, and above all a dependence on third-party marketplaces whose commission rates, in normal times, would have been considered unaffordable.
Several cities responded to that dependence with commission-cap ordinances limiting what marketplaces could charge restaurants during the emergency. That those ordinances existed is the clearest available evidence of the underlying problem: the take rate was high enough that municipal governments treated it as a threat to the survival of local restaurants.
Into that environment came a proposition that looked, on the arithmetic, extremely attractive.
The pitch
The logic is worth restating because it was genuinely good logic, and because you will hear a version of it again in some other format within five years.
A restaurant's most expensive irreversible commitments are the ones the guest sees. Bellwether's \$620,000 is construction \$310,000 · equipment \$185,000 · FF&E \$45,000 · pre-opening \$35,000 · working capital \$45,000 — and a large share of that construction figure exists to make a room somebody wants to sit in: the dining room, the bar, the millwork, the lighting, the restrooms, the finishes, the visible frontage that justifies a retail rent. Bellwether then pays \$95,200 a year for ten years to keep that room, guaranteed personally, with total exposure of \$1,367,600.
Delete the room. Keep the kitchen. Sell into a delivery channel that is, at that moment, the only channel there is.
The capital falls by most of an order of magnitude. The lease becomes a license agreement measured in months. The front-of-house payroll disappears. The retail rent premium disappears. You can be producing in weeks rather than in the eight-to-twelve months a build-out takes.
Facility operators built capacity accordingly. Restaurant companies launched delivery-only brands out of existing kitchens. A great deal of money moved into the format very quickly.
What broke
Dine-in came back.
And when it did, the format's single structural dependency became visible, and it is exactly the one this chapter's Figure 30.5 isolates: a delivery-only business has one sales channel, and it is the channel with the worst take rate in the industry.
Run the logic. A conventional restaurant sells through multiple channels with different economics — a walk-in guest at roughly zero acquisition cost, a reservation, a bar seat, a private event at Chapter 29's margins, a takeout order at a card fee, and a marketplace delivery order at 25% to 30%. If the marketplace channel gets worse, the restaurant shifts weight. It has somewhere to shift to, because it owns a room.
A delivery-only operation has nowhere. It cannot open a counter. It cannot host an event. It cannot win a second visit through hospitality, because the guest has never met anyone who works there. Its growth comes from platform placement and advertising, which means the platform controls its volume. Its reputation lives in a ratings system it does not administer. And its facility fee is a second landlord, layered on top of the first.
While demand was extraordinary, all of that was survivable. When demand normalized, the format's margin — never large — went negative for a great many operations, and the sector contracted: facilities closed, operators consolidated, and chains quietly retired delivery-only brands.
The lesson is not "ghost kitchens don't work." Some do, and §30.5 shows the three levers that make them work: average order value, channel mix, and volume. The lesson is narrower and more useful: a business whose entire revenue depends on a single channel it does not control has no margin of safety, regardless of how good its prime cost is. Figure 30.5's operation held a 57.0% prime cost — better than the full-service benchmark this book has spent thirty chapters teaching — and lost money anyway.
Part Two: one operator's unwind
\[Labeled composite — constructed teaching example. All figures illustrative. Not a real > business.\]
The operator
A 90-seat casual American restaurant in a mid-size market. Pre-2020: roughly \$1.9 million a year, about 88% dine-in and 12% takeout, run by an owner who had held the business for nine years and knew her numbers.
March 2020 took the dining room. Revenue fell to a fraction of plan. Like a great many operators, she did two things to survive, and it is important to see that they were different decisions with different economics, even though at the time they felt like one decision.
Decision one: three virtual brands out of her own kitchen. A wings brand off the existing fryer station, a family-meal pasta brand, and a breakfast-sandwich brand for the morning hours the kitchen was otherwise dark. Cross-utilized ingredients, existing equipment, existing hood, existing walk-in, existing rent.
Decision two: a licensed suite in a purpose-built delivery-only facility across town, to serve a second delivery zone her own location could not reach inside the platforms' delivery radius. This required equipment, a license agreement, and dedicated staff.
Both decisions produced revenue. Only one of them produced money, and it took her two years to see it clearly because she was measuring total off-premise revenue rather than measuring each channel as its own business line — which is precisely the error Chapter 28 warned about.
The two extensions, separated
Here is the year she finally separated them.
The three virtual brands, out of her own kitchen:
Orders 6,200
Average order subtotal $28.00
Revenue $173,600
Commission, blended 22.0% ($38,192)
Food cost at 28.0% ($48,608)
Packaging at 5.2% ($9,027)
Incremental labor ($34,000)
Tech, photography, supplies ($9,000)
────────────────────────────────────────────────
Facility cost $0 ← the whole point
────────────────────────────────────────────────
CONTRIBUTION +$34,773 (20.0%)
The licensed suite, across town:
Orders 9,600
Average order subtotal $31.00
Revenue $297,600
Commission, blended 25.0% ($74,436)
marketplace 8,448 × $31 × 28.0% = $73,329
first-party 1,152 × $31 × 3.1% = $1,107
Food cost at 29.0% ($86,304)
Packaging at 5.6% ($16,666)
Labor, all-in ($96,500)
Facility license $3,100 × 12 ($37,200)
Utilities and pass-throughs $700 × 12 ($8,400)
Platform advertising ($18,300)
Tech, insurance, supplies, permits,
admin ($21,400)
────────────────────────────────────────────────
OPERATING RESULT ($61,606) (20.7%)
Equipment note, $62,000 over 4 yrs
at ~10% — $1,572/month ($18,864)
────────────────────────────────────────────────
ANNUAL CASH BLEED ($80,470)
Same food. Same brands. Same platforms. Same commission structure, roughly. One made \$34,773 and one lost \$80,470, and the difference was who owned the kitchen.
