Chapter 28: Delivery, Takeout, and Off-Premise: The Channel That Changed the Industry (and How to Make It Profitable)

"The app says thirty minutes. The kitchen says twelve. The driver says he's four blocks away, and he has been four blocks away for nine minutes. Guess whose review it is." — constructed; a general manager, standing at the pass, describing the channel

Overview

Somebody is going to tell you that delivery is free money. The argument sounds airtight: your rent is already paid, your hood is already running, your salaried chef is already standing there, so every incremental order absorbs fixed cost you have already committed to. Take the order. Take all of them.

The argument is not wrong. It is conditional, and almost nobody states the condition. Incremental revenue absorbs fixed cost only when the resource it consumes was going to sit idle. The moment the channel starts consuming a resource that something else wanted — a hearth slot, a pair of hands at the pass, the expediter's attention at 7:40 on a Saturday — it stops being incremental and starts being substitution. And it substitutes at roughly half the rate, because you sold food without selling a seat, a bottle of wine, or an evening.

So the honest question for this chapter is not "how do we grow delivery." It is a harder one, and it is the one the restaurant in this book actually has to answer: should we do this at all?

Bellwether is a 68-seat, wood-fired, chef-driven room with twenty-two dinner items and a line of four people at peak. There is no fifth person to make delivery orders. Its revenue bridge contains exactly one takeout line — $31,200 a year, fifty-two weeks at $600 — which is 2.0% of the plan's $1,550,000. That line was written before anybody costed it. By the end of this chapter you will have costed it, and you will either defend it or replace it with something you can prove.

Along the way you will learn a way of thinking that outlives any particular platform. Every channel a restaurant adds — delivery, catering, retail, a second daypart — is a separate business line with its own margin structure that happens to share your kitchen. The discipline is to price each one honestly, on its own numbers, against the constraint it consumes. Operators who do this find one or two channels that genuinely pay. Operators who don't find themselves doing 30% of their volume at a contribution rate that quietly funds somebody else's growth.

In this chapter, you will learn to:

• Trace how off-premise moved from a side channel to a structural one, and name what it changed about the economics of a restaurant — including the two changes that have nothing to do with commission.
• Compute the true contribution of a specific order from a specific menu, after commission, packaging, packing labor, processing, and refunds, and compare it to the same food dining in.
• Read a platform payout statement and find the gap between the commission line and the effective take rate.
• Compute a break-even cannibalization rate and design an experiment that tests incrementality instead of assuming it.
• Choose between marketplace and first-party ordering on reach, data, cost, and control — and understand what "who owns the guest" costs in dollars per year.
• Sort a menu by how it travels, and compute the price uplift a commission rate would require.
• Protect the dine-in guest operationally, and know the volume at which off-premise earns a dedicated position.

Learning Paths

🏗️ Opening — this is a plan decision, not an operations decision. Weight §28.2, §28.3, and the Business Plan checkpoint. Do the arithmetic on your own menu before you sign anything. 📋 Managing — weight §28.6 and §28.8. If you already run a delivery program, §28.2 will tell you what it is really contributing and §28.3 will tell you whether it is adding anything at all. 🍸 Beverage — off-premise is the channel where your best margin disappears: no wine, no cocktails, no second round. §28.2 quantifies it. §28.9 covers alcohol-to-go, which varies enormously by jurisdiction. 🚚 Small Format — this chapter is nearly your whole business model. Read all of it, and read §28.4 twice: a small format that does not own its guest list owns nothing. Chapter 30 builds on it.

28.1 How off-premise became structural, and what it did to restaurant economics

For most of the twentieth century, off-premise dining was a narrow, well-understood category. Off-premise simply means food prepared in your kitchen and consumed somewhere else — takeout, curbside, drive-through, delivery, catering. In a full-service American restaurant it was pizza, it was Chinese food, it was a paper menu wedged under a door, and it was a small percentage of a small number of restaurants' sales. If a neighborhood bistro delivered at all, it delivered with its own staff in somebody's Corolla, within eight blocks, and reluctantly.

Two things changed that, in sequence.

The first was aggregation. Beginning in the mid-2010s, third-party marketplaces — DoorDash, Uber Eats, and Grubhub are the ones an American operator will actually meet — built a demand-side product that was genuinely good. The guest gets one app, every restaurant in the area, no phone call, a saved card, a live map, and a review system. That is a real service, and it created real demand that did not exist before. Restaurants signed up because the alternative was being invisible in the place where a growing number of people were deciding what to eat.

The second was the COVID-19 shutdowns of 2020, which are among the most thoroughly documented events in the industry's history. Dining rooms were closed by public order across most of the United States. Off-premise stopped being a channel and became the channel, essentially overnight. Restaurants that had never packed a box learned to in a week. Operators who had spent a career insisting their food could not travel discovered that the alternative was zero.

Here is the part that matters for your plan: the technology was not the structural change. The habit was. By the time dining rooms reopened, a very large number of people had a delivery app on their phone with a saved card and a default address, and they had used it forty times. Habit is stickier than technology. That is why off-premise did not snap back to 2019 levels, and it is why this chapter exists as a permanent part of a restaurant-management curriculum rather than a footnote.

What it actually changed

Four things, and only the first one gets discussed.

One: it inserted an intermediary that charges a percentage of revenue. Marketplace commissions commonly run somewhere in the 15–30% range depending on the service tier you select — a basic tier that gives you a listing, a middle tier that adds delivery-fee subsidies and better placement, a top tier that adds more marketing and sometimes a delivery-radius extension. Rates vary by platform, by market, by contract, and over time, so treat any specific number you hear as one restaurant's deal rather than a law of nature. What does not vary is the arithmetic: a 25% commission levied on a business that keeps five cents on the dollar is not a fee. It is a different business model. Chapter 1 established that a full-service restaurant clears three to seven points of operating profit. There is no version of that P&L where you absorb twenty-five points of revenue and carry on as before.

Two: it took the guest's identity. In the dining room you can learn a guest's name, remember that they sat at 14 and drank Sancerre, and put it in the notes field of your reservation system. On a marketplace you get an order number and, if you are lucky, a first name. The platform has the email, the phone number, the order history, the frequency, and the ability to send that person a promotion for the restaurant across the street. Chapter 23 built the recognition systems that produce a second visit and Chapter 27 built the owned channels — the email list, the SMS list — that make marketing free. A marketplace order contributes to neither. You served a guest and did not meet them.

Three: it changed the labor shape of the business. An off-premise order consumes back-of-house labor and almost no front-of-house labor. On paper that looks like a labor-cost win, and sometimes it is. But the FOH labor you saved was the labor that sold the wine, the second cocktail, and the dessert — which is why the beverage attachment on an off-premise order is close to zero and the check average is lower than it looks. It also moved the tip. On a delivery order the tip goes to a driver you do not employ, do not train, and cannot discipline, and whose performance the guest will attribute to you.

Four: it changed the design of restaurants. Pickup shelves in the vestibule. A second entrance so drivers stop crossing the dining room. Menus written for boxes rather than plates. At the far end of that logic, kitchens with no dining room at all — ghost kitchens and virtual brands, which are Chapter 30's subject and which exist because somebody followed this channel's economics to their conclusion.

⚠️ Where the Money Leaks

"It's incremental, so any margin is good margin."

This is the most expensive sentence in the chapter, and it is half true, which is what makes it dangerous.

The fixed-cost-absorption argument is legitimate: if your rent, your salaried chef, your hood, and your walk-in are already paid for, then an order that returns anything above its variable cost makes you better off. That is real. It is why a Tuesday takeout order at Bellwether is a genuinely good idea.

The argument fails the moment the order consumes a resource that was already spoken for. Then you are not absorbing idle capacity — you are reallocating scarce capacity from a high-contribution use to a low-contribution one, and you are doing it without noticing, because both transactions show up as sales.

The test is one question: what would that resource have done in the next ten minutes if this order had not existed? If the answer is "nothing," take the order. If the answer is "a dine-in entrée," you just made a trade, and §28.2 prices it.

Why operators said yes anyway

It is worth being fair to the thousands of operators who signed marketplace agreements without running this arithmetic, because the incentives were genuinely difficult.

In 2020 the choice was frequently between a low-contribution channel and no channel. That is not a close call; a restaurant with 40% of its former revenue at 40% contribution survives, and one with zero revenue does not. Many operators are still running programs they adopted under those conditions, priced under those conditions, and never re-examined. A decision that was correct under emergency conditions is not automatically correct in year four. Re-running the numbers is not disloyalty to a decision that saved your business. It is what you owe the business now.

There is also a visibility argument, and it is not nothing. Being absent from the app where a neighborhood decides what to eat has a cost that never appears on any statement. §28.4 treats this properly, because it is the strongest argument for the marketplace and it deserves to be made well rather than dismissed.

28.2 The channel math: commission, packaging, and the true contribution of a delivery order

Enough theory. Take a real order off Bellwether's menu and follow the money.

Two people order dinner on a Thursday. They pick the signature — the Hearth Chicken at $29.00, whose cost card Chapter 11 froze at $8.52 — plus the hand-cut pasta and a salad. This is a completely ordinary order and it is representative of what the channel actually sells: entrées, no alcohol, one shared starter.

(The Hearth Chicken carries its frozen cost card. The other two are modeled at the menu's 30% food-cost target rather than costed individually — the point here is the channel, not the recipe.)

