Direct ordering vs apps: the equation that decides your margin in 2026

Chapter 1 — Commission is not a marketing expense: it is contribution margin that never comes back

A 27% marketplace commission does not belong on the marketing line of your P&L, it belongs to the variable cost of the dish, and that change of column decides whether you understand your business or run it blind. Take a dish with a 30% food cost: sold through an app it leaves 43 gross points to cover payroll, rent, utilities and profit; the same dish sold through your own channel, with a 3.4% payment gateway, leaves 66.6 points. That is a 23-point gap on every single order, and in a business that rarely clears 10% EBITDA, 23 points are the whole business. The size of the channel explains why nobody negotiates it: DoorDash moved a marketplace GOV of roughly US$80.2 billion in 2024 (DoorDash, 2024 results), so the fee is not discussed, it is accepted. Rent gets renegotiated every year. Commission, curiously, does not.

Chapter 2 — Should you switch the apps off? No, and whoever does it overnight loses sales they never recover

Switching off the apps is a cash-flow mistake dressed up as a margin decision, and I hold that view even though the opposite argument sounds braver. Platforms generated nearly US$60 billion in sales for local merchants during 2024 (DoorDash, 2024): part of that demand is incremental, people who would never have called your kitchen. Play out the whole scenario: you cut the marketplace on a Monday, you recover 27 points per order and you lose 40% of your orders; with that mix your total contribution margin FALLS, because 66.6 points on 60 orders yield less than 43 points on 100. The arithmetic only flips once your own channel already recovers at least half of that volume. The right sequence is to build first and disconnect later, never the other way round.

Chapter 3 — Customer data is the asset, and today it lives in a database you do not control

What your own channel buys is not the commission saving, it is the name, the phone number and the ordering frequency of whoever eats your food, and that is the asset somebody values when they put a price on your restaurant. According to the National Restaurant Association (2025), nearly 75% of sales already happen off-premise: the growing share of the business is precisely the one sitting on somebody else's server. A buyer pays no multiple for rented demand. Diego F. Parra keeps pushing an uncomfortable test at Masterestaurant: if your brand vanished from the marketplace tomorrow, how many customers would know how to write to you directly? When you count the answer on your fingers, you do not own a restaurant with delivery, you are a kitchen supplier for a platform. Email still works: 25.1% average open rate in 2023 (Omnisend, 2024).

Chapter 4 — The real cost of a direct order: the gateway is cheap, traffic is not free

Your own channel does not cost 3.4%, it costs 3.4% plus the work of getting somebody to place the order, and that second line is the one almost nobody budgets. Add gateway, ordering platform, in-house or outsourced courier per order and the upfront acquisition spend, and a mature direct order settles between 12% and 18% of ticket, against 25% to 32% for the marketplace. You still win, but by 10 to 14 points, not by 25. Your job is not to create the habit, it is to intercept it with a repeat order. A customer who orders four times pays back the cost of capture; one who orders once does not.

Chapter 5 — Every revenue band plays a different game, and mixing them up gets expensive

The right mix shifts dramatically with your annual revenue, and applying the big operator's recipe to a small one ruins both. Below 500 thousand dollars a year, the minimum viable direct channel is WhatsApp with a catalogue and a payment link: cost close to zero, while the marketplace stays at 50% or more of digital sales. Between 500 thousand and 1 million the breaking point shows up: a CRM plus an owned platform run between 3,000 and 8,000 dollars a year, a figure that pays for itself once the direct channel passes 20% of delivery. Above 1 million, every point of mix moved from app to owned channel is worth between 2,300 and 2,700 dollars of annual margin. The small band must not imitate the big one: it should collect the data and wait.

Chapter 6 — Above 5 million the problem changes its name: it is capacity, not commission

Once revenue passes 5 million a year, commission stops being the main enemy and the bottleneck moves to the kitchen and to last-mile operations. In that band a partial in-house fleet already makes sense, and the break-even for a salaried courier shows up around 2.5 to 3 sustained orders per hour; below that, the third party is cheaper. High-volume formats prove it from another angle: over 50% of a US QSR's revenue comes through the drive-thru (Business Research Insights, 2024) and an independent drive-thru-only unit posted a median US$9,227 million in 2024 (QSR Magazine, 2024). That channel is owned, measurable and free of intermediaries. Chipotle opens between 315 and 345 locations in 2025, over 80% of them with their own lane (Chipotle, Q4 2024). No coincidence there.

Chapter 7 — The celebrity-chef restaurant pays costs that commission never even touches

In the celebrity and large-format themed segment, above 5 million a year, app commission is a third-order problem next to that business's own cost structure. Skilled payroll, the wine list and the upkeep of the spectacle weigh more than any fee: alcohol, named by 46% of operators among the highest-margin categories (Technomic / Nation's Restaurant News, 2024), is only sold in the dining room, and no marketplace delivers it. That is why the correct strategy in that band is asymmetric: apps for the packable, low-ticket product line, owned channel and direct booking for the full experience where the margin actually lives. Selling delivery of the signature menu through an app in that band gives away prestige and margin at the same time, and I have argued that decision more often than I would like.

Chapter 8 — What moves the needle in the first quarter: three numbers and one transfer rule

Start by measuring three things and leave the ideological debate for later: each channel's share of sales, contribution margin per order in each one, and customers with their own data captured this month. With that you can set a transfer target of 5% of mix per quarter, which is aggressive but reachable, and check it against the P&L instead of against your gut. Cost pressure does not wait: in Colombia menu prices rose 9.8% from February 2025 to sustain 98,000 jobs (ACODRES, 2025), so margin lost to commission will not be recovered by raising prices again. This week, drop a QR code with a 10% incentive inside every package that goes out through an app and measure how many of those customers come back directly within 60 days. That number, not the opinion, decides your mix for the year.

Chapter 9 — The three differences that move EBITDA

The first difference is accounting, not marketing: marketplace commission is not a selling expense, it is a direct reduction of every dish's contribution margin. A dish with 30% food cost and 27% commission leaves 43 gross points to cover payroll, rent and utilities; the same dish sold through your own channel, with a 3.4% gateway, leaves 66.6 points. That 23-point gap separates single-digit EBITDA from double-digit EBITDA. The second is asset ownership. Off-premise sales are already near 75% of the business, according to the National Restaurant Association (2025): when that growing share of the business lives in a database you do not control, the sale value of the restaurant falls, because no buyer pays for rented demand. The customer record is the only digital asset that appreciates over time. The third is territory risk. Whoever fails to capture it is not facing a missing market, they simply never built the local engine that intercepts the customer before the app opens.