That sentence is the whole case, and it is the chapter's conclusion arriving as an operator's experience rather than as a principle: small-format profit comes from selling into capacity somebody has already paid for. Her own kitchen's rent, hood, walk-in, and insurance were sunk. The suite's were not — and \$45,600 a year of license and utilities, layered under a 25% commission, was more than the channel could carry at the volume it actually produced.
The contested decision
Fourteen months remained on the license agreement. She had three options, and the argument inside her own head is the one Chapter 39 describes: this is either a problem I can fix or a business that is over, and I have to decide which.
Option A — run it to term. Fourteen months at the current bleed rate: $\$80{,}470 \times 14 \div 12 = \$93{,}882$.
Option B — fix it. Attack the three levers from §30.5: raise average order value from \$31 toward \$38 with bundles; drive first-party share from 12% toward 35%; grow volume to spread the \$67,000 facility-and-overhead block. All three are real levers and all three were available. The problem was the fourth variable she could not change: the suite existed to reach a second delivery zone, and that zone's demand had normalized downward while her own kitchen had recaptured the nearer half of it. She was, in effect, competing with herself in the geography where she was already profitable. Chapter 35 has a word for this: cannibalization.
Option C — close it now. Worst case, pay out the remaining license in full: $14 \times \$3{,}100 = \$43{,}400$. Sell the equipment: an estimated \$19,000 against \$26,000 of remaining note principal, a \$7,000 shortfall. Redeploy the staff into her own kitchen, which was busy again. **Total: \$50,400.**
Run to term (14 months) $93,882
Close now, license paid in full $50,400
─────────────────────────────────────────────────
SAVED BY CLOSING $43,482
She closed it. And she discovered, in the process, two things worth carrying forward.
First, the license agreement had no assignment right. She had assumed she could hand the suite to another operator — there were operators who wanted it — and the agreement did not permit it without consent she could not obtain quickly. This is Chapter 6's lesson arriving in a document nobody thought of as a lease: a shorter term is not the same as an easier exit, and the clauses that matter later are the ones you read last. Read the assignment and early-termination provisions of a ghost-kitchen license before you sign it, with the same attention Chapter 6 tells you to give a percentage-rent clause.
Second, the dine-in guest had been paying for the brands all along. When she pulled her weekly numbers back three years, peak-hour ticket times had risen from about 15 minutes to about 23 during the period the brands ran unrestricted — and her review scores under the restaurant's own name had drifted down in the same window. That cost never appeared on any channel P&L. She capped the virtual brands by daypart — off during peak dinner service, on during the closed morning and the mid-afternoon — and ticket times settled at about 17.
The three brands out of her own kitchen still run. They contributed \$34,773 in the year above and they cost her nothing in rent.
What it shows
1. "Off-premise revenue" is not a line. It is several businesses. Measuring them together is what hid a \$80,470 bleed inside a growing top line for two years. Chapter 28 said to model each channel as its own business, and this is what happens when you don't.
2. A short term is not a cheap exit. The format's headline advantage over a ten-year lease is real, but a license agreement is still a contract with a term, and if it has no assignment right your only exits are running it out or buying it out. Price both before you sign.
3. Capacity you already own is the best capacity there is. Nothing else in this case comes close in explanatory power. The extension into her own kitchen worked. The extension into somebody else's did not. Same food, same commissions.
4. The sunk-cost trap is arithmetic, and the arithmetic usually says stop sooner than feels tolerable. Closing fourteen months early and paying the license in full still saved \$43,482. She came within a board meeting of running it to term "because we've already put so much into it," and Chapter 39 exists because that sentence has closed more restaurants than any recipe.
5. Prime cost is necessary and insufficient. Figure 30.5's operation was better than benchmark on both halves of prime cost and lost money. In channel-heavy formats, channel cost belongs on the weekly flash report next to food and labor. If your flash report has no line for it, add one this week.
Discussion questions
-
The composite operator's two extensions had nearly identical prime-cost structures and opposite outcomes. Write the one-sentence diagnostic she should have applied at the moment she was considering the suite, and explain why "we're already good at delivery" was the wrong reason to sign.
-
Option B — fix it — was genuinely available, and the chapter's §30.5 levers were all real. Make the strongest case for Option B, including the specific dollar improvements the three levers would have needed to produce to break even. Then say why cannibalization defeats the case.
-
She discovered the assignment problem after signing. Draft the five questions you would ask about a ghost-kitchen license agreement before signing, in priority order, and say which Chapter 6 clause each one echoes.
-
The rise in peak ticket times from 15 to 23 minutes never appeared on any P&L. Design the one metric and one threshold you would put on a weekly report to catch that cost in month two rather than year three. What action does crossing the threshold trigger, and who has authority to trigger it?
-
Several cities capped delivery commissions during the emergency. Using this case, argue both sides: what did the caps protect, and what would a permanent cap do to the availability of the channel for a small operator? Be honest about the uncertainty.
-
Apply it to the plan. Bellwether's small-format contingency (this chapter's checkpoint) proposes a delivery-only second brand out of Bellwether's own kitchen at roughly \$39,000 of annual contribution, and explicitly rejects a separate facility. Using this case, list the three specific guardrails you would attach to that recommendation before it goes in the plan, and state the metric that would tell the partners to shut it off.