Before a single channel cost, that order carries a contribution margin of **$65.00 − $19.32 = $45.68**. Chapter 12 taught you to bank dollars, not percentages, and $45.68 is a healthy number.

Now start subtracting.

Packaging is a cost card

Most operators guess at packaging, which means most operators are wrong about it. Build it the way Chapter 11 taught you to build a recipe cost card: line by line, at the price you actually pay.

Packaging cost is the fully loaded per-order cost of every disposable that leaves the building with the food — containers, lids, bags, cutlery, napkins, sauce cups, tamper seals, and labels.

Notice the structure, because it is how you forecast packaging for a whole year: a base of $0.60 that every order carries regardless of size (bag, cutlery, seal, label) plus **$1.64 of containers that scales with the number of items. Bellwether's average off-premise order will run about 2.4 containers, so the planning figure is $0.60 + $1.25 = $1.85 per order** — call it 3.7% of a $50 ticket. That is a real cost line, it is larger than most operators assume, and §28.7 will show you why buying the cheap version of it is usually the expensive choice.

Somebody has to pack it

Packing an order is not free either. Pull the items, portion into containers, fill and lid the sauce cups, bag it, seal it, label it, stage it, and hand it to a driver. Call it four minutes of hands on a three-item order. At an illustrative fully loaded support-labor rate of $19.50 an hour — wage plus payroll taxes plus benefits, per Chapter 19's definition of all-in labor — four minutes costs $1.30.

Hold that number. §28.6 will show you that where those four minutes come from is a much harder question than what they cost.

🧮 Run the Numbers

The same $65 of food, through three channels.

The restaurant's own variable costs on this order are identical no matter how it is sold: food $19.32 + packaging $2.24 + packing labor $1.30 = **$22.86**.

Marketplace delivery, at four commission rates across the common range:

	15%	20%	25%	30%
Menu subtotal	$65.00 | $65.00	$65.00 | $65.00
Commission	($9.75) | ($13.00)	($16.25) | ($19.50)
Net remittance	$55.25 | $52.00	$48.75 | $45.50
Food, packaging, packing labor	($22.86) | ($22.86)	($22.86) | ($22.86)
Contribution	$32.39** | **$29.14	$25.89** | **$22.64
as % of the order	49.8%	44.8%	39.8%	34.8%

First-party pickup — your own ordering page, guest collects it:

Menu subtotal	$65.00
Payment processing (2.9% + $0.30) | ($2.19)
Online-ordering platform fee (2%)	($1.30)
Food, packaging, packing labor	($22.86)
Contribution	$38.65

The same two people, dining in. Two covers at the plan's $46.00 average check: $92.00, split 72/28 into $66.24 of food and $25.76 of beverage. Food cost 30% = $19.87; pour cost 22% = $5.67; total COGS $25.54, which is the plan's 27.8% blended rate. Card processing on the charged amount including a 20% tip ($110.40) is about $3.50. Linen, china, glassware, and wash: call it $0.60 a cover, $1.20.

Contribution: $92.00 − $25.54 − $3.50 − $1.20 = $61.76.

Line them up:

Channel	Contribution	Index to dine-in
Dine-in, two covers	$61.76	100%
First-party pickup	$38.65	63%
Marketplace @ 15%	$32.39	52%
Marketplace @ 20%	$29.14	47%
Marketplace @ 25%	$25.89	42%
Marketplace @ 30%	$22.64	37%

A marketplace order at 25% returns about forty-two cents for every dollar of contribution the same two people would have produced at a table. And read the top two rows again: the single largest gap in the whole table is not the commission. It is the $25.76 of beverage that nobody ordered.

That last observation is the one operators consistently miss, so it is worth stating flatly. You do not lose the most money to the platform. You lose it to the fact that nobody drinks wine in their own kitchen. Chapter 15 established that beverage is where the margin lives — 22% pour cost against 30% food cost. An off-premise order is, structurally, a sale with the high-margin half removed.

FIGURE 28.1 — Where a dollar goes, by channel                [the Bellwether plan]

  DINE-IN (two covers, $92.00)
    food & beverage cost   ███████████                28¢
    card processing        ██                          4¢
    china, linen, wash     ▌                           1¢
    ───────────────────────────────────────────────────────
    = contribution         ██████████████████████████ 67¢

  FIRST-PARTY PICKUP ($65.00)
    food cost              ████████████               30¢
    packaging              █▌                          3¢
    packing labor          █                           2¢
    card processing        █▌                          3¢
    ordering platform fee  █                           2¢
    ───────────────────────────────────────────────────────
    = contribution         ████████████████████████   60¢

  MARKETPLACE DELIVERY @ 25% ($65.00)
    commission             ██████████                 25¢
    food cost              ████████████               30¢
    packaging              █▌                          3¢
    packing labor          █                           2¢
    ───────────────────────────────────────────────────────
    = contribution         ████████████████           40¢

  Rounded to whole cents. Contribution here is BEFORE any fixed cost — rent,
  salaried labor, insurance — and before any allowance for orders that go wrong.
  The dine-in column is higher partly because of the channel and mostly because
  of the beverage: 28% of a dine-in check, and roughly 0% of a delivery order.

The commission line is not the take rate

Here is where operators who did do the arithmetic still get surprised. The percentage in your contract is the commission rate — the platform's stated share of the menu subtotal. It is not what the platform actually keeps.

The effective take rate is the total the platform retains — commission plus promotional funding, sponsored-listing spend, error refunds charged back to you, and any per-order or hardware fees — divided by your gross menu sales. On most statements it is a meaningfully larger number than the commission line, and the only way to know yours is to read the payout statement rather than the contract.

🧾 Read the Numbers

```text FIGURE 28.4 — "The weekly payout statement" [constructed teaching example] THE ARTIFACT One week of a third-party marketplace payout statement, as the operator receives it. A 70-seat neighborhood American restaurant in a mid-size market — comparable in size and concept to Bellwether — in its second year of running marketplace delivery. Contract commission: 25%. THE CONTEXT An ordinary week in March. No holidays, no weather event. The operator opted into a "$5 off $30" promotion in January and has not turned it off, and is buying sponsored placement because a competitor started to.

                 Orders                                             118
                 Gross menu sales (item subtotals)            $4,732.00
                 Average order                                    $40.10

                 Commission @ 25% of menu subtotal            ($1,183.00)
                 Promotion funding (restaurant share)            ($185.00)
                 Sponsored listing / ad spend                    ($142.00)
                 Error refunds charged back                      ($168.44)
                 Adjustments and credits                          $21.00
                 ──────────────────────────────────────────────────────
                 NET REMITTANCE                                $3,074.56

                 Restaurant's own costs on those 118 orders
                   Food cost @ 30% of menu sales               $1,419.60
                   Packaging @ $1.85 x 118                       $218.30
                   Packing labor @ $0.98 x 118                   $115.64
                 ──────────────────────────────────────────────────────
                 CONTRIBUTION                                  $1,321.02
                   per order                                      $11.20
                   as % of menu sales                              27.9%

WHAT IT SHOWS The contract says 25%. The statement says 35.0% — the platform retained $1,657.44 of $4,732.00. Ten points of that gap are things the operator chose (promotions, ads) or absorbed (refunds), not things the contract imposed. After the restaurant's own costs, 118 orders and a week of the kitchen's attention produced $1,321.02. The same $4,732.00 sold in the dining room, at the plan's 72.2% contribution rate, would have produced roughly $3,416 — about 2.6 times as much. WHAT IT DOESN'T It says nothing about incrementality. If most of those 118 orders came from households that would otherwise have booked a table, this statement is a record of a business converting good revenue into worse revenue. It also does not show ticket-time damage to the dining room during those 118 orders, it does not show a single guest's email address, and the $168.44 refund line does not say which orders failed or why. THE DECISION Monday: turn off the standing promotion and the sponsored listing — together $327, or 6.9 points of take rate, and neither has ever been measured against orders it generated. Then pull the refund detail and find out whether the $168.44 is one systematic packing error or fifteen unrelated ones. Do not touch the commission rate; you cannot negotiate it at this volume (§28.9). THE LESSON Never manage a channel from the contract. Manage it from the payout statement, monthly, the same way Chapter 31 makes you manage food cost from a count rather than from an invoice total. ```

Two details in that statement deserve a second look.

The promotion and the ad line are opt-ins that behave like defaults. They are easy to turn on during a slow month and almost never turned off, because nothing on the platform's dashboard makes their cost salient. Six point nine points of take rate is more than most operators' entire net margin, spent without an authorization step. Chapter 34's whole argument about comp and discount authorization applies here exactly: a discount nobody had to approve is a discount nobody reviews.

The refund line is a cost of quality. At $168.44 on $4,732.00 it is 3.6% of menu sales — larger than packaging and packing labor combined. §28.8 treats it as what it is: a measurable defect rate with a dollar value attached.

⚠️ Where the Money Leaks

Commission is charged on the menu subtotal, and it does not care about your plate cost.

A 25% commission is 25% of the menu price, which means the dollar cost of the commission scales with price while your food cost scales with the recipe. That has a counterintuitive consequence: your best items subsidize the channel hardest.

The Hearth Chicken at $29.00 pays $7.25 of commission. A $12.00 salad pays $3.00. But the chicken's contribution margin is $20.48 and the salad's is $8.40, so the commission consumes 35% of the chicken's margin and 36% of the salad's — nearly identical.

Now change the item. A high-cost, low-margin item — a $34.00 fish plate at a 38% food cost, so $12.92 of plate cost and $21.08 of margin — pays $8.50 of commission, which is 40% of its margin. The rule: commission hurts in proportion to food cost. Your low-food-cost stars survive the channel best; your expensive-protein items get eaten alive. Chapter 12's matrix does not become irrelevant on a delivery menu. It becomes more important, and the quadrants move.

28.3 Incremental vs. cannibalized: the question almost nobody asks

Every number in §28.2 assumed the off-premise order was additional. If it was not — if the same household would have booked a table that night and did not, because ordering in was easier — then the channel did not add $25.89 of contribution. It subtracted the difference between what they would have spent at a table and what they spent in a box.

This is the single most important question in the chapter and it is asked almost nowhere, because it is genuinely hard to answer and because the answer is frequently unwelcome. The POS shows the delivery sales. It does not show the reservation that never got made.

Two definitions, and get them exactly right:

• Incremental sales are sales that would not have occurred at all without the channel. The household was going to cook, or order from somebody else, or eat nothing memorable. The channel created the transaction.
• Cannibalized sales are sales the channel moved from a higher-contribution channel to a lower one. The household was going to come in. Total revenue looks flat or up; contribution falls.

The break-even cannibalization rate

You can turn this from a debate into an arithmetic problem. Let the break-even cannibalization rate be the fraction of off-premise orders that could replace a dine-in visit before the channel stops adding contribution at all.

  break_even_cannibalization = off_premise_contribution_per_order
                               ------------------------------------
                               contribution_of_the_dine_in_visit_it_replaces

You need the denominator to be honest, so size the dine-in visit to the off-premise order rather than to a fixed party of two. Bellwether's modeled average off-premise ticket is $50.00, which is roughly 1.7 people's worth of food. The same 1.7 covers dining in:

Build the COGS from its two components rather than applying the blended 27.8% to the total — the shortcut gives $21.74 and a contribution of $52.42, three cents off, which does not matter here and will matter somewhere. Chapter 11's discipline holds: build the number from its parts, then check it against the ratio, never the other way around.

Now the break-evens, using the per-order contributions from §28.2 scaled to a $50 ticket (§28.9's Business Plan section does this properly; take the results on faith for a page):

Read those as sentences. If more than about a third of your marketplace orders would otherwise have been dine-in visits, the marketplace is destroying contribution — even though sales are up. First-party pickup gets a much wider margin of error: it can cannibalize more than half its own volume and still pay.

That difference — 34% versus 55% — is the entire argument for owning your ordering channel, expressed as a tolerance for being wrong. The cheaper channel is not just more profitable. It is more forgiving of the thing you cannot measure.

🧮 Run the Numbers

What cannibalization costs, on Bellwether's line.

The plan books 624 off-premise orders a year (12 a week × 52). Assume first-party pickup at $28.82 of contribution each, and a dine-in visit worth $52.45.

Cannibalization rate	Contribution per order, net	Annual contribution
0% (fully incremental)	$28.82 | $17,984
15%	$28.82 − $7.87 = $20.95 | $13,073
25%	$28.82 − $13.11 = $15.71 | $9,803
40%	$28.82 − $20.98 = $7.84 | $4,892
55% (break-even)	≈ $0 | ≈ $0
70%	$28.82 − $36.72 = ($7.90) | **($4,930)**

Now run the same thing on a 25% marketplace at $18.07 a order:

Cannibalization rate	Net per order	Annual contribution
0%	$18.07 | $11,276
25%	$18.07 − $13.11 = $4.96 | $3,095
34% (break-even)	≈ $0 | ≈ $0
50%	$18.07 − $26.23 = ($8.16) | **($5,092)**

The same channel, the same sales line, and a swing of more than twenty thousand dollars depending entirely on a variable nobody measures. The full swing is $23,076 — from $17,984 down to a loss of $5,092. On a business whose entire operating profit at 16.8% of $1,550,000 is $261,020 before debt service, that is nearly nine percent of the profit, decided by an assumption nobody wrote down.

How to actually measure it

Here is the method. None of it is difficult; all of it requires deciding in advance to look.

1. The baseline test. Before you launch, record dine-in covers by daypart for eight weeks. After you launch, record them for eight more. Compare dine-in covers, not total sales — total sales will rise by construction and tell you nothing. Control for the obvious confounds: season, weather, a marketing push, a competitor opening, a road closure. This is crude, it is confounded, and it is still better than the nothing most operators have.

2. The constraint test. Ask whether the channel operates during hours when any resource is binding — the fire, the line, the seats, the parking, the host stand. If yes, assume high cannibalization by default and make the channel prove otherwise. If no, assume high incrementality. This is the cheapest test and it usually gives you the right answer. §28.6 applies it to Bellwether.

3. The geography test. Plot the delivery addresses. Orders arriving from outside your walk-in trade area — Chapter 2's radius — are much more likely to be genuinely new demand; you could not have served those households at all. Orders from the six blocks that already walk in are the ones to worry about. If 60% of your delivery volume comes from within four blocks, you are largely delivering to your own regulars.

4. The guest-overlap test. Match off-premise phone numbers and emails against your reservation book and your loyalty list. High overlap is a cannibalization signal. On a marketplace you often cannot run this test at all, which is itself informative: a channel that prevents you from measuring cannibalization has made an assumption on your behalf.

5. The daypart-shape test. Chart when off-premise orders arrive against when dine-in demand arrives. Orders that cluster in your dine-in peak are the expensive ones. Orders in the 5:00 shoulder or after 9:00 are mostly free. This chart takes twenty minutes to build from POS data and changes more decisions than any other single artifact in this chapter.

6. The blackout experiment. The strongest test, and almost nobody runs it: turn the channel off for two weeks on Friday and Saturday only, keep it on Tuesday through Thursday, and measure dine-in covers on the blacked-out nights against the same nights the prior month and the same nights last year. If dine-in covers rise, you were cannibalizing. If they do not move, the channel was incremental and you have just proved it with evidence rather than argument.

The blackout is uncomfortable because it means deliberately refusing revenue, and because the platform's algorithm may punish a pause with lower placement when you return. Both are real costs. They are also small compared to running a channel for three years without knowing whether it adds anything.

🔍 Check Your Understanding

1. A restaurant's delivery sales grew from $0 to $6,000 a month while total sales stayed flat. What happened to contribution, and why is "sales are flat, so it's a wash" wrong?
2. A first-party pickup order contributes $31 and the dine-in visit it might replace contributes $58. What is the break-even cannibalization rate?
3. Why does the constraint test (method 2) usually give you the right answer faster than the baseline test (method 1)?

(1: Contribution fell. Flat sales with a channel shift means dollars moved from ~67¢ of contribution to ~40¢; on $6,000 a month that is roughly $1,600 a month of contribution gone with no change in revenue. 2: $31 ÷ $58 = 53%. 3: Because cannibalization is caused by resource contention, and you can observe whether a resource is contended today — you do not need sixteen weeks of confounded before-and-after data to know that your fire is over rate at 7:30 on a Saturday.)

28.4 Marketplace vs. first-party: reach, data, cost, and control

There are two fundamentally different ways to sell food for consumption elsewhere, and confusing them is the most common strategic error in this chapter.

A third-party marketplace is a platform that lists many restaurants, owns the guest relationship, takes the order, processes the payment, dispatches a driver, and remits you a net figure. DoorDash, Uber Eats, and Grubhub are the ones you will meet. You are a supplier on somebody else's shelf.

First-party ordering is a guest ordering directly from you — your own website, your own ordering page, a link on your Google Business Profile, a QR code on the check. You own the menu, the pricing, the guest data, the payment relationship, and the service recovery. Direct delivery is the sub-case where you also handle the drive, either with your own employed drivers or by buying a white-label delivery-as-a-service drop from a logistics provider.

Here is the honest comparison.

That last row is the whole thing. Neither channel is better in the abstract. The marketplace sells you demand; first-party sells you margin. An operator with plenty of demand and no margin should not be on a marketplace. An operator nobody has heard of, in a market where the app is how people decide, may genuinely need one.

The marketplace as an acquisition channel

Here is the framing that makes the decision tractable, and it comes straight from Chapter 27's cost per cover acquired.

Stop thinking of the commission as a fulfillment cost. Think of it as an acquisition cost, and then ask whether you are acquiring anything.

🧮 Run the Numbers

What a marketplace order is worth if you convert the guest — and if you don't.

A $50.00 order on a 25% marketplace costs you $12.50 in commission. Treat that as the price of meeting a household you would not otherwise have met.

If they never convert: you pay $12.50 every single time they order, forever. Over four orders a year for three years, that is $150.00 of commission for twelve transactions, and at the end you still do not have their email address.

If they convert to first-party: say the bag contained a card — "order direct next time, 10% off, here is the QR" — and they do. You paid $12.50 once. They then order four more times a year at $28.82 of contribution each, which is **$115.28 a year**, and you can email them about the patio opening.

The marketplace is a rational acquisition channel and an irrational fulfillment channel. The conversion rate is what decides which one you are running. If you list on a marketplace and do not put a conversion mechanism in every single bag, you have chosen the irrational version by default.

One honest caveat: platform agreements sometimes restrict what you may insert or how you may solicit. Read yours (§28.9), and treat a clause that forbids you from identifying yourself to your own guest as material information about the relationship.

What "who owns the guest" costs, in dollars

Abstractions are cheap. Put a number on it.

Bellwether's planned off-premise line is **$31,200** — 624 orders a year at a $50 average ticket. Run that identical volume through both channels:

Difference: $6,708 a year, on identical sales, for identical food, out of the same kitchen. That is 21.5% of the entire revenue line, and it is what the reach is costing. On a business plan with a $1,550,000 top line, $6,708 is not decisive on its own. But it is roughly the annual cost of the scheduling and inventory software in Chapter 26's stack, and no operator would sign that contract without reading it.

🤝 Hospitality

You cannot recover a service failure you never hear about.

Chapter 23 made the case that service recovery is one of the highest-return activities in a restaurant: a guest whose problem is fixed well often becomes more loyal than a guest who never had one. The mechanism requires two things — you find out, and you can do something.

A marketplace order breaks both. The guest reports the problem to the app. The app issues a refund and charges it to you. You learn about it, if at all, as a line item on a statement eleven days later, with no name, no contact, and no opportunity. The guest's experience is: the restaurant sent me the wrong thing and a company I don't work for gave me my money back. You paid for the apology and received none of its value.

On a first-party order the same failure is a phone number in your system. You call. You apologize. You put a credit on the account and a note on their profile. Chapter 23's arithmetic says that guest is now more likely to return than one who never had a problem.

This is the part of "who owns the guest" that never shows up in a commission calculation, and it may be the largest part. Off-premise strips out the room, the server, the music, the moment the plate lands — everything this book means by hospitality. What is left is the food and the way you handle it when the food is wrong. If you cannot handle it, you have nothing left.

The hybrid most operators land on

For what it is worth, the arrangement that survives contact with reality in most independents is not either/or:

1. List on one marketplace, not three. Multiple tablets, multiple menus, and multiple 86 lists is how accuracy dies (§28.8). Pick the one with the demand in your specific neighborhood.
2. Treat it as marketing spend, budgeted and reviewed monthly against the plan's marketing line from Chapter 27 — not as a revenue channel that runs itself.
3. Put a conversion card in every bag, with an offer that is worth the commission you paid.
4. Build first-party properly — integrated with the POS so the 86 list is live, with the ordering link everywhere Chapter 27 put you: the Google Business Profile, the email footer, the check presenter, the window.
5. Measure the mix quarterly. The number that tells you whether the strategy is working is not total off-premise sales. It is the share of off-premise orders that came direct, and it should rise every quarter or the acquisition thesis is false.

28.5 Menu strategy for delivery: what travels, what doesn't, and price parity

A menu written for a plate that will be carried nine feet is not a menu written for a box that will sit in a bag for twenty minutes. Almost every quality failure in off-premise is decided at the menu stage, not in the kitchen.

The physics are simple and unforgiving. Hot food in a closed container generates steam. Steam condenses on the lid, runs back down, and lands on whatever is underneath it. Anything crisp becomes soft. Anything soft becomes wet. Meanwhile the food keeps cooking in its own residual heat — carryover that a plate dissipates into a dining room and a container traps. Emulsified sauces sit and break. Dressed leaves wilt. Starch retrogrades and sets.

FIGURE 28.2 — The twenty-minute clock                       [constructed teaching example]

  minutes after the item leaves the pass
   0      4      8      12     20     30     45
   |      |      |      |      |      |      |
   crisp skin / fried crust ...X                    gone by about 4 minutes. No exceptions.
   dressed leaves ..........X                       wilt starts ~6, collapse by ~10
   fresh hand-cut pasta ........X                   carryover cooking; starch sets ~10-12
   emulsified or mounted sauce .....X               breaks or separates ~12-15
   whole-muscle roast, sliced ...............X      still good at 20; tired by 30
   braise, stew, ragu .........................X    essentially unharmed to 45
   set dessert (tart, custard, cake) .........X     fine. Anything that melts is not.

   X marks the point at which the guest is eating a materially different dish
   from the one you would send to a table.

   And remember the staging: 6-10 minutes from pass to driver on a busy night,
   THEN the drive. Twenty minutes is an optimistic total, not a typical one.

Bellwether's menu, honestly assessed

Twenty-two dinner items. Not twenty-two off-premise items.

The Hearth Chicken travels — with a compromise you have to name out loud. The meat is fine; a half bird holds heat well and arrives hot. The roasted roots are better than fine — they were cooked in a hearth and they hold. The salsa verde must go in a sauce cup, not on the plate, or it steams the skin and turns gray. And the skin is the casualty. The crackling, blistered, wood-fired skin that is the entire reason this dish is the signature is gone by minute four regardless of what you do, because physics. You can vent the container, you can hold it in a perforated insert, you can put a paper liner underneath — those buy you texture at minute eight instead of minute four. They do not buy you the dish.

So the question is not "can we ship the Hearth Chicken." It is "do we want the version of the Hearth Chicken that arrives in a box to be some guests' only experience of it?" That is a brand question (Chapter 3), not a margin question, and it deserves a real answer from the chef-owner rather than a default yes.

The hand-cut pasta does not travel, and it should not be on the list. It fails on three separate mechanisms at once: carryover cooking takes it past al dente somewhere around minute ten, the sauce breaks, and the starch sets so that what arrives is a solid mass. There is no packaging solution. A restaurant that ships hand-cut pasta on a twenty-minute drive is shipping a bad review with a delay fuse.

FIGURE 28.5 — THE OFF-PREMISE MENU MATRIX                   [constructed teaching example]

                        TRAVELS BADLY              TRAVELS WELL
                 ┌─────────────────────────┬─────────────────────────┐
  HIGH           │      BRAND RISK         │        THE LIST         │
  contribution   │  ships a worse version  │  put it on, photograph  │
                 │  of your best dish;     │  it, and let it carry   │
                 │  fix the packaging or   │  the whole channel      │
                 │  leave it off           │                         │
                 │  (the Hearth Chicken)   │  (braises, roasts,      │
                 │                         │   roasted vegetables)   │
                 ├─────────────────────────┼─────────────────────────┤
  LOW            │        NEVER            │       THE FILLER        │
  contribution   │  you lose the guest     │  safe, doesn't pay;     │
                 │  AND the margin. There  │  use it only to let a   │
                 │  is no argument for     │  guest build a complete │
                 │  this quadrant.         │  meal                   │
                 │  (fried items, crudo,   │  (grain salads, breads, │
                 │   dressed leaf salads)  │   set desserts)         │
                 └─────────────────────────┴─────────────────────────┘

  Same axes as Chapter 12's engineering matrix, with popularity replaced by
  travel behavior — because on this channel, an item that arrives wrong has a
  negative popularity: it costs you the next order too.

Work the quadrants the way Chapter 12 taught you:

• The List is your off-premise menu. At Bellwether that is a short set: the wood-roasted vegetables, a braise, a grain salad, the bread service, a set dessert, and one or two roasts served sliced with the sauce separate. Call it nine of the twenty-two items.
• Brand Risk items get one of three dispositions, decided deliberately: re-engineer the packaging and accept a documented compromise; re-engineer the dish for the channel (sauce on the side, a different garnish, a finish-at-home instruction); or leave it off. All three are defensible. The default — ship it and hope — is not.
• Filler items exist so a guest can assemble a real meal. They do not need to be exciting; they need to be right.
• Never is not a judgment call. Fried items, crudo, and dressed leaf salads do not go in a box on a thirty-minute journey, and no packaging vendor is going to solve it for you.

👨‍🍳 On the Line

Testing it, which takes one afternoon and almost nobody does.

Before you publish an off-premise menu, run this. It is the cheapest quality-control exercise in the book.

Fire every candidate item at 2:00 on a Tuesday. Pack each one exactly as it would go out — same container, same lid, same bag, same sauce cup. Put the bags in a car. Drive a real route: twelve minutes out, park, wait four minutes the way a driver waits, twelve minutes back. Do not open anything.

At 2:30, unpack everything onto plates and eat it — the chef, the sous, the FOH partner, and one server who will have to describe it to guests. Rate each item honestly on three axes: temperature, texture, and appearance. Photograph what it actually looks like coming out of the box, not what it looks like on the pass.

Two things will happen. First, three or four items you assumed were fine will be visibly bad, and one item you assumed would fail will be perfect. Second, somebody will say "we could fix that with a different container," and they will usually be right, and that is exactly the conversation §28.7 exists to have.

Repeat it every time the menu changes seasonally. Ten items, ninety minutes, four people. Compare that to the cost of a season's worth of two-star ratings on a dish that never had a chance.

Price parity, and the uplift arithmetic

Delivery price parity means charging the same menu prices on an off-premise channel as in the dining room. The alternative — an off-premise uplift — means listing higher prices on the channel to recover its costs.

Both are defensible; the arithmetic tells you what the uplift would have to be, and the number is usually sobering.

To hold contribution constant, the listed price must satisfy:

  listed_price = (dine_in_price + per_item_packaging_and_labor) / (1 - commission_rate)

Run it on the Hearth Chicken at a 25% commission. Its share of packaging and packing labor is $1.20 — the large vented container $0.62, one sauce cup $0.09, its share of the base bag/cutlery/seal $0.20, and $0.29 of packing labor.

| Listed price | Commission @ 25% | Remitted | − plate $8.52 | − pack/labor $1.20 | Contribution | vs. dine-in CM $20.48 | |---|---|---|---|---|---|---| | $29.00 (parity) | $7.25 | $21.75 | $13.23 | | $12.03 | 59% | | $33.00 (+13.8%) | $8.25 | $24.75 | $16.23 | | $15.03 | 73% | | $40.27 (+38.9%) | $10.07 | $30.20 | $21.68 | | $20.48 | 100% |

**To make a delivered Hearth Chicken contribute what a plated one contributes, you would have to list it at $40.27 — a 39% increase.** Nobody is paying $40 for a delivered half chicken from a neighborhood restaurant. That single line is very close to a complete answer to the delivery question for this menu.

And the table is generous, because it compares contribution on the chicken alone. The dine-in guest who ordered that chicken also bought a glass of wine.

⚠️ Where the Money Leaks

The uplift you can actually charge, and the trust it costs.

Most operators who use an uplift land somewhere around 10–15% — enough to recover a meaningful share of commission, small enough that it does not read as gouging. On a 25% commission that recovers roughly half of what you gave up, and you should model it that way rather than pretending it is neutral.

Three things to know before you do it:

• Guests notice. Your dine-in menu is on your website and on the wall. A regular who sees $29 in the room and $33 on the app draws a conclusion, and it is not "ah, commission structures."
• Platform agreements have historically contained price-parity language, and where they do, an uplift may be a contract issue. Read yours.
• Some jurisdictions have taken up menu-price transparency on delivery platforms. Rules differ and change; verify locally before you build a pricing strategy on top of one.

The honest position, and the one that survives a regular reading your menu: an uplift is a fee disclosure, so disclose it. A line on your own site — "prices on delivery apps are higher to cover their commission; order direct and pay our menu price" — turns a discovered discrepancy into a reason to use your first-party channel. That is Chapter 27's owned-channel argument doing real work.

28.6 Operations: throughput, staging, driver flow, and protecting the dine-in guest

This is where the chapter stops being financial and starts being physical, and it is where Bellwether's answer actually gets decided.

Throughput impact is the reduction in a kitchen's dine-in production capacity caused by off-premise production competing for the same constrained resource. It is a capacity cost, it is almost never measured, and at Bellwether it is the largest cost in the entire channel — larger than commission, larger than packaging, larger than labor.

Find the constraint, then price it

Chapter 10 flagged the throughput risk when the menu was built: the hearth sustains 28 items an hour, and 44% of entrées fire to order on it. Chapter 22 established the shape of the week — Tuesday through Thursday the room is demand-constrained (there are empty seats and nobody to put in them), while Friday and Saturday are capacity-constrained. Chapter 24 put the operating ceiling at about 132 covers a night.

Put those together for a Saturday.

FIGURE 28.3 — The Saturday hearth at the binding hour        [the Bellwether plan]

  7:00-8:00 p.m., the busiest hour of the week

    46 covers seated in the hour
    46 entrees ordered
    x 44% fire on the hearth              = 20 hearth entree items
    + flatbread, wood-roasted vegetables,
      hearth bread service                =  9 hearth support items
                                            ─────────────────────────
    TOTAL HEARTH LOAD                       29 items in the hour

    sustainable rate  ████████████████████████████       28 / hour
    actual load       █████████████████████████████      29 / hour   = 104%

  The fire is already over rate. It is not "busy." It is producing more than it
  can sustain, which is why ticket times stretch, why the pass backs up, and why
  Chapter 22 found the room 43% empty on average across the service while the
  kitchen was drowning: demand arrives in a two-hour spike the fire cannot absorb.

Now price a hearth slot. During that hour, 46 covers at the plan's $46.00 check produce $46.00 − $12.77 of COGS = $33.23 of contribution each**, or **$1,528.58 for the hour. Those 46 covers consumed 29 hearth items. So the fire, at the binding hour, is earning:

  contribution per hearth item = $1,528.58 / 29 items = $52.71

That figure is the correct denominator for every off-premise decision on a Saturday night, and it is the number the fixed-cost-absorption argument in §28.1 forgets to compute.

🧮 Run the Numbers

What one delivery order costs at 7:40 on a Saturday.

A marketplace order comes in at 7:40 p.m. containing a Hearth Chicken. It occupies one hearth slot. From §28.2, at a 25% commission that order contributes $25.89.

The hearth slot it consumed was, during that hour, generating $52.71 of contribution.

Contribution earned by the delivery order	$25.89
Contribution the hearth slot was producing	($52.71)
Net effect on the business	($26.82)

That order made the restaurant $26.82 worse off, and both the POS and the platform statement will record it as $65.00 of sales.

Now the same order at 6:15 on a Tuesday. The fire is running at roughly 45% of rate. The slot it consumes was going to sit idle. The displacement cost is zero, and the order's $25.89 — or $38.65 if it came through your own site — is real, incremental contribution against fixed costs you have already paid.

Identical order. Identical food. Identical commission. A swing of $52.71 in what it does to your business, decided entirely by the clock.

This is theme four of this book — every seat-hour is inventory you can't store — extended one step. At Bellwether the perishable inventory is not only the seat. It is the hearth-minute, and on Friday and Saturday the hearth-minute is scarcer than the seat.

That is the whole answer, and it is not "yes" or "no." It is a schedule.

FIGURE 28.6 — The off-premise window                          [the Bellwether plan]

  Hearth load as a % of the 28-item/hour sustainable rate.
  ● = off-premise ordering open

              5-6pm     6-7pm     7-8pm     8-9pm     9-10pm
   TUE     ●  25%    ●  45%    ●  60%    ●  50%    ●  25%     fire idle all night
   WED     ●  25%    ●  45%    ●  65%    ●  50%    ●  25%     fire idle all night
   THU     ●  30%    ●  55%    ●  80%    ●  60%    ●  30%     headroom, narrowing
   FRI     ●  40%       75%      100%       95%    ●  45%     closed 6:00-8:45
   SAT     ●  45%       85%      104%       95%    ●  50%     closed 6:00-8:45
                                    ^
                          over rate; every hearth slot
                          is already spoken for

  The channel is open 23 of the 25 service hours in a week and closed for the
  two-and-three-quarter hours that matter. That is not a compromise. It is the
  entire difference between a line that adds $18,000 and one that subtracts.

Who packs it

Now the harder question, and the one that decides whether the line survives at all.

Chapter 14 built Bellwether's line: four people at peak. Grill/hearth, sauté, garde manger, and the chef expediting at the pass. There is no fifth person. So when an order needs to be portioned into containers, sauced, lidded, bagged, sealed, labeled, and staged, one of four things happens, and each one has a cost:

🧮 Run the Numbers

The fifth person, priced honestly.

Hire someone to pack off-premise orders. Five hours a night, five nights, at the illustrative all-in support rate of $19.50 an hour:

text $19.50/hr x 5 hrs x 5 nights = $487.50 per week

Bellwether's planned off-premise line is 12 orders a week. Each contributes $28.82 *after* $0.98 of absorbed packing labor, so before that labor each contributes $29.80:

text 12 orders x $29.80 = $357.60 of weekly contribution

The dedicated packer costs $129.90 a week more than the entire channel produces. Not "reduces the margin." Consumes it, plus a hundred and thirty dollars.

Work out the volume at which a dedicated position starts to pay: $487.50 ÷ $29.80 = 16.4 orders a week just to cover the wage, contributing nothing. Somewhere north of 30 orders a week — about $1,500 of weekly off-premise sales, roughly 5% of Bellwether's total sales — the position starts to earn a modest positive return, and even there it is thin.

The rule of thumb, labeled illustrative: off-premise does not justify dedicated labor until it is somewhere around 5% of sales. Below that, it either gets absorbed by existing labor with genuine slack — or it should not exist.

And "existing labor with genuine slack" is a precise phrase. It means Tuesday at 6:15, not Saturday at 7:40. The finance answer and the operations answer converge on exactly the same window.

Staging and driver flow

Two physical problems, both of which are solved in the floor plan (Chapter 7) or not at all.

Staging. A packed order that sits waiting for a driver is a dish getting worse on the twenty-minute clock from Figure 28.2, and the clock started when it left the pass, not when the driver arrived. Six to ten minutes of staging on a busy night is normal. Design for it:

• A dedicated pickup shelf, not the pass and not the service station. Orders on the pass get bumped, mixed up with dine-in food, and picked up by the wrong person.
• Within sight of the host stand, so somebody is accountable for handing it over and nobody walks out with a bag that is not theirs. Unattended shelves generate a specific and expensive failure: the order that was stolen, which you refund and rate poorly on.
• Fire to the driver's ETA, not to the order time. This is the single highest-leverage operational fix in off-premise and most kitchens do the opposite. If the platform says eleven minutes out, the chicken should hit the fire so that it finishes at minute nine, not at minute one. It requires a separate ticket queue and a discipline about reading the ETA, and it converts a twenty-eight-minute food-age into a fifteen-minute one.
• No heat lamps on packed orders. A lamp on a closed container cooks the food and drives more condensation. Insulated bags are better than heat; they hold what you have rather than adding to it.

Driver flow. A delivery driver walking through your dining room in a branded jacket, looking at tables, asking a server where the bags are, is a hospitality failure that your guests will register and your reviews will reflect. Bellwether's 2,800 sq ft second-generation space has a fixed shell, so the question for the plan is concrete: is there a route from the street to a pickup point that does not cross the guest path? If yes, the answer is a small vestibule shelf and clear signage. If no, the answer is a shelf immediately inside the door in the host's sightline, and the honest acknowledgment that on a busy Saturday there will be a driver standing in your entry — which is another reason the channel closes from 6:00 to 8:45 on Friday and Saturday.

👨‍🍳 On the Line

The 86 problem, which is the one that actually generates the refunds.

It is 8:10 on a Thursday. The chicken is 86'd — you have four birds left and eleven on order. The chef calls it, the expediter calls it, the servers stop selling it, and the room adapts in ninety seconds. That is Chapter 14's system working exactly as designed.

The platform does not know. The platform will keep selling the Hearth Chicken until somebody walks to a tablet, finds the item, and toggles it off — and on a busy pass, "somebody" is a person who does not exist. So at 8:14 an order comes in for a chicken you do not have, and now you have three bad options: substitute without asking (a refund and a one-star), cancel the order (a refund, a one-star, and a ding to your acceptance metrics), or call the guest through the platform's system, which sometimes works.

This is why a POS-integrated ordering channel is worth real money. When the item is 86'd in the POS, the first-party menu updates instantly and the integration pushes it to the marketplace, if the integration exists and works. When it isn't integrated, your 86 list lives in two places, one of which is a tablet nobody is holding.

Ask this question in every technology demo (Chapter 26): when I 86 an item at the POS, how many seconds until it disappears from every channel, and what happens to an order that was already in the cart? Watch them answer it. The answer tells you more about the product than the price does.

28.7 Packaging: cost, sustainability, and the food arriving as intended

Packaging is the only part of your product that the guest touches before the food, and in most restaurants it is the only line item purchased entirely on price.

Three jobs, in order of how often they get forgotten:

1. Deliver the food in the condition you cooked it. Vent what should stay crisp, seal what should stay hot, separate what should not touch.
2. Survive the journey. A container that leaks in a bag ruins the order, the bag, sometimes the driver's car, and always the review.
3. Represent the restaurant. This is a brand surface (Chapter 3). A guest who has never been in your dining room forms their entire impression of you from a bag on a kitchen counter.

The cost, honestly

From §28.2, Bellwether's modeled three-item order carries $2.24 of packaging and the planning average is $1.85** per order. Against a $50 average ticket that is 3.7% of off-premise sales** — a real line, and one that belongs in your chart of accounts as its own account rather than buried in "supplies," so that Chapter 31's flash report can see it move.

Two things distort packaging budgets predictably:

Under-counting the base. Operators price containers and forget bags, cutlery, napkins, sauce cups, lids, seals, and labels. That base was $0.60 of a $2.24 order — 27% of the packaging cost, in items that cost pennies and are ordered by somebody who is not looking at the total.

Over-counting the savings from cheap containers. Save $0.14 a container by buying the thinner grade, on 624 orders averaging 2.4 containers, and you save $210 a year. One leaked order costs you a $50 refund and a rating; two a month costs you $1,200 a year plus the ratings. This is a false economy with a legible number attached, which makes it a good teaching case for Chapter 13's principle: a spec is a decision about total cost, not about unit price.

⚠️ Where the Money Leaks

Utensils in every bag.

Roughly 70–80% of off-premise orders are eaten at home, where the household owns forks. At $0.22 a set on 624 orders, unconditional cutlery costs Bellwether about $137 a year, of which perhaps $100 goes straight into a drawer or a bin.

That is a trivial number at Bellwether's volume and a serious one at scale — which is exactly why several jurisdictions now require that single-use utensils be provided only on request, and why most platforms have a guest-facing toggle for it. Turn the toggle on. Default it to off. You save money, you reduce waste, and you comply with a rule that may already apply to you.

The same logic runs through the whole category: the cheapest packaging is the packaging you did not send. No bag inside a bag. No lid on a lidded container. No sauce cup for a dish with no sauce.

Sustainability, without the greenwash

Chapter 38 handles sustainability properly. Here is what belongs in this chapter.

Off-premise is, environmentally, the worst thing a restaurant does per dollar of revenue: every order generates single-use packaging and a vehicle trip. That is not an argument against the channel, but it is an argument for being straight about it.

The honest options, in rough order of impact:

• Send less. Right-sized containers, no unnecessary components, utensils on request. Costs nothing, saves money, works immediately.
• Match the material to the disposal reality in your market. Compostable containers are only compostable where commercial composting exists and accepts them. Recyclable containers are only recycled where they are collected and where food residue does not disqualify them. A compostable container in a market with no organics collection is a landfilled container that cost you more. Find out what your municipality actually does before you buy the more expensive box.
• Comply with the material bans. A growing number of states, counties, and cities restrict expanded polystyrene foam service ware, require recyclable or compostable containers, or mandate utensils-on-request. These vary enormously and change; verify locally.
• Say only what is true. "Compostable where facilities exist" is accurate and unglamorous. "Eco-friendly packaging" on a container that goes to a landfill is a claim a guest can check, and the cost of being caught is larger than the credit for being green. Chapter 38 makes this case at length.

⚖️ Code and Compliance

Food safety does not stop at your door, and neither does your liability.

Chapter 25's framework applies to off-premise with one hard difference: once the bag leaves the building you have lost control of time and temperature, and the FDA Food Code framework the health department applies to you — cold holding at or below 41°F, hot holding at or above 135°F, poultry cooked to 165°F, and the temperature danger zone in between — does not care that a driver stopped for gas.

What a responsible operator does:

• Set a delivery radius and defend it. Your radius is a food-safety control, not a marketing decision. If the drive is thirty-five minutes, the food is not going to be right and it may not be safe.
• Package hot and cold separately. Never in the same bag. This is the most common violation of basic sense in the category.
• Use tamper-evident seals, and record that you did. Several jurisdictions require them for third-party delivery; all of them make the question "did somebody open this?" answerable.
• Label allergens on the container. There is no server to have the conversation Chapter 25 describes. On an off-premise order, the label is the conversation. Name the nine major allergens where they are present, and treat an allergen note on an order ticket as a stop-and-check step, not a preference.
• Know who is licensed for what. Rules on third-party delivery of prepared food, on required permits for the delivery service, and above all on alcohol to go and alcohol delivery vary enormously by state and locality. Many states expanded alcohol-to-go during the 2020 shutdowns and some made it permanent, frequently with conditions — sealed containers, food purchase required, ID verification at handoff, quantity limits. If your license is the one the alcohol travels under, a driver you do not employ is checking an ID on your behalf. Understand that exposure and the dram-shop implications in Chapter 8 before you sell a single cocktail to go.

All of this varies by jurisdiction and changes frequently. Verify locally, in writing, before you launch — and re-verify when the emergency-era rules in your state come up for renewal.

28.8 Accuracy, ratings, and the refund economics of the platforms

In the dining room, a mistake is recoverable in ninety seconds. The server sees the face, apologizes, fires the correct dish, and Chapter 23's arithmetic takes over: handled well, that guest may end up more loyal than one who never had a problem.

Off-premise, a mistake is discovered by a stranger in their own kitchen, twenty-five minutes after it left your hands, with no one present to fix it. You find out when the money is taken back.

Accuracy is a process control, not an attitude

The failures are boringly consistent and therefore fixable: a missing item, the wrong modification (the allergy note, the no-onion, the sauce on the side), the missing side or sauce cup, the wrong order entirely, and the order handed to the wrong driver.

Every one of those is caused by the same thing — a single person packing under time pressure with no verification step — and every one is prevented by the same thing:

1. Print or display the ticket at the pack station. Not from memory, not from the KDS across the kitchen.
2. Pack to the ticket, top to bottom, checking each line as it goes in the bag.
3. A second person reads the ticket aloud against the sealed bag before it is staged. This is the step everyone skips and the step that removes most of the defect rate. It takes twenty seconds.
4. Seal and label with the order name and item count. "3 items" written on the bag lets the driver and the guest both verify without opening it.
5. Log every failure — what was missing, which shift, which item, which packer. A defect log turns "we sometimes forget things" into "we lose sauce cups on Thursdays when the sous is off," which is a problem you can solve.

That is Chapter 14's mise en place discipline and Chapter 34's separation-of-duties principle applied to a bag. It is not sophisticated. It is just never assigned to anyone.

The refund economics

Go back to the payout statement in Figure 28.4: $168.44 of error refunds on $4,732.00 of menu sales — 3.6%. That is larger than packaging and packing labor combined, and it is entirely a quality cost.

Model it explicitly rather than discovering it. Two allowances belong in any off-premise pro forma:

That last clause matters. On many platforms, the restaurant bears the cost of a refund for problems it did not cause — food that arrived cold after a driver took a second pickup, an order the driver delivered to the wrong building, a bag left in a lobby. Dispute processes exist and you should use them, systematically, with your defect log as evidence. You will win some and not others. Budget for a residual you cannot win, because pretending it is zero is how a 25% commission quietly becomes a 35% take rate.

⚠️ Where the Money Leaks

The two ratings you are being judged on are not the same rating.

Chapter 27 taught you to manage your public reviews — the ones a guest reads before deciding where to eat. A marketplace maintains a separate score, computed from its own inputs, which drives your placement in its search results and therefore your volume on that channel. It is opaque, it usually mixes food quality with things you do not control (driver time, platform errors), and you frequently cannot see its components.

Three consequences an operator should plan for:

• You are penalized for the driver's performance. A guest rating a cold arrival is rating a thirty-four-minute drive; the score lands on you.
• Pausing or turning off the channel can hurt placement. Which is a real cost of the blackout experiment in §28.3, and a reason to run it deliberately rather than by accident.
• Operational metrics you did not know you had — acceptance rate, cancellation rate, prep-time accuracy — feed the score. Marking every order "ready" when it isn't, to keep a metric clean, is the kind of gaming that destroys the food. Set an honest prep time and hold it.

The defensible posture is the same one Chapter 27 recommends for reviews: manage the inputs you control, respond where you can, and never make an operational decision whose only purpose is to move a score you do not understand.

🔍 Check Your Understanding

1. A restaurant's marketplace statement shows 25% commission but the platform kept 34% of menu sales. Name three lines that could account for the gap.
2. Why is a first-party error allowance of 1.5% defensible when the marketplace allowance is 3–4%, given that the kitchen makes the same mistakes either way?
3. Your marketplace score is falling and the largest driver appears to be late deliveries. What can you actually change, and what should you stop trying to change?

(1: Promotional funding you opted into, sponsored-listing/ad spend, error refunds charged back, and any per-order or hardware fees. 2: Because the failure is caught at handoff while the guest is present — the cost is a remake, not a full refund plus a rating. 3: You can change prep-time accuracy, staging discipline, packaging that holds heat, and delivery radius. You cannot change driver supply or routing, and building operational decisions around it — firing early so food sits, for instance — makes the food worse to improve a number.)

28.9 Commission caps, regulation, and negotiating leverage

Off-premise is the rare corner of restaurant operations where public policy moved fast and visibly, and where the results are instructive about how two-sided markets behave.

What actually happened

During the 2020 shutdowns, when dining rooms were closed by order and off-premise was many restaurants' only revenue, several U.S. cities capped the commissions third-party platforms could charge restaurants. New York City, San Francisco, and Seattle were among them. The common shape of these ordinances was a cap around 15% for delivery services, with a few additional percentage points permitted for other services such as marketing or payment processing — but the specifics differed city by city and you should read the actual ordinance for any market you operate in rather than trusting a summary, including this one.

Most were emergency measures tied to the state of emergency. Some cities later made their caps permanent, and at least one permanent cap drew litigation from the platforms, who argued the caps were unconstitutional interference in private contracts. Platforms also responded in some capped markets by adding or raising fees on the guest side — which is the predictable outcome when you regulate one side of a two-sided market and not the other, and which is worth understanding before you campaign for a cap in your own city.

For your purposes as an operator, three practical conclusions:

1. Find out whether a cap applies where you operate, and whether it is permanent, expired, or in litigation. The answer changes your channel economics by up to ten points of revenue.
2. A cap does not make the channel free. Fifteen percent on a business that keeps five is still an enormous number, and every other cost in §28.2 is unchanged.
3. Regulation is not a strategy. A cap is a floor under a bad deal, not a reason to build your business on somebody else's platform.

⚖️ Code and Compliance

The regulatory surface of off-premise is wider than commission caps.

Several areas have drawn legislative attention and vary substantially by jurisdiction:

• Listing without consent. Platforms have historically listed restaurants they had no agreement with, sometimes with wrong menus, wrong prices, and wrong hours — and the restaurant absorbed the guest's disappointment. Several states have since enacted laws requiring a written agreement before a third-party service may list a restaurant. If you find yourself listed without one, that is worth a call to an attorney, not a shrug.
• Menu-price transparency. Some jurisdictions have taken up disclosure when platform prices differ from in-store prices. Verify locally before building a pricing strategy on an uplift.
• Marketplace facilitator tax rules. In many states the platform, not the restaurant, collects and remits sales tax on marketplace orders — which changes what you owe, what you report, and how your bookkeeper reconciles the deposit. Chapter 31's reminder applies with force: sales tax is never your money, and misunderstanding who remitted it is a way to owe it twice.
• Fee disclosure to the guest. Rules on how delivery, service, and small-order fees must be presented have been the subject of both litigation and legislation.
• Driver classification. Whether delivery drivers are employees or independent contractors has been litigated and legislated repeatedly. If you employ your own drivers, Chapter 20's wage-and-hour framework applies to them fully — including mileage reimbursement rules, which are a genuine and frequently missed liability.

Every item above varies by state, county, and city and several are actively changing. Verify locally, and use an attorney for anything with money attached.

Negotiating leverage: what you actually have

Now the unglamorous part. Independent operators routinely ask how to negotiate a better commission rate, and the honest answer for most of them is: you cannot, because you have no leverage, and pretending otherwise wastes time you could spend on the things you can control.

Leverage on a marketplace comes from exactly three places:

• Volume. A multi-unit group doing thousands of orders a week is a customer. A 68-seat restaurant doing twelve is a listing.
• Desirability in a thin market. If you are one of four restaurants of your type in a suburb, the platform needs you more than the arithmetic suggests. If you are one of ninety in a dense urban neighborhood, it does not.
• Willingness to leave. Which is only credible if you have a first-party channel that works — which is the argument of §28.4 arriving from a different direction.

Bellwether has none of the three. That is not a failure; it is a fact about its size, and it should inform the plan rather than being wished away.

What you can control, in descending order of value:

1. Which tier you buy. Tiered pricing is the platform's own published structure, and the higher tiers buy marketing and placement. Buy the lowest tier that produces orders, measure for a month, and move up only against evidence.
2. Whether promotions and ads are on. Figure 28.4's 6.9 points of self-inflicted take rate is the single largest lever most operators have and it takes two minutes to pull.
3. What is on the menu. §28.5. Removing items that travel badly improves ratings, cuts refunds, and raises the average contribution per order without negotiating anything.
4. Your radius and your hours. Both are yours to set, both are food-safety and quality controls, and both are how you keep the channel out of your binding hour.

⚠️ Where the Money Leaks

The contract terms that cost money later.

Before signing any third-party or ordering-platform agreement, find and read these clauses. Chapter 8 made the general case about contracts that auto-renew; here are the specific ones in this category.

• Term, auto-renewal, and the notice window to cancel. Diary the notice date the day you sign.
• Which fees are in the headline rate and which are extra — per-order fees, payment-processing fees, tablet or hardware rental, printer paper, "activation" charges.
• Marketing opt-ins and how they default. Whether a campaign can be enabled without a specific authorization, and who at your restaurant can enable one.
• The refund and chargeback policy, including the dispute window and what evidence is accepted.
• Exclusivity, in any form. A discount for listing on one platform only is a discount for surrendering your remaining leverage.
• Data ownership. Who owns the order data and the guest contact information, and what you are permitted to do with it.
• Menu and photo rights. Whether the platform may use, edit, or license your photography and menu content, and whether it may set prices you did not approve.
• Ratings on pause. What happens to your score and placement if you turn the channel off — which determines the real cost of the blackout experiment in §28.3.

None of these are exotic. All of them are in the agreement. Almost nobody reads it, because it arrives as a click-through at a moment when the operator is trying to solve a revenue problem quickly — which is precisely when a contract should be read most carefully.

🍽️ The Business Plan

Checkpoint 28 of 40 — the Off-Premise section.

What was on the page before this chapter

Chapter 4 built Bellwether's sales forecast bottom-up: 68 seats, 1.4 turns, a $46.00 dinner check, five dinners and two brunches a week, which totaled $1,410,760. The plan's headline year-one number is $1,550,000**, and the difference — **$139,240 — is a bridge made of lines added on top of the bottom-up build. One of those lines reads:

  Takeout        52 weeks x $600 per week  =  $31,200      (2.0% of sales)

It was written before anybody costed it. This chapter's job is to defend it or replace it.

The decision: keep the number, rebuild the line

The $31,200 stands. Everything behind it changes.

The dollar figure survives because it is the right size. Twelve orders a week at a $50 average ticket is a takeout line for a 68-seat neighborhood restaurant with a four-person line. It is small enough to absorb into existing labor on nights when the fire has headroom, and it is 2.0% of sales — too small to distort the plan and large enough to be worth building properly. The temptation in a revenue bridge is always to make a soft line bigger. We are declining it, and §28.3 explains why: this line's value swings by more than twenty thousand dollars on a variable we cannot measure until we open. A plan should not book revenue from a channel it has not tested.

What changes is the specification.

1. First-party pickup only in year one. No marketplace. No delivery.

$6,708 a year — 21.5% of the entire revenue line — for identical food and identical sales. At twelve orders a week Bellwether has no negotiating leverage (§28.9) and no volume to make a marketplace's reach worth its take rate. It also has no ability to deliver: no driver, no radius, no insurance for one, and a wood-fired menu whose signature item degrades on the twenty-minute clock.

Here is the contribution build, so a lender or a partner can check it:

2. The channel closes during the binding hour. Ordering is open all service Tuesday, Wednesday, and Thursday, and on Friday and Saturday only before 6:00 p.m. and after 8:45 p.m. — Figure 28.6. This is not a courtesy to the kitchen; it is the difference between a channel that adds $17,984 and one that destroys $26.82 of contribution every time it puts a chicken on a fire that is already at 104% of rate (§28.6).

3. A nine-item off-premise menu, drawn from the twenty-two dinner items using Figure 28.5: the braise, the roasts served sliced with sauce separate, the wood-roasted vegetables, a grain salad, the bread, and a set dessert. The hand-cut pasta is not on it and will not be. The Hearth Chicken is, with a documented compromise — vented container, salsa verde in a separate cup, and the chef-owner's explicit acknowledgment that the skin does not survive. That decision gets revisited after the first quarter against actual ratings.

4. Price parity. Off-premise prices equal dining-room prices, because there is no commission to recover on a first-party pickup order. If a marketplace is ever added, §28.5's uplift table is the starting point, and any uplift gets disclosed on our own site.

5. Labor is absorbed, not added. A dedicated packer at $487.50 a week costs $129.90 a week more than this entire channel produces (§28.6). At twelve orders a week the work is absorbed by the host and, on Tuesday and Wednesday, by a line that is not at capacity. The trigger to revisit is 30 orders a week — about 5% of sales. Below that, no position.

What this section does not settle

Incrementality is an assumption, not a finding. The plan assumes 25% cannibalization — that one in four of these orders replaces a dine-in visit that would otherwise have happened. At that rate the line's net contribution is not $17,984 but roughly **$9,803, and $7,800 of the $31,200 sales line is transfer from the dine-in forecast rather than new revenue.** The assumptions register (Chapter 4) carries both figures explicitly. We are not claiming the full $31,200 as growth, and any reader who wants to haircut the bridge by $7,800 is entitled to.

The break-even is 55% cannibalization for this channel as specified. If more than about half these orders would have been dine-in visits, the line contributes nothing, and we will not know until we can run the tests in §28.3.

We have not tested whether the food is good enough. The tasting drive in §28.5 happens in pre-opening (Chapter 9), before a single item is published, and the nine-item list is provisional until it does.

The physical route is unresolved. The floor plan (Chapter 7) needs a pickup point in the host's sightline that does not cross the guest path. If the second-generation shell does not permit one, the Friday and Saturday shoulder windows close too, and the line drops to Tuesday through Thursday — which would be roughly $18,700 of sales, not $31,200.

The year-two test, written down now so it happens

Marketplace listing is deferred, not rejected. The plan commits to a defined re-entry test in year two:

1. A six-week windowed pilot on one platform, lowest tier, no promotions, no sponsored placement, open only during the hours in Figure 28.6.
2. A blackout control: two of the six weeks run with the channel off on Friday and Saturday, with dine-in covers measured against the same nights in the surrounding weeks.
3. Go/no-go on two thresholds, decided before the pilot starts: contribution per order net of the measured cannibalization must exceed zero, and no hearth item may be produced for the channel during an hour running above 90% of the sustainable 28-item rate.
4. Every bag carries a first-party conversion card, and the share of off-premise orders arriving direct is reported monthly.

Open questions carried forward

1. Does the shell permit a pickup point off the guest path, or does the Friday/Saturday shoulder window die with the floor plan? (Chapter 7, revisited)
2. What is the actual cannibalization rate once we can measure it, and does the 25% assumption hold? (Chapters 31, 32 — and the tests in §28.3)
3. Does the catering and private-event line make more of the same kitchen capacity than off-premise does, and should the two compete for the same Tuesday? (Chapter 29)
4. If the room underperforms on weeknights, is off-premise the right response — or is the right response a pricing or daypart change? (Chapters 24, 39)
5. Does the technology stack integrate the 86 list across the POS and the ordering channel, and what does that integration cost? (Chapter 26, revisited)

Conclusion

Off-premise is not free revenue and it is not a mistake. It is a second business line with a different margin structure that happens to share your kitchen, and it has to be priced like one.

The arithmetic for Bellwether came out clearly. The same $65 of food contributes $61.76 sold to two people at a table, $38.65 through our own pickup channel, and $25.89 through a marketplace at 25% — and the largest single gap in that comparison is not the commission but the $25.76 of wine nobody ordered. At the binding hour on a Saturday, when the hearth is producing 29 items against a sustainable 28, a delivery order that consumes one hearth slot earns $25.89 in place of the $52.71 the fire was already earning, and makes the business $26.82 worse off while recording $65.00 of sales.

Which is why the answer was never yes or no. It was a schedule. The channel is welcome Tuesday through Thursday, welcome in the Friday and Saturday shoulders, and barred from 6:00 to 8:45 on the two nights that carry the week. The plan keeps its $31,200 and rebuilds the line underneath it: first party only, nine items, parity pricing, labor absorbed, and a written year-two test for the marketplace with a blackout control and thresholds set in advance.

Three principles generalize past this chapter and past this platform generation.

Find the constraint and compute contribution per unit of it. Contribution per order is the wrong denominator whenever something is scarce. Bellwether's scarce resource is a hearth-minute, and once you price it, the entire decision resolves in an afternoon. Theme four of this book said every seat-hour is inventory you can't store; a hearth-minute is the same thing wearing an apron.

Never manage a channel from the contract. Manage it from the payout statement. The commission line said 25% and the statement said 35%, and ten of those points were choices nobody had to authorize. That is Chapter 34's argument about comps and voids, arriving through the back door.

Ask who owns the guest, and price the answer. $6,708 a year at Bellwether's tiny volume, and much more than that at any real one — plus the service recovery you cannot perform for a guest whose name you never learned. Theme three of this book is that you sell hospitality, not plates. Off-premise strips out the room, the server, and the moment the plate lands, and leaves you with the food and how you behave when the food is wrong. If a third party stands between you and that conversation, you have sold the plates and kept nothing.

Chapter 29 turns to the other way of selling the kitchen you already pay for — catering, private events, and banquets, where the covers are known, the menu is fixed, the deposit is collected in advance, and the margin is frequently the best in the building. It is also, at Bellwether, a direct competitor with off-premise for the same idle Tuesday. Keep the contribution-per-constraint habit; you are about to need it again.

Key Terms

Off-premise — food prepared in a restaurant's kitchen and consumed somewhere else: takeout, curbside, drive-through, delivery, and catering. Structurally a separate business line with its own margin structure sharing the same kitchen. (Ch. 28)

Third-party marketplace — a platform that lists many restaurants, owns the guest relationship, takes the order, processes payment, dispatches a driver, and remits a net figure to the restaurant. You are a supplier on somebody else's shelf. (Ch. 28)

Direct delivery — delivery fulfilled by the restaurant itself, either with employed drivers or by purchasing a white-label delivery-as-a-service drop, while the restaurant retains the guest relationship and the order data. (Ch. 28)

First-party ordering — a guest ordering directly from the restaurant's own channel — its website, ordering page, or a link from its Google Business Profile — so that the restaurant owns the menu, pricing, payment relationship, guest data, and service recovery. (Ch. 28)

Commission rate — a third-party marketplace's stated percentage of the menu subtotal, commonly in the 15–30% range depending on service tier, market, and contract. It is charged on menu price, not on margin. (Ch. 28)

Effective take rate — the total a platform actually retains — commission plus promotional funding, sponsored-listing spend, error refunds charged back, and per-order or hardware fees — divided by gross menu sales. Routinely and materially higher than the commission rate; readable only from the payout statement. (Ch. 28)

Commission caps — municipal ordinances limiting what third-party delivery platforms may charge restaurants. Adopted by several U.S. cities during the 2020 shutdowns, commonly around 15% for delivery with a few additional points allowed for other services; some were later made permanent and at least one was litigated. Specifics vary by city — read the ordinance. (Ch. 28)

Delivery price parity — charging the same menu prices on an off-premise channel as in the dining room. The alternative is an off-premise uplift, which recovers part of the commission at the cost of a discrepancy guests can see. (Ch. 28)

Incremental sales — sales that would not have occurred at all without the channel; the transaction was created rather than moved. (Ch. 28)

Cannibalized sales — sales the channel moved from a higher-contribution channel to a lower one. Total revenue looks flat or up while contribution falls, which is why sales figures alone cannot detect it. (Ch. 28)

Break-even cannibalization rate — the fraction of off-premise orders that could replace a dine-in visit before the channel stops adding contribution: off-premise contribution per order divided by the contribution of the dine-in visit it replaces. (Ch. 28)

Packaging cost — the fully loaded per-order cost of every disposable that leaves the building with the food: containers, lids, bags, cutlery, napkins, sauce cups, tamper seals, and labels. Built as a fixed base per order plus a variable per-container cost. (Ch. 28)

Throughput impact — the reduction in a kitchen's dine-in production capacity caused by off-premise production competing for the same constrained resource. Measured as contribution per unit of the constraint, not per order. (Ch. 28)

Spaced Review

1. Without looking back: name the four things off-premise changed about restaurant economics, and say which one never appears on a payout statement.
2. A delivery order contributes $22 and consumes one slot on a piece of equipment that, during that hour, is producing $48 of contribution per item. What is the net effect on the business, and what does the POS record?
3. From Chapter 12: an item has a $9.00 plate cost and sells for $27.00. What is its contribution margin dining in? On a 25% marketplace at parity pricing, with $1.15 of packaging and packing labor allocated to it, what does it contribute — and what percentage of the dine-in figure is that?
4. From Chapters 1 and 11: a restaurant adds a delivery channel that runs at a 40% contribution rate and grows to 20% of sales, replacing dine-in sales that ran at 67%. Total sales are unchanged. Estimate the effect on prime cost percentage and on operating profit, and explain why the two move differently.
5. The recurring question: you have been asked to justify keeping an off-premise line in a business plan. You have twelve months of sales data and no ability to run an experiment. What are the three strongest pieces of evidence you can assemble for incrementality, and what is the honest thing to say about the ones you cannot get?